orpo chinese phi3 4K

README.md

@@ -1,3 +1,62 @@
 ---
-license: mit
+license: other
+base_model: /data/user/chengrui/project/mergekit/Phi-3-mini-128k-instruct
+tags:
+- llama-factory
+- full
+- generated_from_trainer
+model-index:
+- name: phi3-chinese-orpo
+  results: []
 ---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+# phi3-chinese-orpo
+
+This model is a fine-tuned version of [/data/user/chengrui/project/mergekit/Phi-3-mini-128k-instruct](https://huggingface.co//data/user/chengrui/project/mergekit/Phi-3-mini-128k-instruct) on the dpo_mix_en and the dpo_mix_zh datasets.
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 5e-06
+- train_batch_size: 1
+- eval_batch_size: 1
+- seed: 42
+- distributed_type: multi-GPU
+- num_devices: 6
+- gradient_accumulation_steps: 8
+- total_train_batch_size: 48
+- total_eval_batch_size: 6
+- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
+- lr_scheduler_type: cosine
+- lr_scheduler_warmup_ratio: 0.1
+- lr_scheduler_warmup_steps: 20
+- num_epochs: 3.0
+- mixed_precision_training: Native AMP
+
+### Training results
+
+
+
+### Framework versions
+
+- Transformers 4.40.0
+- Pytorch 2.1.0+cu121
+- Datasets 2.15.0
+- Tokenizers 0.19.1

added_tokens.json

@@ -0,0 +1,40 @@
+{
+  "<|/code|>": 32014,
+  "<|/data|>": 32033,
+  "<|/inst|>": 32037,
+  "<|/query|>": 32031,
+  "<|/sys|>": 32035,
+  "<|assistant_mask|>": 32017,
+  "<|assistant|>": 32001,
+  "<|calc|>": 32012,
+  "<|code|>": 32013,
+  "<|continue|>": 32009,
+  "<|data|>": 32032,
+  "<|diff_marker|>": 32025,
+  "<|disc_sep|>": 32029,
+  "<|disc_start|>": 32028,
+  "<|disc_thread|><|query|>": 32030,
+  "<|endoftext|>": 32000,
+  "<|end|>": 32007,
+  "<|fim_middle|>": 32021,
+  "<|fim_prefix|>": 32020,
+  "<|fim_suffix|>": 32022,
+  "<|function_call|>": 32005,
+  "<|function_list|>": 32011,
+  "<|function_output|>": 32003,
+  "<|ghissue|>": 32026,
+  "<|ghreview|>": 32027,
+  "<|inst|>": 32036,
+  "<|ipynb_marker|>": 32024,
+  "<|message|>": 32019,
+  "<|meta_start|>": 32023,
+  "<|raw|>": 32008,
+  "<|resource|>": 32016,
+  "<|start|>": 32018,
+  "<|step|>": 32002,
+  "<|summary|>": 32015,
+  "<|system|>": 32006,
+  "<|sys|>": 32034,
+  "<|tag|>": 32004,
+  "<|user|>": 32010
+}

all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 2.994601079784043,
+    "total_flos": 132590267662336.0,
+    "train_loss": 0.7937506708579186,
+    "train_runtime": 49781.9259,
+    "train_samples_per_second": 1.205,
+    "train_steps_per_second": 0.025
+}

config.json

@@ -0,0 +1,138 @@
+{
+  "_name_or_path": "/data/user/chengrui/project/mergekit/Phi-3-mini-128k-instruct",
+  "architectures": [
+    "Phi3ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_phi3.Phi3Config",
+    "AutoModel": "modeling_phi3.Phi3ForCausalLM",
+    "AutoModelForCausalLM": "modeling_phi3.Phi3ForCausalLM"
+  },
+  "bos_token_id": 1,
+  "embd_pdrop": 0.0,
+  "eos_token_id": 32000,
+  "hidden_act": "silu",
+  "hidden_size": 3072,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "max_position_embeddings": 131072,
+  "model_type": "phi3",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 32,
+  "original_max_position_embeddings": 4096,
+  "pad_token_id": 32000,
+  "resid_pdrop": 0.0,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": {
+    "long_factor": [
+      1.0299999713897705,
+      1.0499999523162842,
+      1.0499999523162842,
+      1.0799999237060547,
+      1.2299998998641968,
+      1.2299998998641968,
+      1.2999999523162842,
+      1.4499999284744263,
+      1.5999999046325684,
+      1.6499998569488525,
+      1.8999998569488525,
+      2.859999895095825,
+      3.68999981880188,
+      5.419999599456787,
+      5.489999771118164,
+      5.489999771118164,
+      9.09000015258789,
+      11.579999923706055,
+      15.65999984741211,
+      15.769999504089355,
+      15.789999961853027,
+      18.360000610351562,
+      21.989999771118164,
+      23.079999923706055,
+      30.009998321533203,
+      32.35000228881836,
+      32.590003967285156,
+      35.56000518798828,
+      39.95000457763672,
+      53.840003967285156,
+      56.20000457763672,
+      57.95000457763672,
+      59.29000473022461,
+      59.77000427246094,
+      59.920005798339844,
+      61.190006256103516,
+      61.96000671386719,
+      62.50000762939453,
+      63.3700065612793,
+      63.48000717163086,
+      63.48000717163086,
+      63.66000747680664,
+      63.850006103515625,
+      64.08000946044922,
+      64.760009765625,
+      64.80001068115234,
+      64.81001281738281,
+      64.81001281738281
+    ],
+    "short_factor": [
+      1.05,
+      1.05,
+      1.05,
+      1.1,
+      1.1,
+      1.1500000000000001,
+      1.2000000000000002,
+      1.2500000000000002,
+      1.3000000000000003,
+      1.3500000000000003,
+      1.5000000000000004,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.000000000000001,
+      2.0500000000000007,
+      2.0500000000000007,
+      2.0500000000000007,
+      2.1000000000000005,
+      2.1000000000000005,
+      2.1000000000000005,
+      2.1500000000000004,
+      2.1500000000000004,
+      2.3499999999999996,
+      2.549999999999999,
+      2.5999999999999988,
+      2.5999999999999988,
+      2.7499999999999982,
+      2.849999999999998,
+      2.849999999999998,
+      2.9499999999999975
+    ],
+    "type": "su"
+  },
+  "rope_theta": 10000.0,
+  "sliding_window": 262144,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.40.0",
+  "use_cache": false,
+  "vocab_size": 32064
+}

configuration_phi3.py

@@ -0,0 +1,213 @@
+# coding=utf-8
+# Copyright 2024 Microsoft and the HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+""" Phi-3 model configuration"""
+
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+PHI3_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "microsoft/Phi-3-mini-4k-instruct": "https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/resolve/main/config.json",
+    "microsoft/Phi-3-mini-128k-instruct": "https://huggingface.co/microsoft/Phi-3-mini-128k-instruct/resolve/main/config.json",
+}
+
+
+class Phi3Config(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`Phi3Model`]. It is used to instantiate a Phi-3
+    model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
+    defaults will yield a similar configuration to that of the
+    [microsoft/Phi-3-mini-4k-instruct](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct).
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 32064):
+            Vocabulary size of the Phi-3 model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`Phi3Model`].
+        hidden_size (`int`, *optional*, defaults to 3072):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 8192):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer decoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer decoder.
+        num_key_value_heads (`int`, *optional*):
+            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
+            `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
+            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
+            by meanpooling all the original heads within that group. For more details checkout [this
+            paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
+            `num_attention_heads`.
+        resid_pdrop (`float`, *optional*, defaults to 0.0):
+            Dropout probability for mlp outputs.
+        embd_pdrop (`int`, *optional*, defaults to 0.0):
+            The dropout ratio for the embeddings.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio after computing the attention scores.
+        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (function or string) in the decoder.
+        max_position_embeddings (`int`, *optional*, defaults to 4096):
+            The maximum sequence length that this model might ever be used with.
+        original_max_position_embeddings (`int`, *optional*, defaults to 4096):
+            The maximum sequence length that this model was trained with. This is used to determine the size of the
+            original RoPE embeddings when using long scaling.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-05):
+            The epsilon value used for the RMSNorm.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`. Whether to tie weight embeddings or not.
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether to tie weight embeddings
+        rope_theta (`float`, *optional*, defaults to 10000.0):
+            The base period of the RoPE embeddings.
+        rope_scaling (`dict`, *optional*):
+            The scaling strategy for the RoPE embeddings. If `None`, no scaling is applied. If a dictionary, it must
+            contain the following keys: `type`, `short_factor` and `long_factor`. The `type` must be either `su` or `yarn` and
+            the `short_factor` and `long_factor` must be lists of numbers with the same length as the hidden size
+            divided by the number of attention heads divided by 2.
+        bos_token_id (`int`, *optional*, defaults to 1):
+            The id of the "beginning-of-sequence" token.
+        eos_token_id (`int`, *optional*, defaults to 32000):
+            The id of the "end-of-sequence" token.
+        pad_token_id (`int`, *optional*, defaults to 32000):
+            The id of the padding token.
+        sliding_window (`int`, *optional*):
+            Sliding window attention window size. If `None`, no sliding window is applied.
+
+    Example:
+
+    ```python
+    >>> from transformers import Phi3Model, Phi3Config
+
+    >>> # Initializing a Phi-3 style configuration
+    >>> configuration = Phi3Config.from_pretrained("microsoft/Phi-3-mini-4k-instruct")
+
+    >>> # Initializing a model from the configuration
+    >>> model = Phi3Model(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "phi3"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=32064,
+        hidden_size=3072,
+        intermediate_size=8192,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=None,
+        resid_pdrop=0.0,
+        embd_pdrop=0.0,
+        attention_dropout=0.0,
+        hidden_act="silu",
+        max_position_embeddings=4096,
+        original_max_position_embeddings=4096,
+        initializer_range=0.02,
+        rms_norm_eps=1e-5,
+        use_cache=True,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        bos_token_id=1,
+        eos_token_id=32000,
+        pad_token_id=32000,
+        sliding_window=None,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.resid_pdrop = resid_pdrop
+        self.embd_pdrop = embd_pdrop
+        self.attention_dropout = attention_dropout
+        self.hidden_act = hidden_act
+        self.max_position_embeddings = max_position_embeddings
+        self.original_max_position_embeddings = original_max_position_embeddings
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.rope_scaling = rope_scaling
+        self._rope_scaling_validation()
+        self.sliding_window = sliding_window
+
+        super().__init__(
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            pad_token_id=pad_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+    def _rope_scaling_validation(self):
+        """
+        Validate the `rope_scaling` configuration.
+        """
+        if self.rope_scaling is None:
+            return
+
+        if not isinstance(self.rope_scaling, dict) or len(self.rope_scaling) != 3:
+            raise ValueError(
+                "`rope_scaling` must be a dictionary with three fields, `type`, `short_factor` and `long_factor`, "
+                f"got {self.rope_scaling}"
+            )
+        rope_scaling_type = self.rope_scaling.get("type", None)
+        rope_scaling_short_factor = self.rope_scaling.get("short_factor", None)
+        rope_scaling_long_factor = self.rope_scaling.get("long_factor", None)
+        if rope_scaling_type is None or rope_scaling_type not in ["su", "yarn"]:
+            raise ValueError(f"`rope_scaling`'s type field must be one of ['su', 'yarn'], got {rope_scaling_type}")
+        if not (
+            isinstance(rope_scaling_short_factor, list)
+            and all(isinstance(x, (int, float)) for x in rope_scaling_short_factor)
+        ):
+            raise ValueError(
+                f"`rope_scaling`'s short_factor field must be a list of numbers, got {rope_scaling_short_factor}"
+            )
+        if not len(rope_scaling_short_factor) == self.hidden_size // self.num_attention_heads // 2:
+            raise ValueError(
+                f"`rope_scaling`'s short_factor field must have length {self.hidden_size // self.num_attention_heads // 2}, got {len(rope_scaling_short_factor)}"
+            )
+        if not (
+            isinstance(rope_scaling_long_factor, list)
+            and all(isinstance(x, (int, float)) for x in rope_scaling_long_factor)
+        ):
+            raise ValueError(
+                f"`rope_scaling`'s long_factor field must be a list of numbers, got {rope_scaling_long_factor}"
+            )
+        if not len(rope_scaling_long_factor) == self.hidden_size // self.num_attention_heads // 2:
+            raise ValueError(
+                f"`rope_scaling`'s long_factor field must have length {self.hidden_size // self.num_attention_heads // 2}, got {len(rope_scaling_long_factor)}"
+            )

generation_config.json

@@ -0,0 +1,11 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": [
+    32000,
+    32001,
+    32007
+  ],
+  "pad_token_id": 32000,
+  "transformers_version": "4.40.0"
+}

model-00001-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a5de46e0a3c8c853a9fb520b403c406cb1254fb754022ed56ef38ada6613888
+size 4972489200

model-00002-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d42e717d30a6f95a5312fb8d8ebe283253417fac308597c8673bcb1678ac959
+size 2669692488

model.safetensors.index.json

@@ -0,0 +1,202 @@
+{
+  "metadata": {
+    "total_size": 7642159104
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00002-of-00002.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
+    "model.norm.weight": "model-00002-of-00002.safetensors"
+  }
+}

modeling_phi3.py

@@ -0,0 +1,1606 @@
+# coding=utf-8
+# Copyright 2024 Microsoft and the HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+""" PyTorch Phi-3 model."""
+
+import inspect
+import math
+import warnings
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
+
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache
+from transformers.modeling_attn_mask_utils import _prepare_4d_causal_attention_mask
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPast,
+    CausalLMOutputWithPast,
+    SequenceClassifierOutputWithPast,
+    TokenClassifierOutput,
+)
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import (
+    add_code_sample_docstrings,
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    is_flash_attn_2_available,
+    is_flash_attn_greater_or_equal_2_10,
+    logging,
+    replace_return_docstrings,
+)
+from .configuration_phi3 import Phi3Config
+
+
+logger = logging.get_logger(__name__)
+
+# Transformers scans dependencies in the modeling file, causing issues on conditional loading. The regex only ignores try/catch blocks, but not if statements
+# if is_flash_attn_2_available():
+_flash_supports_window_size = False
+try:
+    from flash_attn import flash_attn_func, flash_attn_varlen_func
+    from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input  # noqa
+
+    _flash_supports_window_size = "window_size" in list(inspect.signature(flash_attn_func).parameters)
+except ImportError as error:
+    logger.warning(
+        f"`flash-attention` package not found, consider installing for better performance: {error}."
+    )
+    if not _flash_supports_window_size:
+        logger.warning(
+            "Current `flash-attenton` does not support `window_size`. Either upgrade or use `attn_implementation='eager'`."
+        )
+
+_CHECKPOINT_FOR_DOC = "microsoft/Phi-3-mini-4k-instruct"
+_CONFIG_FOR_DOC = "Phi3Config"
+
+PHI3_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "microsoft/Phi-3-mini-4k-instruct",
+    "microsoft/Phi-3-mini-128k-instruct",
+    # See all Phi-3 models at https://huggingface.co/models?filter=Phi-3
+]
+
+
+# Copied from transformers.models.llama.modeling_llama.LlamaRMSNorm with Llama->Phi3
+class Phi3RMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        Phi3RMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(torch.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+        return self.weight * hidden_states.to(input_dtype)
+
+
+# Copied from transformers.models.llama.modeling_llama._get_unpad_data
+def _get_unpad_data(attention_mask):
+    seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
+    indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
+    max_seqlen_in_batch = seqlens_in_batch.max().item()
+    cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
+    return (
+        indices,
+        cu_seqlens,
+        max_seqlen_in_batch,
+    )
+
+
+# Copied from transformers.models.gemma.modeling_gemma.GemmaRotaryEmbedding with gemma->phi3, Gemma->Phi3
+class Phi3RotaryEmbedding(nn.Module):
+    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None):
+        super().__init__()
+
+        self.dim = dim
+        self.max_position_embeddings = max_position_embeddings
+        self.base = base
+        self.register_buffer("inv_freq", None, persistent=False)
+
+    @torch.no_grad()
+    def forward(self, x, position_ids, seq_len=None):
+        # x: [bs, num_attention_heads, seq_len, head_size]
+        if self.inv_freq is None:
+            self.inv_freq = 1.0 / (
+                self.base ** (torch.arange(0, self.dim, 2, dtype=torch.int64, device=x.device).float() / self.dim)
+            )
+        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1)
+        position_ids_expanded = position_ids[:, None, :].float()
+        # Force float32 since bfloat16 loses precision on long contexts
+        # See https://github.com/huggingface/transformers/pull/29285
+        device_type = x.device.type
+        device_type = device_type if isinstance(device_type, str) and device_type != "mps" else "cpu"
+        with torch.autocast(device_type=device_type, enabled=False):
+            freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
+            emb = torch.cat((freqs, freqs), dim=-1)
+            cos = emb.cos()
+            sin = emb.sin()
+        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
+
+
+class Phi3SuScaledRotaryEmbedding(Phi3RotaryEmbedding):
+    def __init__(self, dim, config, device=None):
+        super().__init__(dim, config.max_position_embeddings, config.rope_theta, device)
+
+        self.short_factor = config.rope_scaling["short_factor"]
+        self.long_factor = config.rope_scaling["long_factor"]
+        self.original_max_position_embeddings = config.original_max_position_embeddings
+
+    @torch.no_grad()
+    def forward(self, x, position_ids, seq_len=None):
+        seq_len = torch.max(position_ids) + 1
+        if seq_len > self.original_max_position_embeddings:
+            ext_factors = torch.tensor(self.long_factor, dtype=torch.float32, device=x.device)
+        else:
+            ext_factors = torch.tensor(self.short_factor, dtype=torch.float32, device=x.device)
+
+        inv_freq_shape = torch.arange(0, self.dim, 2, dtype=torch.int64, device=x.device).float() / self.dim
+        self.inv_freq = 1.0 / (ext_factors * self.base**inv_freq_shape)
+
+        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1)
+        position_ids_expanded = position_ids[:, None, :].float()
+
+        # Force float32 since bfloat16 loses precision on long contexts
+        # See https://github.com/huggingface/transformers/pull/29285
+        device_type = x.device.type
+        device_type = device_type if isinstance(device_type, str) and device_type != "mps" else "cpu"
+        with torch.autocast(device_type=device_type, enabled=False):
+            freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
+            emb = torch.cat((freqs, freqs), dim=-1)
+
+            scale = self.max_position_embeddings / self.original_max_position_embeddings
+            if scale <= 1.0:
+                scaling_factor = 1.0
+            else:
+                scaling_factor = math.sqrt(1 + math.log(scale) / math.log(self.original_max_position_embeddings))
+
+            cos = emb.cos() * scaling_factor
+            sin = emb.sin() * scaling_factor
+        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
+
+
+class Phi3YarnScaledRotaryEmbedding(Phi3RotaryEmbedding):
+    def __init__(self, dim, config, device=None):
+        super().__init__(dim, config.max_position_embeddings, config.rope_theta, device)
+
+        self.short_factor = config.rope_scaling["short_factor"]
+        self.long_factor = config.rope_scaling["long_factor"]
+        self.original_max_position_embeddings = config.original_max_position_embeddings
+
+    @torch.no_grad()
+    def forward(self, x, position_ids, seq_len=None):
+        seq_len = torch.max(position_ids) + 1
+        if seq_len > self.original_max_position_embeddings:
+            ext_factors = torch.tensor(self.long_factor, dtype=torch.float32, device=x.device)
+        else:
+            ext_factors = torch.tensor(self.short_factor, dtype=torch.float32, device=x.device)
+
+        inv_freq_shape = torch.arange(0, self.dim, 2, dtype=torch.int64, device=x.device).float() / self.dim
+        self.inv_freq = 1.0 / (ext_factors * self.base**inv_freq_shape)
+
+        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1)
+        position_ids_expanded = position_ids[:, None, :].float()
+
+        # Force float32 since bfloat16 loses precision on long contexts
+        # See https://github.com/huggingface/transformers/pull/29285
+        device_type = x.device.type
+        device_type = device_type if isinstance(device_type, str) and device_type != "mps" else "cpu"
+        with torch.autocast(device_type=device_type, enabled=False):
+            freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
+            emb = torch.cat((freqs, freqs), dim=-1)
+
+            scale = self.max_position_embeddings / self.original_max_position_embeddings
+            if scale <= 1.0:
+                scaling_factor = 1.0
+            else:
+                scaling_factor = 0.1 * math.log(scale) + 1.0
+
+            cos = emb.cos() * scaling_factor
+            sin = emb.sin() * scaling_factor
+        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
+
+
+# Copied from transformers.models.llama.modeling_llama.rotate_half
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2 :]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+# Copied from transformers.models.llama.modeling_llama.apply_rotary_pos_emb
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
+    """Applies Rotary Position Embedding to the query and key tensors.
+
+    Args:
+        q (`torch.Tensor`): The query tensor.
+        k (`torch.Tensor`): The key tensor.
+        cos (`torch.Tensor`): The cosine part of the rotary embedding.
+        sin (`torch.Tensor`): The sine part of the rotary embedding.
+        position_ids (`torch.Tensor`, *optional*):
+            Deprecated and unused.
+        unsqueeze_dim (`int`, *optional*, defaults to 1):
+            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
+            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
+            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
+            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
+            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
+            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
+    Returns:
+        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
+    """
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+class Phi3MLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+
+        self.config = config
+        self.gate_up_proj = nn.Linear(config.hidden_size, 2 * config.intermediate_size, bias=False)
+        self.down_proj = nn.Linear(config.intermediate_size, config.hidden_size, bias=False)
+
+        self.activation_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, hidden_states: torch.FloatTensor) -> torch.FloatTensor:
+        up_states = self.gate_up_proj(hidden_states)
+
+        gate, up_states = up_states.chunk(2, dim=-1)
+        up_states = up_states * self.activation_fn(gate)
+
+        return self.down_proj(up_states)
+
+
+# Copied from transformers.models.llama.modeling_llama.repeat_kv with llama->phi
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """
+    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+class Phi3Attention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: Phi3Config, layer_idx: Optional[int] = None):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        if layer_idx is None:
+            logger.warning_once(
+                f"Instantiating {self.__class__.__name__} without passing a `layer_idx` is not recommended and will "
+                "lead to errors during the forward call if caching is used. Please make sure to provide a `layer_idx` "
+                "when creating this class."
+            )
+
+        self.attention_dropout = config.attention_dropout
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.hidden_size // self.num_heads
+        self.num_key_value_heads = config.num_key_value_heads
+        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
+        self.max_position_embeddings = config.max_position_embeddings
+        self.original_max_position_embeddings = config.original_max_position_embeddings
+        self.rope_theta = config.rope_theta
+        self.rope_scaling = config.rope_scaling
+        self.is_causal = True
+
+        if (self.head_dim * self.num_heads) != self.hidden_size:
+            raise ValueError(
+                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
+                f" and `num_heads`: {self.num_heads})."
+            )
+
+        op_size = self.num_heads * self.head_dim + 2 * (self.num_key_value_heads * self.head_dim)
+        self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
+        self.qkv_proj = nn.Linear(self.hidden_size, op_size, bias=False)
+        self._init_rope()
+
+    def _init_rope(self):
+        if self.rope_scaling is None:
+            self.rotary_emb = Phi3RotaryEmbedding(
+                self.head_dim,
+                max_position_embeddings=self.max_position_embeddings,
+                base=self.rope_theta,
+            )
+        else:
+            scaling_type = self.config.rope_scaling["type"]
+            if scaling_type == "su":
+                self.rotary_emb = Phi3SuScaledRotaryEmbedding(self.head_dim, self.config)
+            elif scaling_type == "yarn":
+                self.rotary_emb = Phi3YarnScaledRotaryEmbedding(self.head_dim, self.config)
+            else:
+                raise ValueError(f"Unknown RoPE scaling type {scaling_type}")
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Cache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        logger.warning_once("You are not running the flash-attention implementation, expect numerical differences.")
+
+        bsz, q_len, _ = hidden_states.size()
+
+        qkv = self.qkv_proj(hidden_states)
+        query_pos = self.num_heads * self.head_dim
+        query_states = qkv[..., :query_pos]
+        key_states = qkv[..., query_pos : query_pos + self.num_key_value_heads * self.head_dim]
+        value_states = qkv[..., query_pos + self.num_key_value_heads * self.head_dim :]
+
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            if self.layer_idx is None:
+                raise ValueError(
+                    f"The cache structure has changed since version v4.36. If you are using {self.__class__.__name__} "
+                    "for auto-regressive decoding with k/v caching, please make sure to initialize the attention class "
+                    "with a layer index."
+                )
+            kv_seq_len += past_key_value.get_usable_length(kv_seq_len, self.layer_idx)
+        cos, sin = self.rotary_emb(value_states, position_ids, seq_len=kv_seq_len)
+
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+
+        if past_key_value is not None:
+            cache_kwargs = {"sin": sin, "cos": cos}  # Specific to RoPE models
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        # repeat k/v heads if n_kv_heads < n_heads
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz, self.num_heads, q_len, kv_seq_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+            attn_weights = attn_weights + attention_mask
+
+        # upcast attention to fp32
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(value_states.dtype)
+        attn_weights = nn.functional.dropout(attn_weights, p=self.attention_dropout, training=self.training)
+
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        attn_output = attn_output.transpose(1, 2).contiguous()
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+
+class Phi3FlashAttention2(Phi3Attention):
+    """
+    Phi-3 flash attention module. This module inherits from `Phi3Attention` as the weights of the module stays
+    untouched. The only required change would be on the forward pass where it needs to correctly call the public API of
+    flash attention and deal with padding tokens in case the input contains any of them.
+    """
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2.__init__
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
+        # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
+        # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
+        self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.LongTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Cache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        **kwargs,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        # Phi3FlashAttention2 attention does not support output_attentions
+
+        if not _flash_supports_window_size:
+            logger.warning_once(
+                "The current flash attention version does not support sliding window attention. Please use `attn_implementation='eager'` or upgrade flash-attn library."
+            )
+            raise ValueError("The current flash attention version does not support sliding window attention.")
+
+        output_attentions = False
+
+        if "padding_mask" in kwargs:
+            warnings.warn(
+                "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
+            )
+
+            # overwrite attention_mask with padding_mask
+            attention_mask = kwargs.pop("padding_mask")
+
+        bsz, q_len, _ = hidden_states.size()
+
+        qkv = self.qkv_proj(hidden_states)
+        query_pos = self.num_heads * self.head_dim
+        query_states = qkv[..., :query_pos]
+        key_states = qkv[..., query_pos : query_pos + self.num_key_value_heads * self.head_dim]
+        value_states = qkv[..., query_pos + self.num_key_value_heads * self.head_dim :]
+
+        # Flash attention requires the input to have the shape
+        # batch_size x seq_length x head_dim x hidden_dim
+        # therefore we just need to keep the original shape
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            if self.layer_idx is None:
+                raise ValueError(
+                    f"The cache structure has changed since version v4.36. If you are using {self.__class__.__name__} "
+                    "for auto-regressive decoding with k/v caching, please make sure to initialize the attention class "
+                    "with a layer index."
+                )
+            kv_seq_len += past_key_value.get_usable_length(kv_seq_len, self.layer_idx)
+
+        # Because the input can be padded, the absolute sequence length depends on the max position id.
+        rotary_seq_len = max(kv_seq_len, position_ids[:, -1].max().item()) + 1
+        cos, sin = self.rotary_emb(value_states, position_ids, seq_len=rotary_seq_len)
+
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+
+        use_sliding_windows = (
+            _flash_supports_window_size
+            and getattr(self.config, "sliding_window", None) is not None
+            and kv_seq_len > self.config.sliding_window
+        )
+
+        if past_key_value is not None:
+            # Activate slicing cache only if the config has a value `sliding_windows` attribute
+            cache_has_contents = past_key_value.get_seq_length(self.layer_idx) > 0
+            if (
+                getattr(self.config, "sliding_window", None) is not None
+                and kv_seq_len > self.config.sliding_window
+                and cache_has_contents
+            ):
+                slicing_tokens = 1 - self.config.sliding_window
+
+                past_key = past_key_value[self.layer_idx][0]
+                past_value = past_key_value[self.layer_idx][1]
+
+                past_key = past_key[:, :, slicing_tokens:, :].contiguous()
+                past_value = past_value[:, :, slicing_tokens:, :].contiguous()
+
+                if past_key.shape[-2] != self.config.sliding_window - 1:
+                    raise ValueError(
+                        f"past key must have a shape of (`batch_size, num_heads, self.config.sliding_window-1, head_dim`), got"
+                        f" {past_key.shape}"
+                    )
+
+                if attention_mask is not None:
+                    attention_mask = attention_mask[:, slicing_tokens:]
+                    attention_mask = torch.cat([attention_mask, torch.ones_like(attention_mask[:, -1:])], dim=-1)
+
+            cache_kwargs = {"sin": sin, "cos": cos}  # Specific to RoPE models
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        # repeat k/v heads if n_kv_heads < n_heads
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+        attn_dropout = self.attention_dropout if self.training else 0.0
+
+        # In PEFT, usually we cast the layer norms in float32 for training stability reasons
+        # therefore the input hidden states gets silently casted in float32. Hence, we need
+        # cast them back in the correct dtype just to be sure everything works as expected.
+        # This might slowdown training & inference so it is recommended to not cast the LayerNorms
+        # in fp32.
+
+        if query_states.dtype == torch.float32:
+            if torch.is_autocast_enabled():
+                target_dtype = torch.get_autocast_gpu_dtype()
+            # Handle the case where the model is quantized
+            elif hasattr(self.config, "_pre_quantization_dtype"):
+                target_dtype = self.config._pre_quantization_dtype
+            else:
+                target_dtype = self.qkv_proj.weight.dtype
+
+            logger.warning_once(
+                f"The input hidden states seems to be silently casted in float32, this might be related to"
+                f" the fact you have upcasted embedding or layer norm layers in float32. We will cast back the input in"
+                f" {target_dtype}."
+            )
+
+            query_states = query_states.to(target_dtype)
+            key_states = key_states.to(target_dtype)
+            value_states = value_states.to(target_dtype)
+
+        # Reashape to the expected shape for Flash Attention
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+
+        attn_output = self._flash_attention_forward(
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            q_len,
+            dropout=attn_dropout,
+            use_sliding_windows=use_sliding_windows,
+        )
+
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size).contiguous()
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+    # Copied from transformers.models.mistral.modeling_mistral.MistralFlashAttention2._flash_attention_forward
+    def _flash_attention_forward(
+        self,
+        query_states,
+        key_states,
+        value_states,
+        attention_mask,
+        query_length,
+        dropout=0.0,
+        softmax_scale=None,
+        use_sliding_windows=False,
+    ):
+        """
+        Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
+        first unpad the input, then computes the attention scores and pad the final attention scores.
+
+        Args:
+            query_states (`torch.Tensor`):
+                Input query states to be passed to Flash Attention API
+            key_states (`torch.Tensor`):
+                Input key states to be passed to Flash Attention API
+            value_states (`torch.Tensor`):
+                Input value states to be passed to Flash Attention API
+            attention_mask (`torch.Tensor`):
+                The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
+                position of padding tokens and 1 for the position of non-padding tokens.
+            dropout (`float`):
+                Attention dropout
+            softmax_scale (`float`, *optional*):
+                The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
+            use_sliding_windows (`bool`, *optional*):
+                Whether to activate sliding window attention.
+        """
+        if not self._flash_attn_uses_top_left_mask:
+            causal = self.is_causal
+        else:
+            # TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
+            causal = self.is_causal and query_length != 1
+
+        # Contains at least one padding token in the sequence
+        if attention_mask is not None:
+            batch_size = query_states.shape[0]
+            query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
+                query_states, key_states, value_states, attention_mask, query_length
+            )
+
+            cu_seqlens_q, cu_seqlens_k = cu_seq_lens
+            max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
+
+            if not use_sliding_windows:
+                attn_output_unpad = flash_attn_varlen_func(
+                    query_states,
+                    key_states,
+                    value_states,
+                    cu_seqlens_q=cu_seqlens_q,
+                    cu_seqlens_k=cu_seqlens_k,
+                    max_seqlen_q=max_seqlen_in_batch_q,
+                    max_seqlen_k=max_seqlen_in_batch_k,
+                    dropout_p=dropout,
+                    softmax_scale=softmax_scale,
+                    causal=causal,
+                )
+            else:
+                attn_output_unpad = flash_attn_varlen_func(
+                    query_states,
+                    key_states,
+                    value_states,
+                    cu_seqlens_q=cu_seqlens_q,
+                    cu_seqlens_k=cu_seqlens_k,
+                    max_seqlen_q=max_seqlen_in_batch_q,
+                    max_seqlen_k=max_seqlen_in_batch_k,
+                    dropout_p=dropout,
+                    softmax_scale=softmax_scale,
+                    causal=causal,
+                    window_size=(self.config.sliding_window, self.config.sliding_window),
+                )
+
+            attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
+        else:
+            if not use_sliding_windows:
+                attn_output = flash_attn_func(
+                    query_states,
+                    key_states,
+                    value_states,
+                    dropout,
+                    softmax_scale=softmax_scale,
+                    causal=causal,
+                )
+            else:
+                attn_output = flash_attn_func(
+                    query_states,
+                    key_states,
+                    value_states,
+                    dropout,
+                    softmax_scale=softmax_scale,
+                    causal=causal,
+                    window_size=(self.config.sliding_window, self.config.sliding_window),
+                )
+
+        return attn_output
+
+    # Copied from transformers.models.mistral.modeling_mistral.MistralFlashAttention2._upad_input
+    def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
+        batch_size, kv_seq_len, num_heads, head_dim = key_layer.shape
+
+        # On the first iteration we need to properly re-create the padding mask
+        # by slicing it on the proper place
+        if kv_seq_len != attention_mask.shape[-1]:
+            attention_mask_num_tokens = attention_mask.shape[-1]
+            attention_mask = attention_mask[:, attention_mask_num_tokens - kv_seq_len :]
+
+        indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
+
+        key_layer = index_first_axis(key_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k)
+        value_layer = index_first_axis(value_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k)
+
+        if query_length == kv_seq_len:
+            query_layer = index_first_axis(
+                query_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k
+            )
+            cu_seqlens_q = cu_seqlens_k
+            max_seqlen_in_batch_q = max_seqlen_in_batch_k
+            indices_q = indices_k
+        elif query_length == 1:
+            max_seqlen_in_batch_q = 1
+            cu_seqlens_q = torch.arange(
+                batch_size + 1, dtype=torch.int32, device=query_layer.device
+            )  # There is a memcpy here, that is very bad.
+            indices_q = cu_seqlens_q[:-1]
+            query_layer = query_layer.squeeze(1)
+        else:
+            # The -q_len: slice assumes left padding.
+            attention_mask = attention_mask[:, -query_length:]
+            query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
+
+        return (
+            query_layer,
+            key_layer,
+            value_layer,
+            indices_q,
+            (cu_seqlens_q, cu_seqlens_k),
+            (max_seqlen_in_batch_q, max_seqlen_in_batch_k),
+        )
+
+
+# copied from transformers.models.llama.modeling_llama.LlamaSdpaAttention with Llama->Phi3
+# TODO @Arthur no longer copied from LLama after static cache
+class Phi3SdpaAttention(Phi3Attention):
+    """
+    Phi3 attention module using torch.nn.functional.scaled_dot_product_attention. This module inherits from
+    `Phi3Attention` as the weights of the module stays untouched. The only changes are on the forward pass to adapt to
+    SDPA API.
+    """
+
+    # Adapted from Phi3Attention.forward
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Cache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        if output_attentions:
+            # TODO: Improve this warning with e.g. `model.config.attn_implementation = "manual"` once this is implemented.
+            logger.warning_once(
+                "Phi3Model is using Phi3SdpaAttention, but `torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to the manual attention implementation, "
+                'but specifying the manual implementation will be required from Transformers version v5.0.0 onwards. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
+            )
+            return super().forward(
+                hidden_states=hidden_states,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_value,
+                output_attentions=output_attentions,
+                use_cache=use_cache,
+            )
+
+        bsz, q_len, _ = hidden_states.size()
+
+        qkv = self.qkv_proj(hidden_states)
+        query_pos = self.num_heads * self.head_dim
+        query_states = qkv[..., :query_pos]
+        key_states = qkv[..., query_pos : query_pos + self.num_key_value_heads * self.head_dim]
+        value_states = qkv[..., query_pos + self.num_key_value_heads * self.head_dim :]
+
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value.get_usable_length(kv_seq_len, self.layer_idx)
+        cos, sin = self.rotary_emb(value_states, position_ids, seq_len=kv_seq_len)
+
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+
+        if past_key_value is not None:
+            cache_kwargs = {"sin": sin, "cos": cos}  # Specific to RoPE models
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+
+        # SDPA with memory-efficient backend is currently (torch==2.1.2) bugged with non-contiguous inputs with custom attn_mask,
+        # Reference: https://github.com/pytorch/pytorch/issues/112577.
+        if query_states.device.type == "cuda" and attention_mask is not None:
+            query_states = query_states.contiguous()
+            key_states = key_states.contiguous()
+            value_states = value_states.contiguous()
+
+        attn_output = torch.nn.functional.scaled_dot_product_attention(
+            query_states,
+            key_states,
+            value_states,
+            attn_mask=attention_mask,
+            dropout_p=self.attention_dropout if self.training else 0.0,
+            # The q_len > 1 is necessary to match with AttentionMaskConverter.to_causal_4d that does not create a causal mask in case q_len == 1.
+            is_causal=self.is_causal and attention_mask is None and q_len > 1,
+        )
+
+        attn_output = attn_output.transpose(1, 2).contiguous()
+        attn_output = attn_output.view(bsz, q_len, self.hidden_size)
+
+        attn_output = self.o_proj(attn_output)
+
+        return attn_output, None, past_key_value
+
+
+PHI3_ATTENTION_CLASSES = {
+    "eager": Phi3Attention,
+    "flash_attention_2": Phi3FlashAttention2,
+    "sdpa": Phi3SdpaAttention,
+}
+
+
+class Phi3DecoderLayer(nn.Module):
+    def __init__(self, config: Phi3Config, layer_idx: int):
+        super().__init__()
+
+        self.config = config
+        self.self_attn = PHI3_ATTENTION_CLASSES[config._attn_implementation](config, layer_idx=layer_idx)
+
+        self.mlp = Phi3MLP(config)
+        self.input_layernorm = Phi3RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+        self.resid_attn_dropout = nn.Dropout(config.resid_pdrop)
+        self.resid_mlp_dropout = nn.Dropout(config.resid_pdrop)
+        self.post_attention_layernorm = Phi3RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+        **kwargs,
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        if "padding_mask" in kwargs:
+            warnings.warn(
+                "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
+            )
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`):
+                input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
+                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
+            position_ids (`torch.LongTensor` of shape `({0})`, *optional*):
+                Indices of positions of each input sequence tokens in the position embeddings. Selected in the range
+                `[0, config.n_positions - 1]`. [What are position IDs?](../glossary#position-ids)
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            use_cache (`bool`, *optional*):
+                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
+                (see `past_key_values`).
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+        """
+
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+
+        # Self Attention
+        attn_outputs, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+        )
+
+        hidden_states = residual + self.resid_attn_dropout(attn_outputs)
+
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + self.resid_mlp_dropout(hidden_states)
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+
+        return outputs
+
+
+PHI3_START_DOCSTRING = r"""
+    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
+    etc.)
+
+    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`Phi3Config`]):
+            Model configuration class with all the parameters of the model. Initializing with a config file does not
+            load the weights associated with the model, only the configuration. Check out the
+            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+
+@add_start_docstrings(
+    "The bare Phi-3 model outputting raw hidden-states without any specific head on top.",
+    PHI3_START_DOCSTRING,
+)
+class Phi3PreTrainedModel(PreTrainedModel):
+    config_class = Phi3Config
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["Phi3DecoderLayer"]
+    _skip_keys_device_placement = "past_key_values"
+    _supports_flash_attn_2 = True
+    _supports_sdpa = False
+    _supports_cache_class = True
+
+    _version = "0.0.5"
+
+    def _init_weights(self, module):
+        std = self.config.initializer_range
+        if isinstance(module, nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+
+PHI3_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+            Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
+            it.
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            If `past_key_values` is used, optionally only the last `input_ids` have to be input (see
+            `past_key_values`).
+
+            If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`]
+            and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more
+            information on the default strategy.
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+        position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+            config.n_positions - 1]`.
+
+            [What are position IDs?](../glossary#position-ids)
+        past_key_values (`Cache` or `tuple(tuple(torch.FloatTensor))`, *optional*):
+            Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
+            blocks) that can be used to speed up sequential decoding. This typically consists in the `past_key_values`
+            returned by the model at a previous stage of decoding, when `use_cache=True` or `config.use_cache=True`.
+
+            Two formats are allowed:
+            - a [`~cache_utils.Cache`] instance;
+            - Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of
+            shape `(batch_size, num_heads, sequence_length, embed_size_per_head)`). This is also known as the legacy
+            cache format.
+
+            The model will output the same cache format that is fed as input. If no `past_key_values` are passed, the
+            legacy cache format will be returned.
+
+            If `past_key_values` are used, the user can optionally input only the last `input_ids` (those that don't
+            have their past key value states given to this model) of shape `(batch_size, 1)` instead of all `input_ids`
+            of shape `(batch_size, sequence_length)`.
+        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
+            model's internal embedding lookup matrix.
+        use_cache (`bool`, *optional*):
+            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
+            `past_key_values`).
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare Phi-3 model outputting raw hidden-states without any specific head on top.",
+    PHI3_START_DOCSTRING,
+)
+class Phi3Model(Phi3PreTrainedModel):
+    """
+    Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`Phi3DecoderLayer`]
+
+    Args:
+        config: Phi3Config
+    """
+
+    def __init__(self, config: Phi3Config):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+        self.embed_dropout = nn.Dropout(config.embd_pdrop)
+        self.layers = nn.ModuleList(
+            [Phi3DecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self._attn_implementation = config._attn_implementation
+        self.norm = Phi3RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+        self.gradient_checkpointing = False
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+    @add_start_docstrings_to_model_forward(PHI3_INPUTS_DOCSTRING)
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # retrieve input_ids and inputs_embeds
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            batch_size, seq_length = input_ids.shape[:2]
+        elif inputs_embeds is not None:
+            batch_size, seq_length = inputs_embeds.shape[:2]
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        past_key_values_length = 0
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
+        if use_cache:
+            use_legacy_cache = not isinstance(past_key_values, Cache)
+            if use_legacy_cache:
+                past_key_values = DynamicCache.from_legacy_cache(past_key_values)
+            past_key_values_length = past_key_values.get_usable_length(seq_length)
+
+        if position_ids is None:
+            device = input_ids.device if input_ids is not None else inputs_embeds.device
+            position_ids = torch.arange(
+                past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
+            )
+            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+        else:
+            position_ids = position_ids.view(-1, seq_length).long()
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        if attention_mask is not None and self._attn_implementation == "flash_attention_2" and use_cache:
+            is_padding_right = attention_mask[:, -1].sum().item() != batch_size
+            if is_padding_right:
+                raise ValueError(
+                    "You are attempting to perform batched generation with padding_side='right'"
+                    " this may lead to unexpected behaviour for Flash Attention version of Phi3. Make sure to "
+                    " call `tokenizer.padding_side  = 'left'` before tokenizing the input. "
+                )
+
+        if self._attn_implementation == "flash_attention_2":
+            # 2d mask is passed through the layers
+            attention_mask = attention_mask if (attention_mask is not None and 0 in attention_mask) else None
+        else:
+            # 4d mask is passed through the layers
+            attention_mask = _prepare_4d_causal_attention_mask(
+                attention_mask,
+                (batch_size, seq_length),
+                inputs_embeds,
+                past_key_values_length,
+                sliding_window=self.config.sliding_window,
+            )
+
+        hidden_states = inputs_embeds
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        next_decoder_cache = None
+
+        for decoder_layer in self.layers:
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+                layer_outputs = self._gradient_checkpointing_func(
+                    decoder_layer.__call__,
+                    hidden_states,
+                    attention_mask,
+                    position_ids,
+                    past_key_values,
+                    output_attentions,
+                    use_cache,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    position_ids=position_ids,
+                    past_key_value=past_key_values,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                )
+
+            hidden_states = layer_outputs[0]
+
+            if use_cache:
+                next_decoder_cache = layer_outputs[2 if output_attentions else 1]
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        next_cache = None
+        if use_cache:
+            next_cache = next_decoder_cache.to_legacy_cache() if use_legacy_cache else next_decoder_cache
+        if not return_dict:
+            return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+        )
+
+
+class Phi3ForCausalLM(Phi3PreTrainedModel):
+    _tied_weights_keys = ["lm_head.weight"]
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaForCausalLM.__init__ with Llama->Phi3
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = Phi3Model(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaForCausalLM.get_input_embeddings
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaForCausalLM.set_input_embeddings
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaForCausalLM.get_output_embeddings
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaForCausalLM.set_output_embeddings
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaForCausalLM.set_decoder
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaForCausalLM.get_decoder
+    def get_decoder(self):
+        return self.model
+
+    # Ignore copy
+    @add_start_docstrings_to_model_forward(PHI3_INPUTS_DOCSTRING)
+    @replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        r"""
+        Args:
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, Phi3ForCausalLM
+
+        >>> model = Phi3ForCausalLM.from_pretrained("microsoft/phi-3-mini-4k-instruct")
+        >>> tokenizer = AutoTokenizer.from_pretrained("microsoft/phi-3-mini-4k-instruct")
+
+        >>> prompt = "This is an example script ."
+        >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        'This is an example script .\n Certainly! Below is a sample script that demonstrates a simple task, such as calculating the sum'
+        ```"""
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        hidden_states = outputs[0]
+        logits = self.lm_head(hidden_states)
+        logits = logits.float()
+
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            # Enable model parallelism
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    # Copied from transformers.models.persimmon.modeling_persimmon.PersimmonForCausalLM.prepare_inputs_for_generation
+    def prepare_inputs_for_generation(
+        self, input_ids, past_key_values=None, attention_mask=None, inputs_embeds=None, **kwargs
+    ):
+        if past_key_values is not None:
+            if isinstance(past_key_values, Cache):
+                cache_length = past_key_values.get_seq_length()
+                past_length = past_key_values.seen_tokens
+                max_cache_length = past_key_values.get_max_length()
+            else:
+                cache_length = past_length = past_key_values[0][0].shape[2]
+                max_cache_length = None
+
+            # Keep only the unprocessed tokens:
+            # 1 - If the length of the attention_mask exceeds the length of input_ids, then we are in a setting where
+            # some of the inputs are exclusively passed as part of the cache (e.g. when passing input_embeds as
+            # input)
+            if attention_mask is not None and attention_mask.shape[1] > input_ids.shape[1]:
+                input_ids = input_ids[:, -(attention_mask.shape[1] - past_length) :]
+            # 2 - If the past_length is smaller than input_ids', then input_ids holds all input tokens. We can discard
+            # input_ids based on the past_length.
+            elif past_length < input_ids.shape[1]:
+                input_ids = input_ids[:, past_length:]
+            # 3 - Otherwise (past_length >= input_ids.shape[1]), let's assume input_ids only has unprocessed tokens.
+
+            # If we are about to go beyond the maximum cache length, we need to crop the input attention mask.
+            if (
+                max_cache_length is not None
+                and attention_mask is not None
+                and cache_length + input_ids.shape[1] > max_cache_length
+            ):
+                attention_mask = attention_mask[:, -max_cache_length:]
+
+        position_ids = kwargs.get("position_ids", None)
+        if attention_mask is not None and position_ids is None:
+            # create position_ids on the fly for batch generation
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+            if past_key_values:
+                position_ids = position_ids[:, -input_ids.shape[1] :]
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids}
+
+        model_inputs.update(
+            {
+                "position_ids": position_ids,
+                "past_key_values": past_key_values,
+                "use_cache": kwargs.get("use_cache"),
+                "attention_mask": attention_mask,
+            }
+        )
+        return model_inputs
+
+    @staticmethod
+    # Copied from transformers.models.llama.modeling_llama.LlamaForCausalLM._reorder_cache
+    def _reorder_cache(past_key_values, beam_idx):
+        reordered_past = ()
+        for layer_past in past_key_values:
+            reordered_past += (
+                tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past),
+            )
+        return reordered_past
+
+
+@add_start_docstrings(
+    """
+    The [`Phi3Model`] with a sequence classification head on top (linear layer).
+
+    [`Phi3ForSequenceClassification`] uses the last token in order to do the classification, as other causal models
+    (e.g. GPT-2) do.
+
+    Since it does classification on the last token, it requires to know the position of the last token. If a
+    `pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If
+    no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the
+    padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in
+    each row of the batch).
+    """,
+    PHI3_START_DOCSTRING,
+)
+# Copied from transformers.models.llama.modeling_llama.LlamaForSequenceClassification with Llama->Phi3, LLAMA->PHI3, self.transformer->self.model, transformer_outputs->model_outputs
+class Phi3ForSequenceClassification(Phi3PreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+        self.model = Phi3Model(config)
+        self.score = nn.Linear(config.hidden_size, self.num_labels, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    @add_start_docstrings_to_model_forward(PHI3_INPUTS_DOCSTRING)
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
+            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
+            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        model_outputs = self.model(
+            input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        hidden_states = model_outputs[0]
+        logits = self.score(hidden_states)
+
+        if input_ids is not None:
+            batch_size = input_ids.shape[0]
+        else:
+            batch_size = inputs_embeds.shape[0]
+
+        if self.config.pad_token_id is None and batch_size != 1:
+            raise ValueError("Cannot handle batch sizes > 1 if no padding token is defined.")
+        if self.config.pad_token_id is None:
+            sequence_lengths = -1
+        else:
+            if input_ids is not None:
+                # if no pad token found, use modulo instead of reverse indexing for ONNX compatibility
+                sequence_lengths = torch.eq(input_ids, self.config.pad_token_id).int().argmax(-1) - 1
+                sequence_lengths = sequence_lengths % input_ids.shape[-1]
+                sequence_lengths = sequence_lengths.to(logits.device)
+            else:
+                sequence_lengths = -1
+
+        pooled_logits = logits[torch.arange(batch_size, device=logits.device), sequence_lengths]
+
+        loss = None
+        if labels is not None:
+            labels = labels.to(logits.device)
+            if self.config.problem_type is None:
+                if self.num_labels == 1:
+                    self.config.problem_type = "regression"
+                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
+                    self.config.problem_type = "single_label_classification"
+                else:
+                    self.config.problem_type = "multi_label_classification"
+
+            if self.config.problem_type == "regression":
+                loss_fct = MSELoss()
+                if self.num_labels == 1:
+                    loss = loss_fct(pooled_logits.squeeze(), labels.squeeze())
+                else:
+                    loss = loss_fct(pooled_logits, labels)
+            elif self.config.problem_type == "single_label_classification":
+                loss_fct = CrossEntropyLoss()
+                loss = loss_fct(pooled_logits.view(-1, self.num_labels), labels.view(-1))
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = BCEWithLogitsLoss()
+                loss = loss_fct(pooled_logits, labels)
+        if not return_dict:
+            output = (pooled_logits,) + model_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return SequenceClassifierOutputWithPast(
+            loss=loss,
+            logits=pooled_logits,
+            past_key_values=model_outputs.past_key_values,
+            hidden_states=model_outputs.hidden_states,
+            attentions=model_outputs.attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    [`Phi3Model`] with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for
+    Named-Entity-Recognition (NER) tasks.
+    """,
+    PHI3_START_DOCSTRING,
+)
+# Copied from transformers.models.mpt.modeling_mpt.MptForTokenClassification with Mpt->Phi3,MPT->PHI3,self.transformer->self.model,transformer_outputs->model_outputs
+class Phi3ForTokenClassification(Phi3PreTrainedModel):
+    def __init__(self, config: Phi3Config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+
+        self.model = Phi3Model(config)
+        if hasattr(config, "classifier_dropout") and config.classifier_dropout is not None:
+            classifier_dropout = config.classifier_dropout
+        elif hasattr(config, "hidden_dropout") and config.hidden_dropout is not None:
+            classifier_dropout = config.hidden_dropout
+        else:
+            classifier_dropout = 0.1
+        self.dropout = nn.Dropout(classifier_dropout)
+        self.classifier = nn.Linear(config.hidden_size, config.num_labels)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @add_start_docstrings_to_model_forward(PHI3_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=TokenClassifierOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        labels: Optional[torch.Tensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        **deprecated_arguments,
+    ) -> Union[Tuple[torch.Tensor], TokenClassifierOutput]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
+            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
+            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        model_outputs = self.model(
+            input_ids,
+            past_key_values=past_key_values,
+            attention_mask=attention_mask,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        hidden_states = model_outputs[0]
+        hidden_states = self.dropout(hidden_states)
+        logits = self.classifier(hidden_states)
+
+        loss = None
+        if labels is not None:
+            # move labels to correct device to enable model parallelism
+            labels = labels.to(logits.device)
+            batch_size, seq_length = labels.shape
+            loss_fct = CrossEntropyLoss()
+            loss = loss_fct(
+                logits.view(batch_size * seq_length, self.num_labels), labels.view(batch_size * seq_length)
+            )
+
+        if not return_dict:
+            output = (logits,) + model_outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return TokenClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=model_outputs.hidden_states,
+            attentions=model_outputs.attentions,
+        )

special_tokens_map.json

@@ -0,0 +1,33 @@
+{
+  "additional_special_tokens": [
+    "<|/inst|>"
+  ],
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723

tokenizer_config.json

@@ -0,0 +1,350 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": false
+    },
+    "32000": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32001": {
+      "content": "<|assistant|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32002": {
+      "content": "<|step|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32003": {
+      "content": "<|function_output|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32004": {
+      "content": "<|tag|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32005": {
+      "content": "<|function_call|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32006": {
+      "content": "<|system|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32007": {
+      "content": "<|end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32008": {
+      "content": "<|raw|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32009": {
+      "content": "<|continue|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32010": {
+      "content": "<|user|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32011": {
+      "content": "<|function_list|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32012": {
+      "content": "<|calc|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32013": {
+      "content": "<|code|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32014": {
+      "content": "<|/code|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32015": {
+      "content": "<|summary|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32016": {
+      "content": "<|resource|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32017": {
+      "content": "<|assistant_mask|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32018": {
+      "content": "<|start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32019": {
+      "content": "<|message|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32020": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32021": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32022": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32023": {
+      "content": "<|meta_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32024": {
+      "content": "<|ipynb_marker|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32025": {
+      "content": "<|diff_marker|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32026": {
+      "content": "<|ghissue|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32027": {
+      "content": "<|ghreview|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32028": {
+      "content": "<|disc_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32029": {
+      "content": "<|disc_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32030": {
+      "content": "<|disc_thread|><|query|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32031": {
+      "content": "<|/query|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32032": {
+      "content": "<|data|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32033": {
+      "content": "<|/data|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32034": {
+      "content": "<|sys|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32035": {
+      "content": "<|/sys|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32036": {
+      "content": "<|inst|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32037": {
+      "content": "<|/inst|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<|/inst|>"
+  ],
+  "bos_token": "<s>",
+  "chat_template": "{% if messages[0]['role'] == 'system' %}{% set system_message = messages[0]['content'] %}{% endif %}{% if system_message is defined %}{{ system_message + '\\n' }}{% endif %}{% for message in messages %}{% set content = message['content'] %}{% if message['role'] == 'user' %}{{ 'Human: ' + content + '\\nAssistant: ' }}{% elif message['role'] == 'assistant' %}{{ content + '<|endoftext|>' + '\\n' }}{% endif %}{% endfor %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "legacy": false,
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "split_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 2.994601079784043,
+    "total_flos": 132590267662336.0,
+    "train_loss": 0.7937506708579186,
+    "train_runtime": 49781.9259,
+    "train_samples_per_second": 1.205,
+    "train_steps_per_second": 0.025
+}

trainer_state.json

@@ -0,0 +1,2138 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.994601079784043,
+  "eval_steps": 500,
+  "global_step": 1248,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.02399520095980804,
+      "grad_norm": 24.58741331565172,
+      "learning_rate": 1.0000000000000002e-06,
+      "logits/chosen": -0.5075146555900574,
+      "logits/rejected": -0.31934085488319397,
+      "logps/chosen": -1.394007921218872,
+      "logps/rejected": -1.3630257844924927,
+      "loss": 1.3501,
+      "odds_ratio_loss": 0.8239962458610535,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.06970040500164032,
+      "rewards/margins": -0.0015491036465391517,
+      "rewards/rejected": -0.06815129518508911,
+      "sft_loss": 1.394007921218872,
+      "step": 10
+    },
+    {
+      "epoch": 0.04799040191961608,
+      "grad_norm": 4.281683015852783,
+      "learning_rate": 3.5e-06,
+      "logits/chosen": 0.08614908158779144,
+      "logits/rejected": 0.3013238310813904,
+      "logps/chosen": -1.3080074787139893,
+      "logps/rejected": -1.334457278251648,
+      "loss": 1.2858,
+      "odds_ratio_loss": 0.7804475426673889,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.0654003769159317,
+      "rewards/margins": 0.0013224859721958637,
+      "rewards/rejected": -0.06672286242246628,
+      "sft_loss": 1.3080074787139893,
+      "step": 20
+    },
+    {
+      "epoch": 0.07198560287942411,
+      "grad_norm": 3.830958349381369,
+      "learning_rate": 4.99986910314335e-06,
+      "logits/chosen": 0.3485943675041199,
+      "logits/rejected": 0.6042150855064392,
+      "logps/chosen": -0.9540683627128601,
+      "logps/rejected": -1.1750730276107788,
+      "loss": 0.9904,
+      "odds_ratio_loss": 0.6533687710762024,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.047703422605991364,
+      "rewards/margins": 0.011050237342715263,
+      "rewards/rejected": -0.05875365808606148,
+      "sft_loss": 0.9540683627128601,
+      "step": 30
+    },
+    {
+      "epoch": 0.09598080383923216,
+      "grad_norm": 3.6776666943951675,
+      "learning_rate": 4.998396670920005e-06,
+      "logits/chosen": 0.17601105570793152,
+      "logits/rejected": 0.5272272229194641,
+      "logps/chosen": -0.898045539855957,
+      "logps/rejected": -1.0136868953704834,
+      "loss": 0.9614,
+      "odds_ratio_loss": 0.6860688328742981,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.04490227997303009,
+      "rewards/margins": 0.005782057531177998,
+      "rewards/rejected": -0.05068434029817581,
+      "sft_loss": 0.898045539855957,
+      "step": 40
+    },
+    {
+      "epoch": 0.11997600479904019,
+      "grad_norm": 2.636908991979515,
+      "learning_rate": 4.995289152254744e-06,
+      "logits/chosen": 0.2309066355228424,
+      "logits/rejected": 0.22152824699878693,
+      "logps/chosen": -0.9074997901916504,
+      "logps/rejected": -1.0551084280014038,
+      "loss": 0.9374,
+      "odds_ratio_loss": 0.663613498210907,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": -0.04537498578429222,
+      "rewards/margins": 0.007380434311926365,
+      "rewards/rejected": -0.05275542289018631,
+      "sft_loss": 0.9074997901916504,
+      "step": 50
+    },
+    {
+      "epoch": 0.14397120575884823,
+      "grad_norm": 1.8300107701302537,
+      "learning_rate": 4.990548580876516e-06,
+      "logits/chosen": 0.307407021522522,
+      "logits/rejected": 0.37507694959640503,
+      "logps/chosen": -0.9279610514640808,
+      "logps/rejected": -0.986476719379425,
+      "loss": 0.9464,
+      "odds_ratio_loss": 0.7063499093055725,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.04639805108308792,
+      "rewards/margins": 0.00292578199878335,
+      "rewards/rejected": -0.04932383447885513,
+      "sft_loss": 0.9279610514640808,
+      "step": 60
+    },
+    {
+      "epoch": 0.16796640671865626,
+      "grad_norm": 3.8157191209486507,
+      "learning_rate": 4.9841780592726385e-06,
+      "logits/chosen": 0.19509825110435486,
+      "logits/rejected": 0.2650177776813507,
+      "logps/chosen": -0.9848098754882812,
+      "logps/rejected": -1.0149097442626953,
+      "loss": 0.9578,
+      "odds_ratio_loss": 0.726799488067627,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.04924049228429794,
+      "rewards/margins": 0.0015049913199618459,
+      "rewards/rejected": -0.050745487213134766,
+      "sft_loss": 0.9848098754882812,
+      "step": 70
+    },
+    {
+      "epoch": 0.19196160767846432,
+      "grad_norm": 4.078587531391316,
+      "learning_rate": 4.976181756658363e-06,
+      "logits/chosen": 0.061622969806194305,
+      "logits/rejected": 0.2444450408220291,
+      "logps/chosen": -0.8894473910331726,
+      "logps/rejected": -1.0614734888076782,
+      "loss": 0.9675,
+      "odds_ratio_loss": 0.6382969617843628,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.04447237029671669,
+      "rewards/margins": 0.008601305074989796,
+      "rewards/rejected": -0.05307367444038391,
+      "sft_loss": 0.8894473910331726,
+      "step": 80
+    },
+    {
+      "epoch": 0.21595680863827235,
+      "grad_norm": 2.9874023740770363,
+      "learning_rate": 4.9665649062483115e-06,
+      "logits/chosen": 0.6337467432022095,
+      "logits/rejected": 0.7902036905288696,
+      "logps/chosen": -0.9439412951469421,
+      "logps/rejected": -0.9588793516159058,
+      "loss": 0.9635,
+      "odds_ratio_loss": 0.7716476917266846,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.047197069972753525,
+      "rewards/margins": 0.0007468975381925702,
+      "rewards/rejected": -0.047943972051143646,
+      "sft_loss": 0.9439412951469421,
+      "step": 90
+    },
+    {
+      "epoch": 0.23995200959808038,
+      "grad_norm": 2.3029148332001745,
+      "learning_rate": 4.955333801831578e-06,
+      "logits/chosen": 0.49920982122421265,
+      "logits/rejected": 0.6337569355964661,
+      "logps/chosen": -0.8333128094673157,
+      "logps/rejected": -1.059599757194519,
+      "loss": 0.9453,
+      "odds_ratio_loss": 0.6517213582992554,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.041665639728307724,
+      "rewards/margins": 0.011314347386360168,
+      "rewards/rejected": -0.05297998711466789,
+      "sft_loss": 0.8333128094673157,
+      "step": 100
+    },
+    {
+      "epoch": 0.26394721055788845,
+      "grad_norm": 2.8766587489414395,
+      "learning_rate": 4.9424957936527295e-06,
+      "logits/chosen": -0.28645992279052734,
+      "logits/rejected": 0.04107431694865227,
+      "logps/chosen": -0.9429195523262024,
+      "logps/rejected": -0.9936224222183228,
+      "loss": 0.9526,
+      "odds_ratio_loss": 0.705885112285614,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.04714598134160042,
+      "rewards/margins": 0.002535139676183462,
+      "rewards/rejected": -0.04968111589550972,
+      "sft_loss": 0.9429195523262024,
+      "step": 110
+    },
+    {
+      "epoch": 0.28794241151769645,
+      "grad_norm": 2.1411106644617703,
+      "learning_rate": 4.92805928360141e-06,
+      "logits/chosen": -0.29608479142189026,
+      "logits/rejected": -0.21111997961997986,
+      "logps/chosen": -0.888851523399353,
+      "logps/rejected": -1.0842912197113037,
+      "loss": 0.8904,
+      "odds_ratio_loss": 0.5968859195709229,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.04444257169961929,
+      "rewards/margins": 0.009771987795829773,
+      "rewards/rejected": -0.054214559495449066,
+      "sft_loss": 0.888851523399353,
+      "step": 120
+    },
+    {
+      "epoch": 0.3119376124775045,
+      "grad_norm": 2.1891227152981347,
+      "learning_rate": 4.912033719713687e-06,
+      "logits/chosen": 0.49228960275650024,
+      "logits/rejected": 0.5680336952209473,
+      "logps/chosen": -0.9152839779853821,
+      "logps/rejected": -1.0058788061141968,
+      "loss": 0.9427,
+      "odds_ratio_loss": 0.6943625807762146,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.04576420038938522,
+      "rewards/margins": 0.004529745317995548,
+      "rewards/rejected": -0.0502939410507679,
+      "sft_loss": 0.9152839779853821,
+      "step": 130
+    },
+    {
+      "epoch": 0.3359328134373125,
+      "grad_norm": 2.5131225459939,
+      "learning_rate": 4.894429589988739e-06,
+      "logits/chosen": -1.2468726634979248,
+      "logits/rejected": -1.0485397577285767,
+      "logps/chosen": -1.0104249715805054,
+      "logps/rejected": -1.0477244853973389,
+      "loss": 0.949,
+      "odds_ratio_loss": 0.7160865068435669,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.05052124708890915,
+      "rewards/margins": 0.0018649749690666795,
+      "rewards/rejected": -0.05238622426986694,
+      "sft_loss": 1.0104249715805054,
+      "step": 140
+    },
+    {
+      "epoch": 0.3599280143971206,
+      "grad_norm": 2.696319834123575,
+      "learning_rate": 4.875258415524945e-06,
+      "logits/chosen": 0.039508234709501266,
+      "logits/rejected": 0.23594827950000763,
+      "logps/chosen": -0.904223620891571,
+      "logps/rejected": -1.032157063484192,
+      "loss": 0.9533,
+      "odds_ratio_loss": 0.6739581823348999,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.04521118476986885,
+      "rewards/margins": 0.0063966671004891396,
+      "rewards/rejected": -0.051607854664325714,
+      "sft_loss": 0.904223620891571,
+      "step": 150
+    },
+    {
+      "epoch": 0.38392321535692864,
+      "grad_norm": 2.241170193835809,
+      "learning_rate": 4.85453274297985e-06,
+      "logits/chosen": 0.4507044851779938,
+      "logits/rejected": 0.7088828682899475,
+      "logps/chosen": -0.9252007603645325,
+      "logps/rejected": -1.0105345249176025,
+      "loss": 0.9187,
+      "odds_ratio_loss": 0.6664329171180725,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.0462600402534008,
+      "rewards/margins": 0.004266692791134119,
+      "rewards/rejected": -0.050526730716228485,
+      "sft_loss": 0.9252007603645325,
+      "step": 160
+    },
+    {
+      "epoch": 0.40791841631673664,
+      "grad_norm": 1.759854296483571,
+      "learning_rate": 4.832266136358951e-06,
+      "logits/chosen": -0.12876208126544952,
+      "logits/rejected": 0.014335835352540016,
+      "logps/chosen": -0.8540490865707397,
+      "logps/rejected": -0.9863293766975403,
+      "loss": 0.926,
+      "odds_ratio_loss": 0.6714656352996826,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.04270245134830475,
+      "rewards/margins": 0.006614011712372303,
+      "rewards/rejected": -0.04931646212935448,
+      "sft_loss": 0.8540490865707397,
+      "step": 170
+    },
+    {
+      "epoch": 0.4319136172765447,
+      "grad_norm": 2.793191882203603,
+      "learning_rate": 4.808473168138675e-06,
+      "logits/chosen": 0.3617595136165619,
+      "logits/rejected": 0.3396950364112854,
+      "logps/chosen": -0.8613064885139465,
+      "logps/rejected": -1.0067331790924072,
+      "loss": 0.9162,
+      "odds_ratio_loss": 0.6582903861999512,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.04306532442569733,
+      "rewards/margins": 0.007271329872310162,
+      "rewards/rejected": -0.050336651504039764,
+      "sft_loss": 0.8613064885139465,
+      "step": 180
+    },
+    {
+      "epoch": 0.4559088182363527,
+      "grad_norm": 1.7774141067161418,
+      "learning_rate": 4.783169409729363e-06,
+      "logits/chosen": 0.9685203433036804,
+      "logits/rejected": 1.1009634733200073,
+      "logps/chosen": -0.8521540760993958,
+      "logps/rejected": -0.9150575399398804,
+      "loss": 0.9004,
+      "odds_ratio_loss": 0.7224193811416626,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.04260770231485367,
+      "rewards/margins": 0.0031451724935323,
+      "rewards/rejected": -0.0457528755068779,
+      "sft_loss": 0.8521540760993958,
+      "step": 190
+    },
+    {
+      "epoch": 0.47990401919616077,
+      "grad_norm": 2.052107783396207,
+      "learning_rate": 4.756371421284482e-06,
+      "logits/chosen": 0.33597105741500854,
+      "logits/rejected": 0.44187426567077637,
+      "logps/chosen": -0.8725342750549316,
+      "logps/rejected": -0.9003400802612305,
+      "loss": 0.919,
+      "odds_ratio_loss": 0.7135496735572815,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.04362671449780464,
+      "rewards/margins": 0.0013902939390391111,
+      "rewards/rejected": -0.04501700773835182,
+      "sft_loss": 0.8725342750549316,
+      "step": 200
+    },
+    {
+      "epoch": 0.5038992201559688,
+      "grad_norm": 2.3000145040966973,
+      "learning_rate": 4.728096740862778e-06,
+      "logits/chosen": 0.16287042200565338,
+      "logits/rejected": 0.35098087787628174,
+      "logps/chosen": -0.8514264822006226,
+      "logps/rejected": -0.9913795590400696,
+      "loss": 0.9096,
+      "odds_ratio_loss": 0.6634506583213806,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.042571328580379486,
+      "rewards/margins": 0.006997650023549795,
+      "rewards/rejected": -0.04956897348165512,
+      "sft_loss": 0.8514264822006226,
+      "step": 210
+    },
+    {
+      "epoch": 0.5278944211157769,
+      "grad_norm": 1.581079267248328,
+      "learning_rate": 4.698363872950406e-06,
+      "logits/chosen": 0.298981636762619,
+      "logits/rejected": 0.49268895387649536,
+      "logps/chosen": -0.8895601034164429,
+      "logps/rejected": -1.026539921760559,
+      "loss": 0.8744,
+      "odds_ratio_loss": 0.6685082316398621,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.04447800666093826,
+      "rewards/margins": 0.0068489923141896725,
+      "rewards/rejected": -0.051326997578144073,
+      "sft_loss": 0.8895601034164429,
+      "step": 220
+    },
+    {
+      "epoch": 0.5518896220755849,
+      "grad_norm": 1.7094822098553022,
+      "learning_rate": 4.6671922763505915e-06,
+      "logits/chosen": 0.34609514474868774,
+      "logits/rejected": 0.5052930116653442,
+      "logps/chosen": -0.863084614276886,
+      "logps/rejected": -0.9836879968643188,
+      "loss": 0.8905,
+      "odds_ratio_loss": 0.6813028454780579,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.043154239654541016,
+      "rewards/margins": 0.006030158139765263,
+      "rewards/rejected": -0.049184400588274,
+      "sft_loss": 0.863084614276886,
+      "step": 230
+    },
+    {
+      "epoch": 0.5758848230353929,
+      "grad_norm": 1.9367159826113498,
+      "learning_rate": 4.634602351448738e-06,
+      "logits/chosen": 0.286350816488266,
+      "logits/rejected": 0.3788919448852539,
+      "logps/chosen": -0.8919585943222046,
+      "logps/rejected": -0.9452742338180542,
+      "loss": 0.9133,
+      "odds_ratio_loss": 0.6905114650726318,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.04459793120622635,
+      "rewards/margins": 0.0026657807175070047,
+      "rewards/rejected": -0.04726371169090271,
+      "sft_loss": 0.8919585943222046,
+      "step": 240
+    },
+    {
+      "epoch": 0.5998800239952009,
+      "grad_norm": 2.0772847936555636,
+      "learning_rate": 4.6006154268613015e-06,
+      "logits/chosen": 0.4635019898414612,
+      "logits/rejected": 0.5444530248641968,
+      "logps/chosen": -0.8181222081184387,
+      "logps/rejected": -0.9908831715583801,
+      "loss": 0.8927,
+      "odds_ratio_loss": 0.6295598149299622,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.04090610891580582,
+      "rewards/margins": 0.008638045750558376,
+      "rewards/rejected": -0.04954415559768677,
+      "sft_loss": 0.8181222081184387,
+      "step": 250
+    },
+    {
+      "epoch": 0.623875224955009,
+      "grad_norm": 2.084215689408855,
+      "learning_rate": 4.565253745477187e-06,
+      "logits/chosen": 0.40253886580467224,
+      "logits/rejected": 0.4625183045864105,
+      "logps/chosen": -0.9301355481147766,
+      "logps/rejected": -1.0306508541107178,
+      "loss": 0.9162,
+      "odds_ratio_loss": 0.6872043609619141,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.04650677740573883,
+      "rewards/margins": 0.005025765858590603,
+      "rewards/rejected": -0.05153254419565201,
+      "sft_loss": 0.9301355481147766,
+      "step": 260
+    },
+    {
+      "epoch": 0.647870425914817,
+      "grad_norm": 1.9031984888179019,
+      "learning_rate": 4.528540449900799e-06,
+      "logits/chosen": 0.4078219532966614,
+      "logits/rejected": 0.6789823174476624,
+      "logps/chosen": -0.8785255551338196,
+      "logps/rejected": -0.9139087796211243,
+      "loss": 0.9176,
+      "odds_ratio_loss": 0.7333613038063049,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.04392627626657486,
+      "rewards/margins": 0.0017691642278805375,
+      "rewards/rejected": -0.04569543898105621,
+      "sft_loss": 0.8785255551338196,
+      "step": 270
+    },
+    {
+      "epoch": 0.671865626874625,
+      "grad_norm": 2.3067419173621113,
+      "learning_rate": 4.490499567306256e-06,
+      "logits/chosen": 0.304252564907074,
+      "logits/rejected": 0.5160123109817505,
+      "logps/chosen": -0.8951358795166016,
+      "logps/rejected": -0.9636558294296265,
+      "loss": 0.8917,
+      "odds_ratio_loss": 0.69621342420578,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.04475679248571396,
+      "rewards/margins": 0.0034259993117302656,
+      "rewards/rejected": -0.04818279296159744,
+      "sft_loss": 0.8951358795166016,
+      "step": 280
+    },
+    {
+      "epoch": 0.6958608278344331,
+      "grad_norm": 3.1297290877323003,
+      "learning_rate": 4.451155993712711e-06,
+      "logits/chosen": 0.25184166431427,
+      "logits/rejected": 0.43299436569213867,
+      "logps/chosen": -0.808620810508728,
+      "logps/rejected": -0.9780584573745728,
+      "loss": 0.9379,
+      "odds_ratio_loss": 0.6151310205459595,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.04043104499578476,
+      "rewards/margins": 0.008471880108118057,
+      "rewards/rejected": -0.048902928829193115,
+      "sft_loss": 0.808620810508728,
+      "step": 290
+    },
+    {
+      "epoch": 0.7198560287942412,
+      "grad_norm": 2.001570442654457,
+      "learning_rate": 4.410535477691041e-06,
+      "logits/chosen": 0.6736063957214355,
+      "logits/rejected": 0.8922637104988098,
+      "logps/chosen": -0.8743098974227905,
+      "logps/rejected": -1.0198915004730225,
+      "loss": 0.8962,
+      "odds_ratio_loss": 0.6545746326446533,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.043715499341487885,
+      "rewards/margins": 0.0072790831327438354,
+      "rewards/rejected": -0.05099458247423172,
+      "sft_loss": 0.8743098974227905,
+      "step": 300
+    },
+    {
+      "epoch": 0.7438512297540492,
+      "grad_norm": 3.088640251108737,
+      "learning_rate": 4.368664603512586e-06,
+      "logits/chosen": -0.10074709355831146,
+      "logits/rejected": 0.08682968467473984,
+      "logps/chosen": -0.7929955720901489,
+      "logps/rejected": -0.9449365735054016,
+      "loss": 0.8789,
+      "odds_ratio_loss": 0.6474851369857788,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.03964977711439133,
+      "rewards/margins": 0.007597046438604593,
+      "rewards/rejected": -0.047246821224689484,
+      "sft_loss": 0.7929955720901489,
+      "step": 310
+    },
+    {
+      "epoch": 0.7678464307138573,
+      "grad_norm": 2.278875813822025,
+      "learning_rate": 4.325570773750952e-06,
+      "logits/chosen": -0.22130906581878662,
+      "logits/rejected": -0.028980206698179245,
+      "logps/chosen": -0.8826779127120972,
+      "logps/rejected": -1.0213041305541992,
+      "loss": 0.9204,
+      "odds_ratio_loss": 0.6443883180618286,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.04413389414548874,
+      "rewards/margins": 0.006931307725608349,
+      "rewards/rejected": -0.05106520652770996,
+      "sft_loss": 0.8826779127120972,
+      "step": 320
+    },
+    {
+      "epoch": 0.7918416316736653,
+      "grad_norm": 1.6952516043840655,
+      "learning_rate": 4.281282191348289e-06,
+      "logits/chosen": 0.45927032828330994,
+      "logits/rejected": 0.6593443751335144,
+      "logps/chosen": -0.8378440141677856,
+      "logps/rejected": -0.9682254791259766,
+      "loss": 0.8995,
+      "odds_ratio_loss": 0.6620376110076904,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.04189220070838928,
+      "rewards/margins": 0.006519075483083725,
+      "rewards/rejected": -0.04841126874089241,
+      "sft_loss": 0.8378440141677856,
+      "step": 330
+    },
+    {
+      "epoch": 0.8158368326334733,
+      "grad_norm": 2.4806806819218794,
+      "learning_rate": 4.235827841157748e-06,
+      "logits/chosen": 0.01970214769244194,
+      "logits/rejected": 0.11670324951410294,
+      "logps/chosen": -0.8856766819953918,
+      "logps/rejected": -1.0817759037017822,
+      "loss": 0.8834,
+      "odds_ratio_loss": 0.6194185018539429,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.04428383335471153,
+      "rewards/margins": 0.009804959408938885,
+      "rewards/rejected": -0.054088789969682693,
+      "sft_loss": 0.8856766819953918,
+      "step": 340
+    },
+    {
+      "epoch": 0.8398320335932813,
+      "grad_norm": 1.5265892877639438,
+      "learning_rate": 4.1892374709742186e-06,
+      "logits/chosen": -0.7483745813369751,
+      "logits/rejected": -0.42045336961746216,
+      "logps/chosen": -0.7948485016822815,
+      "logps/rejected": -0.9918915033340454,
+      "loss": 0.9474,
+      "odds_ratio_loss": 0.5842909812927246,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.03974242880940437,
+      "rewards/margins": 0.009852146729826927,
+      "rewards/rejected": -0.04959457367658615,
+      "sft_loss": 0.7948485016822815,
+      "step": 350
+    },
+    {
+      "epoch": 0.8638272345530894,
+      "grad_norm": 2.1051154185205543,
+      "learning_rate": 4.141541572065762e-06,
+      "logits/chosen": 0.41192498803138733,
+      "logits/rejected": 0.5341157913208008,
+      "logps/chosen": -0.7971394658088684,
+      "logps/rejected": -0.9216561317443848,
+      "loss": 0.8881,
+      "odds_ratio_loss": 0.69920814037323,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.03985697776079178,
+      "rewards/margins": 0.0062258280813694,
+      "rewards/rejected": -0.04608280584216118,
+      "sft_loss": 0.7971394658088684,
+      "step": 360
+    },
+    {
+      "epoch": 0.8878224355128974,
+      "grad_norm": 2.049071087536336,
+      "learning_rate": 4.092771359218462e-06,
+      "logits/chosen": 0.2649831771850586,
+      "logits/rejected": 0.45568495988845825,
+      "logps/chosen": -0.8466150164604187,
+      "logps/rejected": -1.0025365352630615,
+      "loss": 0.9065,
+      "odds_ratio_loss": 0.629971444606781,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.042330749332904816,
+      "rewards/margins": 0.007796071469783783,
+      "rewards/rejected": -0.0501268208026886,
+      "sft_loss": 0.8466150164604187,
+      "step": 370
+    },
+    {
+      "epoch": 0.9118176364727054,
+      "grad_norm": 3.597524104140319,
+      "learning_rate": 4.04295875030778e-06,
+      "logits/chosen": -0.18752217292785645,
+      "logits/rejected": 0.15378537774085999,
+      "logps/chosen": -0.8704308271408081,
+      "logps/rejected": -0.9513336420059204,
+      "loss": 0.9014,
+      "odds_ratio_loss": 0.6948253512382507,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.043521542102098465,
+      "rewards/margins": 0.004045139066874981,
+      "rewards/rejected": -0.04756668210029602,
+      "sft_loss": 0.8704308271408081,
+      "step": 380
+    },
+    {
+      "epoch": 0.9358128374325135,
+      "grad_norm": 3.1405630532603395,
+      "learning_rate": 3.992136345409765e-06,
+      "logits/chosen": -0.1735876053571701,
+      "logits/rejected": -0.20124337077140808,
+      "logps/chosen": -0.9253339767456055,
+      "logps/rejected": -1.0305973291397095,
+      "loss": 0.9111,
+      "odds_ratio_loss": 0.6636070013046265,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.04626670479774475,
+      "rewards/margins": 0.005263164173811674,
+      "rewards/rejected": -0.051529865711927414,
+      "sft_loss": 0.9253339767456055,
+      "step": 390
+    },
+    {
+      "epoch": 0.9598080383923215,
+      "grad_norm": 2.4716790122788983,
+      "learning_rate": 3.940337405465786e-06,
+      "logits/chosen": 0.26361703872680664,
+      "logits/rejected": 0.44345617294311523,
+      "logps/chosen": -0.8355854153633118,
+      "logps/rejected": -1.0225704908370972,
+      "loss": 0.9062,
+      "odds_ratio_loss": 0.6545855402946472,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.04177927225828171,
+      "rewards/margins": 0.009349259547889233,
+      "rewards/rejected": -0.05112852901220322,
+      "sft_loss": 0.8355854153633118,
+      "step": 400
+    },
+    {
+      "epoch": 0.9838032393521295,
+      "grad_norm": 2.3985102639359406,
+      "learning_rate": 3.887595830514775e-06,
+      "logits/chosen": 0.21671700477600098,
+      "logits/rejected": 0.29912179708480835,
+      "logps/chosen": -0.809670090675354,
+      "logps/rejected": -1.0107569694519043,
+      "loss": 0.9029,
+      "odds_ratio_loss": 0.6326887011528015,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.0404835119843483,
+      "rewards/margins": 0.010054344311356544,
+      "rewards/rejected": -0.05053785443305969,
+      "sft_loss": 0.809670090675354,
+      "step": 410
+    },
+    {
+      "epoch": 1.0077984403119375,
+      "grad_norm": 1.6971594247197401,
+      "learning_rate": 3.833946137507195e-06,
+      "logits/chosen": 0.4990086555480957,
+      "logits/rejected": 0.616361141204834,
+      "logps/chosen": -0.8005359768867493,
+      "logps/rejected": -0.9603840708732605,
+      "loss": 0.8398,
+      "odds_ratio_loss": 0.6354148387908936,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.040026795119047165,
+      "rewards/margins": 0.007992411032319069,
+      "rewards/rejected": -0.04801920801401138,
+      "sft_loss": 0.8005359768867493,
+      "step": 420
+    },
+    {
+      "epoch": 1.0317936412717457,
+      "grad_norm": 2.2002987962167904,
+      "learning_rate": 3.779423437715274e-06,
+      "logits/chosen": 0.7601526975631714,
+      "logits/rejected": 0.8180352449417114,
+      "logps/chosen": -0.6671024560928345,
+      "logps/rejected": -0.9577730298042297,
+      "loss": 0.7742,
+      "odds_ratio_loss": 0.5807942152023315,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.03335512429475784,
+      "rewards/margins": 0.014533529989421368,
+      "rewards/rejected": -0.047888655215501785,
+      "sft_loss": 0.6671024560928345,
+      "step": 430
+    },
+    {
+      "epoch": 1.0557888422315538,
+      "grad_norm": 1.5148819350515028,
+      "learning_rate": 3.7240634137542864e-06,
+      "logits/chosen": 0.19566980004310608,
+      "logits/rejected": 0.3528198003768921,
+      "logps/chosen": -0.6874720454216003,
+      "logps/rejected": -1.0558958053588867,
+      "loss": 0.7663,
+      "odds_ratio_loss": 0.48211669921875,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.034373603761196136,
+      "rewards/margins": 0.01842118799686432,
+      "rewards/rejected": -0.052794791758060455,
+      "sft_loss": 0.6874720454216003,
+      "step": 440
+    },
+    {
+      "epoch": 1.0797840431913617,
+      "grad_norm": 1.6130353172110996,
+      "learning_rate": 3.6679022962299054e-06,
+      "logits/chosen": 0.8750432133674622,
+      "logits/rejected": 0.8553866147994995,
+      "logps/chosen": -0.7515122890472412,
+      "logps/rejected": -0.9563247561454773,
+      "loss": 0.7745,
+      "odds_ratio_loss": 0.5920617580413818,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.037575613707304,
+      "rewards/margins": 0.010240620002150536,
+      "rewards/rejected": -0.047816235572099686,
+      "sft_loss": 0.7515122890472412,
+      "step": 450
+    },
+    {
+      "epoch": 1.1037792441511698,
+      "grad_norm": 1.8444047185661667,
+      "learning_rate": 3.6109768400269336e-06,
+      "logits/chosen": 0.21664266288280487,
+      "logits/rejected": 0.3455556333065033,
+      "logps/chosen": -0.7820109128952026,
+      "logps/rejected": -1.1722263097763062,
+      "loss": 0.7949,
+      "odds_ratio_loss": 0.5249099731445312,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -0.03910055011510849,
+      "rewards/margins": 0.019510772079229355,
+      "rewards/rejected": -0.05861131474375725,
+      "sft_loss": 0.7820109128952026,
+      "step": 460
+    },
+    {
+      "epoch": 1.127774445110978,
+      "grad_norm": 1.923809039800638,
+      "learning_rate": 3.5533243002549044e-06,
+      "logits/chosen": -0.051299355924129486,
+      "logits/rejected": 0.12599964439868927,
+      "logps/chosen": -0.6766480803489685,
+      "logps/rejected": -0.9556339979171753,
+      "loss": 0.769,
+      "odds_ratio_loss": 0.5771059989929199,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.03383240848779678,
+      "rewards/margins": 0.013949294574558735,
+      "rewards/rejected": -0.047781698405742645,
+      "sft_loss": 0.6766480803489685,
+      "step": 470
+    },
+    {
+      "epoch": 1.1517696460707858,
+      "grad_norm": 2.0416324249302593,
+      "learning_rate": 3.4949824078663214e-06,
+      "logits/chosen": 0.3260158598423004,
+      "logits/rejected": 0.4627075791358948,
+      "logps/chosen": -0.6955934762954712,
+      "logps/rejected": -1.0405316352844238,
+      "loss": 0.7744,
+      "odds_ratio_loss": 0.5207543969154358,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.03477967530488968,
+      "rewards/margins": 0.017246905714273453,
+      "rewards/rejected": -0.05202658101916313,
+      "sft_loss": 0.6955934762954712,
+      "step": 480
+    },
+    {
+      "epoch": 1.175764847030594,
+      "grad_norm": 2.159701142475688,
+      "learning_rate": 3.4359893449634713e-06,
+      "logits/chosen": 0.10285909473896027,
+      "logits/rejected": 0.18586108088493347,
+      "logps/chosen": -0.7835036516189575,
+      "logps/rejected": -0.9662873148918152,
+      "loss": 0.7699,
+      "odds_ratio_loss": 0.6257883310317993,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.03917517885565758,
+      "rewards/margins": 0.009139184840023518,
+      "rewards/rejected": -0.04831436648964882,
+      "sft_loss": 0.7835036516189575,
+      "step": 490
+    },
+    {
+      "epoch": 1.1997600479904018,
+      "grad_norm": 1.905386181833648,
+      "learning_rate": 3.3763837198099807e-06,
+      "logits/chosen": 0.2618166208267212,
+      "logits/rejected": 0.403994083404541,
+      "logps/chosen": -0.7472913861274719,
+      "logps/rejected": -0.9723391532897949,
+      "loss": 0.8034,
+      "odds_ratio_loss": 0.5758217573165894,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.03736456483602524,
+      "rewards/margins": 0.011252395808696747,
+      "rewards/rejected": -0.048616960644721985,
+      "sft_loss": 0.7472913861274719,
+      "step": 500
+    },
+    {
+      "epoch": 1.22375524895021,
+      "grad_norm": 1.8483335773730425,
+      "learning_rate": 3.3162045415634793e-06,
+      "logits/chosen": -0.06936601549386978,
+      "logits/rejected": 0.15932008624076843,
+      "logps/chosen": -0.7298214435577393,
+      "logps/rejected": -0.989848792552948,
+      "loss": 0.764,
+      "odds_ratio_loss": 0.5586143136024475,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.036491066217422485,
+      "rewards/margins": 0.013001373037695885,
+      "rewards/rejected": -0.04949244111776352,
+      "sft_loss": 0.7298214435577393,
+      "step": 510
+    },
+    {
+      "epoch": 1.247750449910018,
+      "grad_norm": 1.4105189905656275,
+      "learning_rate": 3.255491194745878e-06,
+      "logits/chosen": -0.0699717178940773,
+      "logits/rejected": 0.11926586925983429,
+      "logps/chosen": -0.7712666988372803,
+      "logps/rejected": -1.0007984638214111,
+      "loss": 0.7514,
+      "odds_ratio_loss": 0.576269805431366,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -0.03856333717703819,
+      "rewards/margins": 0.011476586572825909,
+      "rewards/rejected": -0.050039924681186676,
+      "sft_loss": 0.7712666988372803,
+      "step": 520
+    },
+    {
+      "epoch": 1.2717456508698262,
+      "grad_norm": 1.5086406745902339,
+      "learning_rate": 3.1942834134680123e-06,
+      "logits/chosen": -0.4110763669013977,
+      "logits/rejected": -0.197097510099411,
+      "logps/chosen": -0.7337836027145386,
+      "logps/rejected": -1.0581499338150024,
+      "loss": 0.747,
+      "odds_ratio_loss": 0.5731949806213379,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.03668918460607529,
+      "rewards/margins": 0.016218315809965134,
+      "rewards/rejected": -0.05290750414133072,
+      "sft_loss": 0.7337836027145386,
+      "step": 530
+    },
+    {
+      "epoch": 1.295740851829634,
+      "grad_norm": 2.007767969966132,
+      "learning_rate": 3.13262125542547e-06,
+      "logits/chosen": 0.24464428424835205,
+      "logits/rejected": 0.42607539892196655,
+      "logps/chosen": -0.8008230328559875,
+      "logps/rejected": -1.019913911819458,
+      "loss": 0.7839,
+      "odds_ratio_loss": 0.5772299766540527,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.04004114866256714,
+      "rewards/margins": 0.010954543016850948,
+      "rewards/rejected": -0.05099569633603096,
+      "sft_loss": 0.8008230328559875,
+      "step": 540
+    },
+    {
+      "epoch": 1.3197360527894422,
+      "grad_norm": 2.031522996603775,
+      "learning_rate": 3.0705450756826707e-06,
+      "logits/chosen": -0.6761570572853088,
+      "logits/rejected": -0.5336428880691528,
+      "logps/chosen": -0.7791737914085388,
+      "logps/rejected": -0.9758432507514954,
+      "loss": 0.7734,
+      "odds_ratio_loss": 0.5955380201339722,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.03895869478583336,
+      "rewards/margins": 0.009833470918238163,
+      "rewards/rejected": -0.04879216477274895,
+      "sft_loss": 0.7791737914085388,
+      "step": 550
+    },
+    {
+      "epoch": 1.34373125374925,
+      "grad_norm": 1.8127230145286217,
+      "learning_rate": 3.00809550026231e-06,
+      "logits/chosen": 0.7122937440872192,
+      "logits/rejected": 0.8374090194702148,
+      "logps/chosen": -0.7448546290397644,
+      "logps/rejected": -1.0183660984039307,
+      "loss": 0.7313,
+      "odds_ratio_loss": 0.5605376362800598,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.03724273294210434,
+      "rewards/margins": 0.01367556769400835,
+      "rewards/rejected": -0.050918303430080414,
+      "sft_loss": 0.7448546290397644,
+      "step": 560
+    },
+    {
+      "epoch": 1.3677264547090582,
+      "grad_norm": 1.6102410365866324,
+      "learning_rate": 2.9453133995574955e-06,
+      "logits/chosen": 0.1695878505706787,
+      "logits/rejected": 0.34987810254096985,
+      "logps/chosen": -0.7041548490524292,
+      "logps/rejected": -1.1295292377471924,
+      "loss": 0.7529,
+      "odds_ratio_loss": 0.5541011095046997,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.03520774096250534,
+      "rewards/margins": 0.02126871421933174,
+      "rewards/rejected": -0.05647646263241768,
+      "sft_loss": 0.7041548490524292,
+      "step": 570
+    },
+    {
+      "epoch": 1.3917216556688663,
+      "grad_norm": 2.0516481147792964,
+      "learning_rate": 2.8822398615839337e-06,
+      "logits/chosen": -0.15236589312553406,
+      "logits/rejected": 0.005555987358093262,
+      "logps/chosen": -0.7019264698028564,
+      "logps/rejected": -0.9463084936141968,
+      "loss": 0.7377,
+      "odds_ratio_loss": 0.5546727180480957,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.03509632498025894,
+      "rewards/margins": 0.012219103053212166,
+      "rewards/rejected": -0.04731542617082596,
+      "sft_loss": 0.7019264698028564,
+      "step": 580
+    },
+    {
+      "epoch": 1.4157168566286742,
+      "grad_norm": 2.5703275268486463,
+      "learning_rate": 2.8189161650897045e-06,
+      "logits/chosen": 0.09915417432785034,
+      "logits/rejected": 0.2876579761505127,
+      "logps/chosen": -0.7416352033615112,
+      "logps/rejected": -0.9542354345321655,
+      "loss": 0.7748,
+      "odds_ratio_loss": 0.5765627026557922,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.0370817631483078,
+      "rewards/margins": 0.010630009695887566,
+      "rewards/rejected": -0.04771176725625992,
+      "sft_loss": 0.7416352033615112,
+      "step": 590
+    },
+    {
+      "epoch": 1.4397120575884823,
+      "grad_norm": 1.6574957139548097,
+      "learning_rate": 2.7553837525402095e-06,
+      "logits/chosen": 0.14950448274612427,
+      "logits/rejected": 0.14670611917972565,
+      "logps/chosen": -0.7459922432899475,
+      "logps/rejected": -0.9438718557357788,
+      "loss": 0.764,
+      "odds_ratio_loss": 0.6029990911483765,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.037299610674381256,
+      "rewards/margins": 0.009893985465168953,
+      "rewards/rejected": -0.04719359427690506,
+      "sft_loss": 0.7459922432899475,
+      "step": 600
+    },
+    {
+      "epoch": 1.4637072585482904,
+      "grad_norm": 1.5955732799355493,
+      "learning_rate": 2.691684202995966e-06,
+      "logits/chosen": 0.43530672788619995,
+      "logits/rejected": 0.4994083344936371,
+      "logps/chosen": -0.8142836689949036,
+      "logps/rejected": -0.9706009030342102,
+      "loss": 0.7559,
+      "odds_ratio_loss": 0.7006958723068237,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.04071418568491936,
+      "rewards/margins": 0.007815859280526638,
+      "rewards/rejected": -0.04853004962205887,
+      "sft_loss": 0.8142836689949036,
+      "step": 610
+    },
+    {
+      "epoch": 1.4877024595080983,
+      "grad_norm": 1.9589861397245603,
+      "learning_rate": 2.6278592049010204e-06,
+      "logits/chosen": -0.19675548374652863,
+      "logits/rejected": -0.004504656884819269,
+      "logps/chosen": -0.7537368535995483,
+      "logps/rejected": -1.0135046243667603,
+      "loss": 0.7741,
+      "odds_ratio_loss": 0.5691729187965393,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.03768684342503548,
+      "rewards/margins": 0.012988388538360596,
+      "rewards/rejected": -0.050675224512815475,
+      "sft_loss": 0.7537368535995483,
+      "step": 620
+    },
+    {
+      "epoch": 1.5116976604679064,
+      "grad_norm": 1.7255875955000524,
+      "learning_rate": 2.5639505287997584e-06,
+      "logits/chosen": 0.3145737051963806,
+      "logits/rejected": 0.47394928336143494,
+      "logps/chosen": -0.7314926385879517,
+      "logps/rejected": -1.001952886581421,
+      "loss": 0.7829,
+      "odds_ratio_loss": 0.5629433393478394,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.03657463565468788,
+      "rewards/margins": 0.013523015193641186,
+      "rewards/rejected": -0.050097644329071045,
+      "sft_loss": 0.7314926385879517,
+      "step": 630
+    },
+    {
+      "epoch": 1.5356928614277146,
+      "grad_norm": 2.504847023988975,
+      "learning_rate": 2.5e-06,
+      "logits/chosen": 0.2320265769958496,
+      "logits/rejected": 0.3284027874469757,
+      "logps/chosen": -0.7656562924385071,
+      "logps/rejected": -1.076923131942749,
+      "loss": 0.7503,
+      "odds_ratio_loss": 0.584337592124939,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.038282815366983414,
+      "rewards/margins": 0.015563338994979858,
+      "rewards/rejected": -0.053846150636672974,
+      "sft_loss": 0.7656562924385071,
+      "step": 640
+    },
+    {
+      "epoch": 1.5596880623875224,
+      "grad_norm": 1.4394266237384084,
+      "learning_rate": 2.436049471200242e-06,
+      "logits/chosen": -0.5206400156021118,
+      "logits/rejected": -0.38631540536880493,
+      "logps/chosen": -0.8094362020492554,
+      "logps/rejected": -0.9923938512802124,
+      "loss": 0.7752,
+      "odds_ratio_loss": 0.5967071056365967,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.04047181457281113,
+      "rewards/margins": 0.00914788618683815,
+      "rewards/rejected": -0.04961969703435898,
+      "sft_loss": 0.8094362020492554,
+      "step": 650
+    },
+    {
+      "epoch": 1.5836832633473306,
+      "grad_norm": 1.7625452374002906,
+      "learning_rate": 2.3721407950989804e-06,
+      "logits/chosen": -0.24351301789283752,
+      "logits/rejected": -0.07003232091665268,
+      "logps/chosen": -0.6876959800720215,
+      "logps/rejected": -0.9035342335700989,
+      "loss": 0.7734,
+      "odds_ratio_loss": 0.5917103290557861,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.034384798258543015,
+      "rewards/margins": 0.010791914537549019,
+      "rewards/rejected": -0.045176707208156586,
+      "sft_loss": 0.6876959800720215,
+      "step": 660
+    },
+    {
+      "epoch": 1.6076784643071385,
+      "grad_norm": 1.6046093499190943,
+      "learning_rate": 2.3083157970040344e-06,
+      "logits/chosen": 0.5633162260055542,
+      "logits/rejected": 0.6462755799293518,
+      "logps/chosen": -0.7524802684783936,
+      "logps/rejected": -1.0558850765228271,
+      "loss": 0.7563,
+      "odds_ratio_loss": 0.552274227142334,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.03762401267886162,
+      "rewards/margins": 0.015170246362686157,
+      "rewards/rejected": -0.05279426649212837,
+      "sft_loss": 0.7524802684783936,
+      "step": 670
+    },
+    {
+      "epoch": 1.6316736652669466,
+      "grad_norm": 2.117352018263469,
+      "learning_rate": 2.2446162474597913e-06,
+      "logits/chosen": 0.43944865465164185,
+      "logits/rejected": 0.5002392530441284,
+      "logps/chosen": -0.7501770257949829,
+      "logps/rejected": -0.9691005945205688,
+      "loss": 0.7699,
+      "odds_ratio_loss": 0.5791727304458618,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.037508852779865265,
+      "rewards/margins": 0.010946177877485752,
+      "rewards/rejected": -0.04845503345131874,
+      "sft_loss": 0.7501770257949829,
+      "step": 680
+    },
+    {
+      "epoch": 1.6556688662267547,
+      "grad_norm": 1.6685249776962552,
+      "learning_rate": 2.1810838349102963e-06,
+      "logits/chosen": 0.16153453290462494,
+      "logits/rejected": 0.20878514647483826,
+      "logps/chosen": -0.7516240477561951,
+      "logps/rejected": -1.0250643491744995,
+      "loss": 0.7666,
+      "odds_ratio_loss": 0.5872852206230164,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.03758120536804199,
+      "rewards/margins": 0.013672016561031342,
+      "rewards/rejected": -0.051253218203783035,
+      "sft_loss": 0.7516240477561951,
+      "step": 690
+    },
+    {
+      "epoch": 1.6796640671865628,
+      "grad_norm": 2.782782057649718,
+      "learning_rate": 2.117760138416067e-06,
+      "logits/chosen": 0.24376201629638672,
+      "logits/rejected": 0.44258540868759155,
+      "logps/chosen": -0.6985687017440796,
+      "logps/rejected": -1.0050299167633057,
+      "loss": 0.7614,
+      "odds_ratio_loss": 0.543103814125061,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -0.03492843732237816,
+      "rewards/margins": 0.015323063358664513,
+      "rewards/rejected": -0.05025150254368782,
+      "sft_loss": 0.6985687017440796,
+      "step": 700
+    },
+    {
+      "epoch": 1.7036592681463707,
+      "grad_norm": 1.5369658154698735,
+      "learning_rate": 2.0546866004425053e-06,
+      "logits/chosen": 0.3964254558086395,
+      "logits/rejected": 0.4900701642036438,
+      "logps/chosen": -0.7590494155883789,
+      "logps/rejected": -1.2440413236618042,
+      "loss": 0.7652,
+      "odds_ratio_loss": 0.5372438430786133,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.037952471524477005,
+      "rewards/margins": 0.024249596521258354,
+      "rewards/rejected": -0.06220207363367081,
+      "sft_loss": 0.7590494155883789,
+      "step": 710
+    },
+    {
+      "epoch": 1.7276544691061788,
+      "grad_norm": 1.9970193945029362,
+      "learning_rate": 1.9919044997376906e-06,
+      "logits/chosen": 0.6031176447868347,
+      "logits/rejected": 0.7783833742141724,
+      "logps/chosen": -0.7290822267532349,
+      "logps/rejected": -1.021554946899414,
+      "loss": 0.7176,
+      "odds_ratio_loss": 0.557815432548523,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.03645411133766174,
+      "rewards/margins": 0.014623639173805714,
+      "rewards/rejected": -0.051077745854854584,
+      "sft_loss": 0.7290822267532349,
+      "step": 720
+    },
+    {
+      "epoch": 1.7516496700659867,
+      "grad_norm": 2.558147455560064,
+      "learning_rate": 1.9294549243173306e-06,
+      "logits/chosen": -0.027294237166643143,
+      "logits/rejected": 0.11035363376140594,
+      "logps/chosen": -0.7765438556671143,
+      "logps/rejected": -1.0300321578979492,
+      "loss": 0.7771,
+      "odds_ratio_loss": 0.5954040884971619,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.03882719203829765,
+      "rewards/margins": 0.012674416415393353,
+      "rewards/rejected": -0.05150160938501358,
+      "sft_loss": 0.7765438556671143,
+      "step": 730
+    },
+    {
+      "epoch": 1.7756448710257948,
+      "grad_norm": 2.346615273317464,
+      "learning_rate": 1.8673787445745298e-06,
+      "logits/chosen": -0.449845552444458,
+      "logits/rejected": -0.3746832311153412,
+      "logps/chosen": -0.7114017605781555,
+      "logps/rejected": -0.928491473197937,
+      "loss": 0.7699,
+      "odds_ratio_loss": 0.5795110464096069,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.035570088773965836,
+      "rewards/margins": 0.010854486376047134,
+      "rewards/rejected": -0.04642457515001297,
+      "sft_loss": 0.7114017605781555,
+      "step": 740
+    },
+    {
+      "epoch": 1.799640071985603,
+      "grad_norm": 1.995371230537378,
+      "learning_rate": 1.805716586531988e-06,
+      "logits/chosen": -0.13443303108215332,
+      "logits/rejected": 0.014731263741850853,
+      "logps/chosen": -0.8079891204833984,
+      "logps/rejected": -1.0810317993164062,
+      "loss": 0.7825,
+      "odds_ratio_loss": 0.6112096309661865,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.0403994545340538,
+      "rewards/margins": 0.013652140274643898,
+      "rewards/rejected": -0.05405158922076225,
+      "sft_loss": 0.8079891204833984,
+      "step": 750
+    },
+    {
+      "epoch": 1.823635272945411,
+      "grad_norm": 1.8742057389590454,
+      "learning_rate": 1.7445088052541218e-06,
+      "logits/chosen": 0.046121031045913696,
+      "logits/rejected": 0.1955467015504837,
+      "logps/chosen": -0.7093559503555298,
+      "logps/rejected": -1.0484099388122559,
+      "loss": 0.7617,
+      "odds_ratio_loss": 0.5657014846801758,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.03546779602766037,
+      "rewards/margins": 0.016952697187662125,
+      "rewards/rejected": -0.05242049694061279,
+      "sft_loss": 0.7093559503555298,
+      "step": 760
+    },
+    {
+      "epoch": 1.847630473905219,
+      "grad_norm": 1.2680203881504901,
+      "learning_rate": 1.6837954584365217e-06,
+      "logits/chosen": 0.4459083080291748,
+      "logits/rejected": 0.5636454224586487,
+      "logps/chosen": -0.7526987195014954,
+      "logps/rejected": -1.009804606437683,
+      "loss": 0.7871,
+      "odds_ratio_loss": 0.5556772947311401,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -0.03763493150472641,
+      "rewards/margins": 0.012855296023190022,
+      "rewards/rejected": -0.050490230321884155,
+      "sft_loss": 0.7526987195014954,
+      "step": 770
+    },
+    {
+      "epoch": 1.8716256748650268,
+      "grad_norm": 1.9254646582677224,
+      "learning_rate": 1.6236162801900191e-06,
+      "logits/chosen": -0.10451897233724594,
+      "logits/rejected": 0.3060254156589508,
+      "logps/chosen": -0.6585639715194702,
+      "logps/rejected": -0.9869001507759094,
+      "loss": 0.71,
+      "odds_ratio_loss": 0.4942260682582855,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -0.03292820230126381,
+      "rewards/margins": 0.016416804865002632,
+      "rewards/rejected": -0.04934500530362129,
+      "sft_loss": 0.6585639715194702,
+      "step": 780
+    },
+    {
+      "epoch": 1.895620875824835,
+      "grad_norm": 1.9904836511656812,
+      "learning_rate": 1.5640106550365298e-06,
+      "logits/chosen": 0.11656351387500763,
+      "logits/rejected": 0.29824742674827576,
+      "logps/chosen": -0.7831540703773499,
+      "logps/rejected": -1.0284688472747803,
+      "loss": 0.7758,
+      "odds_ratio_loss": 0.5839165449142456,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.03915770351886749,
+      "rewards/margins": 0.01226573996245861,
+      "rewards/rejected": -0.051423441618680954,
+      "sft_loss": 0.7831540703773499,
+      "step": 790
+    },
+    {
+      "epoch": 1.919616076784643,
+      "grad_norm": 1.7061927534288226,
+      "learning_rate": 1.5050175921336797e-06,
+      "logits/chosen": 0.14354857802391052,
+      "logits/rejected": 0.27334246039390564,
+      "logps/chosen": -0.7474446892738342,
+      "logps/rejected": -0.9480558633804321,
+      "loss": 0.7575,
+      "odds_ratio_loss": 0.6441240310668945,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.03737223893404007,
+      "rewards/margins": 0.010030550882220268,
+      "rewards/rejected": -0.04740279167890549,
+      "sft_loss": 0.7474446892738342,
+      "step": 800
+    },
+    {
+      "epoch": 1.9436112777444512,
+      "grad_norm": 2.251879648695612,
+      "learning_rate": 1.446675699745097e-06,
+      "logits/chosen": 0.25183239579200745,
+      "logits/rejected": 0.38326969742774963,
+      "logps/chosen": -0.7823570966720581,
+      "logps/rejected": -0.9946805238723755,
+      "loss": 0.8037,
+      "odds_ratio_loss": 0.6080455183982849,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.03911786153912544,
+      "rewards/margins": 0.010616169311106205,
+      "rewards/rejected": -0.049734026193618774,
+      "sft_loss": 0.7823570966720581,
+      "step": 810
+    },
+    {
+      "epoch": 1.9676064787042593,
+      "grad_norm": 1.9391362449031262,
+      "learning_rate": 1.3890231599730674e-06,
+      "logits/chosen": 0.31725913286209106,
+      "logits/rejected": 0.5106421709060669,
+      "logps/chosen": -0.7221857309341431,
+      "logps/rejected": -0.9829575419425964,
+      "loss": 0.7904,
+      "odds_ratio_loss": 0.5538625121116638,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -0.03610928729176521,
+      "rewards/margins": 0.013038587756454945,
+      "rewards/rejected": -0.049147870391607285,
+      "sft_loss": 0.7221857309341431,
+      "step": 820
+    },
+    {
+      "epoch": 1.9916016796640672,
+      "grad_norm": 1.5457295502049215,
+      "learning_rate": 1.3320977037700952e-06,
+      "logits/chosen": 0.8291665315628052,
+      "logits/rejected": 1.1122350692749023,
+      "logps/chosen": -0.6864774227142334,
+      "logps/rejected": -1.0247427225112915,
+      "loss": 0.7452,
+      "odds_ratio_loss": 0.49447354674339294,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.03432386741042137,
+      "rewards/margins": 0.016913266852498055,
+      "rewards/rejected": -0.051237136125564575,
+      "sft_loss": 0.6864774227142334,
+      "step": 830
+    },
+    {
+      "epoch": 2.015596880623875,
+      "grad_norm": 1.5016852289986733,
+      "learning_rate": 1.2759365862457148e-06,
+      "logits/chosen": -0.4956502318382263,
+      "logits/rejected": -0.1621031016111374,
+      "logps/chosen": -0.7308815717697144,
+      "logps/rejected": -0.9828909039497375,
+      "loss": 0.7173,
+      "odds_ratio_loss": 0.5487710237503052,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.0365440808236599,
+      "rewards/margins": 0.012600463815033436,
+      "rewards/rejected": -0.049144547432661057,
+      "sft_loss": 0.7308815717697144,
+      "step": 840
+    },
+    {
+      "epoch": 2.039592081583683,
+      "grad_norm": 1.622924065562837,
+      "learning_rate": 1.2205765622847273e-06,
+      "logits/chosen": -0.12397761642932892,
+      "logits/rejected": 0.08023932576179504,
+      "logps/chosen": -0.6277745962142944,
+      "logps/rejected": -1.0955206155776978,
+      "loss": 0.6995,
+      "odds_ratio_loss": 0.4475070536136627,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -0.03138873726129532,
+      "rewards/margins": 0.023387301713228226,
+      "rewards/rejected": -0.054776035249233246,
+      "sft_loss": 0.6277745962142944,
+      "step": 850
+    },
+    {
+      "epoch": 2.0635872825434913,
+      "grad_norm": 1.4741935497367946,
+      "learning_rate": 1.1660538624928062e-06,
+      "logits/chosen": -0.3639386296272278,
+      "logits/rejected": -0.2011258602142334,
+      "logps/chosen": -0.6642920970916748,
+      "logps/rejected": -1.0270217657089233,
+      "loss": 0.7019,
+      "odds_ratio_loss": 0.4971997141838074,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.03321460261940956,
+      "rewards/margins": 0.018136484548449516,
+      "rewards/rejected": -0.05135108903050423,
+      "sft_loss": 0.6642920970916748,
+      "step": 860
+    },
+    {
+      "epoch": 2.0875824835032994,
+      "grad_norm": 1.7172174730539993,
+      "learning_rate": 1.112404169485226e-06,
+      "logits/chosen": -0.3923923075199127,
+      "logits/rejected": -0.10327514261007309,
+      "logps/chosen": -0.5645719766616821,
+      "logps/rejected": -1.071115255355835,
+      "loss": 0.6681,
+      "odds_ratio_loss": 0.42052555084228516,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.028228599578142166,
+      "rewards/margins": 0.025327179580926895,
+      "rewards/rejected": -0.053555767983198166,
+      "sft_loss": 0.5645719766616821,
+      "step": 870
+    },
+    {
+      "epoch": 2.1115776844631076,
+      "grad_norm": 1.1474314844125568,
+      "learning_rate": 1.0596625945342148e-06,
+      "logits/chosen": -0.008033117279410362,
+      "logits/rejected": 0.16419892013072968,
+      "logps/chosen": -0.7100299000740051,
+      "logps/rejected": -0.9733055233955383,
+      "loss": 0.6813,
+      "odds_ratio_loss": 0.5328400731086731,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -0.03550150245428085,
+      "rewards/margins": 0.013163777068257332,
+      "rewards/rejected": -0.048665277659893036,
+      "sft_loss": 0.7100299000740051,
+      "step": 880
+    },
+    {
+      "epoch": 2.1355728854229152,
+      "grad_norm": 2.1383619388719515,
+      "learning_rate": 1.0078636545902363e-06,
+      "logits/chosen": -0.4247666001319885,
+      "logits/rejected": -0.17631380259990692,
+      "logps/chosen": -0.6582883596420288,
+      "logps/rejected": -1.0547147989273071,
+      "loss": 0.6895,
+      "odds_ratio_loss": 0.47398701310157776,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -0.0329144187271595,
+      "rewards/margins": 0.019821325317025185,
+      "rewards/rejected": -0.05273573845624924,
+      "sft_loss": 0.6582883596420288,
+      "step": 890
+    },
+    {
+      "epoch": 2.1595680863827234,
+      "grad_norm": 1.5320300236939732,
+      "learning_rate": 9.570412496922198e-07,
+      "logits/chosen": -0.27953624725341797,
+      "logits/rejected": -0.08715387433767319,
+      "logps/chosen": -0.5965186357498169,
+      "logps/rejected": -1.154284119606018,
+      "loss": 0.6738,
+      "odds_ratio_loss": 0.4240815043449402,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -0.029825935140252113,
+      "rewards/margins": 0.02788827195763588,
+      "rewards/rejected": -0.05771421268582344,
+      "sft_loss": 0.5965186357498169,
+      "step": 900
+    },
+    {
+      "epoch": 2.1835632873425315,
+      "grad_norm": 1.6204787225170885,
+      "learning_rate": 9.07228640781539e-07,
+      "logits/chosen": 0.368365079164505,
+      "logits/rejected": 0.6101259589195251,
+      "logps/chosen": -0.6893322467803955,
+      "logps/rejected": -1.0903311967849731,
+      "loss": 0.6791,
+      "odds_ratio_loss": 0.4818887710571289,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -0.03446660935878754,
+      "rewards/margins": 0.02004995197057724,
+      "rewards/rejected": -0.054516565054655075,
+      "sft_loss": 0.6893322467803955,
+      "step": 910
+    },
+    {
+      "epoch": 2.2075584883023396,
+      "grad_norm": 1.290844558254926,
+      "learning_rate": 8.584584279342392e-07,
+      "logits/chosen": -0.16083380579948425,
+      "logits/rejected": -0.10739579051733017,
+      "logps/chosen": -0.6938862800598145,
+      "logps/rejected": -0.9513536691665649,
+      "loss": 0.6888,
+      "odds_ratio_loss": 0.5428452491760254,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.034694310277700424,
+      "rewards/margins": 0.012873371131718159,
+      "rewards/rejected": -0.047567687928676605,
+      "sft_loss": 0.6938862800598145,
+      "step": 920
+    },
+    {
+      "epoch": 2.2315536892621477,
+      "grad_norm": 1.5229766148545818,
+      "learning_rate": 8.10762529025782e-07,
+      "logits/chosen": -0.4659739136695862,
+      "logits/rejected": -0.4786594808101654,
+      "logps/chosen": -0.6584521532058716,
+      "logps/rejected": -0.8917843699455261,
+      "loss": 0.65,
+      "odds_ratio_loss": 0.5486137866973877,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.03292260691523552,
+      "rewards/margins": 0.011666612699627876,
+      "rewards/rejected": -0.044589221477508545,
+      "sft_loss": 0.6584521532058716,
+      "step": 930
+    },
+    {
+      "epoch": 2.255548890221956,
+      "grad_norm": 1.7015940933867517,
+      "learning_rate": 7.641721588422526e-07,
+      "logits/chosen": -0.009342163801193237,
+      "logits/rejected": 0.1280032843351364,
+      "logps/chosen": -0.6387184262275696,
+      "logps/rejected": -1.049140453338623,
+      "loss": 0.687,
+      "odds_ratio_loss": 0.4773840010166168,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.0319359228014946,
+      "rewards/margins": 0.020521100610494614,
+      "rewards/rejected": -0.05245702341198921,
+      "sft_loss": 0.6387184262275696,
+      "step": 940
+    },
+    {
+      "epoch": 2.2795440911817635,
+      "grad_norm": 1.4203319350991257,
+      "learning_rate": 7.187178086517116e-07,
+      "logits/chosen": 0.14468683302402496,
+      "logits/rejected": 0.2608656883239746,
+      "logps/chosen": -0.6514204144477844,
+      "logps/rejected": -1.2591578960418701,
+      "loss": 0.6695,
+      "odds_ratio_loss": 0.455849826335907,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -0.03257102146744728,
+      "rewards/margins": 0.03038688376545906,
+      "rewards/rejected": -0.06295789778232574,
+      "sft_loss": 0.6514204144477844,
+      "step": 950
+    },
+    {
+      "epoch": 2.3035392921415716,
+      "grad_norm": 1.7783791010197938,
+      "learning_rate": 6.74429226249049e-07,
+      "logits/chosen": 0.09898465871810913,
+      "logits/rejected": 0.21373791992664337,
+      "logps/chosen": -0.6381307244300842,
+      "logps/rejected": -0.9742431640625,
+      "loss": 0.6712,
+      "odds_ratio_loss": 0.49530988931655884,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.03190653771162033,
+      "rewards/margins": 0.016805628314614296,
+      "rewards/rejected": -0.04871216416358948,
+      "sft_loss": 0.6381307244300842,
+      "step": 960
+    },
+    {
+      "epoch": 2.3275344931013797,
+      "grad_norm": 1.6090454208525553,
+      "learning_rate": 6.313353964874155e-07,
+      "logits/chosen": 0.1333683431148529,
+      "logits/rejected": 0.3417516350746155,
+      "logps/chosen": -0.6887052655220032,
+      "logps/rejected": -1.0016798973083496,
+      "loss": 0.6673,
+      "odds_ratio_loss": 0.5059822797775269,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -0.03443526476621628,
+      "rewards/margins": 0.01564873196184635,
+      "rewards/rejected": -0.05008399486541748,
+      "sft_loss": 0.6887052655220032,
+      "step": 970
+    },
+    {
+      "epoch": 2.351529694061188,
+      "grad_norm": 1.6382111002720514,
+      "learning_rate": 5.894645223089584e-07,
+      "logits/chosen": 0.7236309051513672,
+      "logits/rejected": 0.8550646901130676,
+      "logps/chosen": -0.6779772639274597,
+      "logps/rejected": -1.2183148860931396,
+      "loss": 0.6958,
+      "odds_ratio_loss": 0.448292076587677,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -0.033898863941431046,
+      "rewards/margins": 0.027016881853342056,
+      "rewards/rejected": -0.0609157457947731,
+      "sft_loss": 0.6779772639274597,
+      "step": 980
+    },
+    {
+      "epoch": 2.375524895020996,
+      "grad_norm": 1.680992010239421,
+      "learning_rate": 5.48844006287289e-07,
+      "logits/chosen": 0.12925365567207336,
+      "logits/rejected": 0.3167954981327057,
+      "logps/chosen": -0.6692675352096558,
+      "logps/rejected": -1.0140740871429443,
+      "loss": 0.6691,
+      "odds_ratio_loss": 0.4763975143432617,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.033463381230831146,
+      "rewards/margins": 0.01724032498896122,
+      "rewards/rejected": -0.050703711807727814,
+      "sft_loss": 0.6692675352096558,
+      "step": 990
+    },
+    {
+      "epoch": 2.3995200959808036,
+      "grad_norm": 1.544720546176764,
+      "learning_rate": 5.095004326937445e-07,
+      "logits/chosen": -0.4231066107749939,
+      "logits/rejected": -0.20230142772197723,
+      "logps/chosen": -0.6737790107727051,
+      "logps/rejected": -1.0810075998306274,
+      "loss": 0.6744,
+      "odds_ratio_loss": 0.4769432544708252,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.033688947558403015,
+      "rewards/margins": 0.02036142908036709,
+      "rewards/rejected": -0.05405038595199585,
+      "sft_loss": 0.6737790107727051,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4235152969406117,
+      "grad_norm": 1.7400382431256138,
+      "learning_rate": 4.71459550099202e-07,
+      "logits/chosen": 0.2943962812423706,
+      "logits/rejected": 0.5343393087387085,
+      "logps/chosen": -0.6686779856681824,
+      "logps/rejected": -1.0820672512054443,
+      "loss": 0.7078,
+      "odds_ratio_loss": 0.5010559558868408,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.03343390300869942,
+      "rewards/margins": 0.020669464021921158,
+      "rewards/rejected": -0.054103363305330276,
+      "sft_loss": 0.6686779856681824,
+      "step": 1010
+    },
+    {
+      "epoch": 2.44751049790042,
+      "grad_norm": 1.548219424075948,
+      "learning_rate": 4.347462545228134e-07,
+      "logits/chosen": 0.13567771017551422,
+      "logits/rejected": 0.31968480348587036,
+      "logps/chosen": -0.6244124174118042,
+      "logps/rejected": -1.05476975440979,
+      "loss": 0.6563,
+      "odds_ratio_loss": 0.4984089732170105,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.03122062422335148,
+      "rewards/margins": 0.021517863497138023,
+      "rewards/rejected": -0.052738480269908905,
+      "sft_loss": 0.6244124174118042,
+      "step": 1020
+    },
+    {
+      "epoch": 2.471505698860228,
+      "grad_norm": 1.4610216249122747,
+      "learning_rate": 3.9938457313869914e-07,
+      "logits/chosen": -0.08544759452342987,
+      "logits/rejected": 0.07162941992282867,
+      "logps/chosen": -0.7579829096794128,
+      "logps/rejected": -1.1255767345428467,
+      "loss": 0.6864,
+      "odds_ratio_loss": 0.547897458076477,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -0.03789914771914482,
+      "rewards/margins": 0.01837969198822975,
+      "rewards/rejected": -0.05627884343266487,
+      "sft_loss": 0.7579829096794128,
+      "step": 1030
+    },
+    {
+      "epoch": 2.495500899820036,
+      "grad_norm": 1.6006797776983446,
+      "learning_rate": 3.6539764855126224e-07,
+      "logits/chosen": -0.23340921103954315,
+      "logits/rejected": -0.1814245879650116,
+      "logps/chosen": -0.6439553499221802,
+      "logps/rejected": -1.0276587009429932,
+      "loss": 0.6617,
+      "odds_ratio_loss": 0.5049816370010376,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.03219776228070259,
+      "rewards/margins": 0.019185172393918037,
+      "rewards/rejected": -0.05138293653726578,
+      "sft_loss": 0.6439553499221802,
+      "step": 1040
+    },
+    {
+      "epoch": 2.519496100779844,
+      "grad_norm": 2.318524117790848,
+      "learning_rate": 3.328077236494087e-07,
+      "logits/chosen": -0.12850667536258698,
+      "logits/rejected": 0.07032374292612076,
+      "logps/chosen": -0.5922039747238159,
+      "logps/rejected": -1.0730435848236084,
+      "loss": 0.6694,
+      "odds_ratio_loss": 0.43941235542297363,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -0.029610196128487587,
+      "rewards/margins": 0.024041980504989624,
+      "rewards/rejected": -0.05365217477083206,
+      "sft_loss": 0.5922039747238159,
+      "step": 1050
+    },
+    {
+      "epoch": 2.5434913017396523,
+      "grad_norm": 1.8087989245838814,
+      "learning_rate": 3.0163612704959486e-07,
+      "logits/chosen": -0.6611061692237854,
+      "logits/rejected": -0.5293869376182556,
+      "logps/chosen": -0.6281863451004028,
+      "logps/rejected": -0.9944284558296204,
+      "loss": 0.6705,
+      "odds_ratio_loss": 0.47698038816452026,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.03140931576490402,
+      "rewards/margins": 0.018312102183699608,
+      "rewards/rejected": -0.04972142353653908,
+      "sft_loss": 0.6281863451004028,
+      "step": 1060
+    },
+    {
+      "epoch": 2.56748650269946,
+      "grad_norm": 1.5444353690364836,
+      "learning_rate": 2.71903259137222e-07,
+      "logits/chosen": 0.411745548248291,
+      "logits/rejected": 0.4236873686313629,
+      "logps/chosen": -0.611006498336792,
+      "logps/rejected": -1.0047032833099365,
+      "loss": 0.672,
+      "odds_ratio_loss": 0.48614612221717834,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -0.03055032715201378,
+      "rewards/margins": 0.019684839993715286,
+      "rewards/rejected": -0.050235163420438766,
+      "sft_loss": 0.611006498336792,
+      "step": 1070
+    },
+    {
+      "epoch": 2.591481703659268,
+      "grad_norm": 2.593043127599419,
+      "learning_rate": 2.436285787155185e-07,
+      "logits/chosen": 0.316955029964447,
+      "logits/rejected": 0.47285112738609314,
+      "logps/chosen": -0.6786519885063171,
+      "logps/rejected": -1.2019875049591064,
+      "loss": 0.6881,
+      "odds_ratio_loss": 0.4908427298069,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.03393259644508362,
+      "rewards/margins": 0.026166772469878197,
+      "rewards/rejected": -0.060099370777606964,
+      "sft_loss": 0.6786519885063171,
+      "step": 1080
+    },
+    {
+      "epoch": 2.6154769046190762,
+      "grad_norm": 2.2050381193088207,
+      "learning_rate": 2.168305902706383e-07,
+      "logits/chosen": -0.4541945457458496,
+      "logits/rejected": -0.18702273070812225,
+      "logps/chosen": -0.7026795148849487,
+      "logps/rejected": -0.962356448173523,
+      "loss": 0.6583,
+      "odds_ratio_loss": 0.5365189909934998,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.035133976489305496,
+      "rewards/margins": 0.012983846478164196,
+      "rewards/rejected": -0.04811782017350197,
+      "sft_loss": 0.7026795148849487,
+      "step": 1090
+    },
+    {
+      "epoch": 2.6394721055788843,
+      "grad_norm": 1.6921175899136245,
+      "learning_rate": 1.9152683186132476e-07,
+      "logits/chosen": -0.4067768156528473,
+      "logits/rejected": -0.3039708137512207,
+      "logps/chosen": -0.6328436136245728,
+      "logps/rejected": -1.12655770778656,
+      "loss": 0.6919,
+      "odds_ratio_loss": 0.4709090292453766,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -0.031642183661460876,
+      "rewards/margins": 0.024685706943273544,
+      "rewards/rejected": -0.05632789060473442,
+      "sft_loss": 0.6328436136245728,
+      "step": 1100
+    },
+    {
+      "epoch": 2.663467306538692,
+      "grad_norm": 1.5594348597838832,
+      "learning_rate": 1.6773386364104972e-07,
+      "logits/chosen": -0.1575368195772171,
+      "logits/rejected": -0.003553843591362238,
+      "logps/chosen": -0.6768941879272461,
+      "logps/rejected": -1.032041072845459,
+      "loss": 0.6913,
+      "odds_ratio_loss": 0.50171959400177,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.033844709396362305,
+      "rewards/margins": 0.017757344990968704,
+      "rewards/rejected": -0.05160205811262131,
+      "sft_loss": 0.6768941879272461,
+      "step": 1110
+    },
+    {
+      "epoch": 2.6874625074985,
+      "grad_norm": 1.2735811398241894,
+      "learning_rate": 1.4546725702015096e-07,
+      "logits/chosen": 0.004650235176086426,
+      "logits/rejected": 0.1661575585603714,
+      "logps/chosen": -0.6541981101036072,
+      "logps/rejected": -1.1094247102737427,
+      "loss": 0.6669,
+      "odds_ratio_loss": 0.4492813050746918,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -0.03270990774035454,
+      "rewards/margins": 0.022761326283216476,
+      "rewards/rejected": -0.055471230298280716,
+      "sft_loss": 0.6541981101036072,
+      "step": 1120
+    },
+    {
+      "epoch": 2.7114577084583082,
+      "grad_norm": 2.2135398834819715,
+      "learning_rate": 1.24741584475056e-07,
+      "logits/chosen": -0.07907108962535858,
+      "logits/rejected": 0.08474680036306381,
+      "logps/chosen": -0.6154497861862183,
+      "logps/rejected": -1.0710924863815308,
+      "loss": 0.6491,
+      "odds_ratio_loss": 0.4509805142879486,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -0.030772492289543152,
+      "rewards/margins": 0.022782133892178535,
+      "rewards/rejected": -0.05355461686849594,
+      "sft_loss": 0.6154497861862183,
+      "step": 1130
+    },
+    {
+      "epoch": 2.7354529094181164,
+      "grad_norm": 1.5137426741255027,
+      "learning_rate": 1.0557041001126145e-07,
+      "logits/chosen": 0.3702402710914612,
+      "logits/rejected": 0.6300150156021118,
+      "logps/chosen": -0.5984182357788086,
+      "logps/rejected": -1.115179419517517,
+      "loss": 0.6191,
+      "odds_ratio_loss": 0.41762223839759827,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -0.0299209114164114,
+      "rewards/margins": 0.025838062167167664,
+      "rewards/rejected": -0.05575897544622421,
+      "sft_loss": 0.5984182357788086,
+      "step": 1140
+    },
+    {
+      "epoch": 2.7594481103779245,
+      "grad_norm": 1.565522436867544,
+      "learning_rate": 8.796628028631321e-08,
+      "logits/chosen": 0.17880654335021973,
+      "logits/rejected": 0.1116660013794899,
+      "logps/chosen": -0.6091745495796204,
+      "logps/rejected": -1.0210378170013428,
+      "loss": 0.6583,
+      "odds_ratio_loss": 0.4544963836669922,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.030458729714155197,
+      "rewards/margins": 0.02059316076338291,
+      "rewards/rejected": -0.05105189234018326,
+      "sft_loss": 0.6091745495796204,
+      "step": 1150
+    },
+    {
+      "epoch": 2.7834433113377326,
+      "grad_norm": 1.604017358081912,
+      "learning_rate": 7.19407163985894e-08,
+      "logits/chosen": -0.04378344863653183,
+      "logits/rejected": 0.18321049213409424,
+      "logps/chosen": -0.6626521348953247,
+      "logps/rejected": -1.1215763092041016,
+      "loss": 0.666,
+      "odds_ratio_loss": 0.4741577208042145,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.033132605254650116,
+      "rewards/margins": 0.022946210578083992,
+      "rewards/rejected": -0.05607881397008896,
+      "sft_loss": 0.6626521348953247,
+      "step": 1160
+    },
+    {
+      "epoch": 2.8074385122975407,
+      "grad_norm": 1.4084206676302562,
+      "learning_rate": 5.750420634727083e-08,
+      "logits/chosen": -0.45710262656211853,
+      "logits/rejected": -0.3050076961517334,
+      "logps/chosen": -0.671418309211731,
+      "logps/rejected": -1.1854102611541748,
+      "loss": 0.6842,
+      "odds_ratio_loss": 0.4368383288383484,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.03357091173529625,
+      "rewards/margins": 0.02569960430264473,
+      "rewards/rejected": -0.05927051231265068,
+      "sft_loss": 0.671418309211731,
+      "step": 1170
+    },
+    {
+      "epoch": 2.8314337132573484,
+      "grad_norm": 1.3507137389822068,
+      "learning_rate": 4.4666198168422656e-08,
+      "logits/chosen": 0.33376216888427734,
+      "logits/rejected": 0.41172194480895996,
+      "logps/chosen": -0.6510582566261292,
+      "logps/rejected": -1.0800405740737915,
+      "loss": 0.6747,
+      "odds_ratio_loss": 0.5277644395828247,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.032552916556596756,
+      "rewards/margins": 0.021449116989970207,
+      "rewards/rejected": -0.054002027958631516,
+      "sft_loss": 0.6510582566261292,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8554289142171565,
+      "grad_norm": 1.6874037821147798,
+      "learning_rate": 3.343509375168863e-08,
+      "logits/chosen": 0.20301933586597443,
+      "logits/rejected": 0.32382094860076904,
+      "logps/chosen": -0.6405006647109985,
+      "logps/rejected": -1.0241023302078247,
+      "loss": 0.6718,
+      "odds_ratio_loss": 0.48166948556900024,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.03202503174543381,
+      "rewards/margins": 0.019180091097950935,
+      "rewards/rejected": -0.051205117255449295,
+      "sft_loss": 0.6405006647109985,
+      "step": 1190
+    },
+    {
+      "epoch": 2.8794241151769646,
+      "grad_norm": 1.6417139708130921,
+      "learning_rate": 2.3818243341637293e-08,
+      "logits/chosen": -0.3619822859764099,
+      "logits/rejected": -0.15361133217811584,
+      "logps/chosen": -0.6599988341331482,
+      "logps/rejected": -1.098881483078003,
+      "loss": 0.6565,
+      "odds_ratio_loss": 0.456063449382782,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -0.03299994021654129,
+      "rewards/margins": 0.021944135427474976,
+      "rewards/rejected": -0.054944075644016266,
+      "sft_loss": 0.6599988341331482,
+      "step": 1200
+    },
+    {
+      "epoch": 2.9034193161367727,
+      "grad_norm": 1.648932215503252,
+      "learning_rate": 1.5821940727361874e-08,
+      "logits/chosen": -0.7362561821937561,
+      "logits/rejected": -0.4996170997619629,
+      "logps/chosen": -0.6824958920478821,
+      "logps/rejected": -0.9969790577888489,
+      "loss": 0.7067,
+      "odds_ratio_loss": 0.5307115316390991,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.034124795347452164,
+      "rewards/margins": 0.01572415977716446,
+      "rewards/rejected": -0.049848951399326324,
+      "sft_loss": 0.6824958920478821,
+      "step": 1210
+    },
+    {
+      "epoch": 2.927414517096581,
+      "grad_norm": 1.7678674281978446,
+      "learning_rate": 9.451419123484573e-09,
+      "logits/chosen": -0.15318191051483154,
+      "logits/rejected": 0.047946538776159286,
+      "logps/chosen": -0.6560810804367065,
+      "logps/rejected": -1.0658347606658936,
+      "loss": 0.6692,
+      "odds_ratio_loss": 0.5046226382255554,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.032804060727357864,
+      "rewards/margins": 0.02048768661916256,
+      "rewards/rejected": -0.053291745483875275,
+      "sft_loss": 0.6560810804367065,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9514097180563885,
+      "grad_norm": 1.4413325593301094,
+      "learning_rate": 4.710847745256209e-09,
+      "logits/chosen": 0.12647075951099396,
+      "logits/rejected": 0.2795228958129883,
+      "logps/chosen": -0.6180914640426636,
+      "logps/rejected": -1.0847346782684326,
+      "loss": 0.6722,
+      "odds_ratio_loss": 0.41623228788375854,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -0.030904576182365417,
+      "rewards/margins": 0.02333216182887554,
+      "rewards/rejected": -0.05423673242330551,
+      "sft_loss": 0.6180914640426636,
+      "step": 1230
+    },
+    {
+      "epoch": 2.9754049190161966,
+      "grad_norm": 1.5296676400661524,
+      "learning_rate": 1.603329079994942e-09,
+      "logits/chosen": -0.3425149619579315,
+      "logits/rejected": -0.06856220215559006,
+      "logps/chosen": -0.6569226980209351,
+      "logps/rejected": -1.1020539999008179,
+      "loss": 0.6649,
+      "odds_ratio_loss": 0.4642546772956848,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -0.03284613788127899,
+      "rewards/margins": 0.02225656434893608,
+      "rewards/rejected": -0.055102698504924774,
+      "sft_loss": 0.6569226980209351,
+      "step": 1240
+    },
+    {
+      "epoch": 2.994601079784043,
+      "step": 1248,
+      "total_flos": 132590267662336.0,
+      "train_loss": 0.7937506708579186,
+      "train_runtime": 49781.9259,
+      "train_samples_per_second": 1.205,
+      "train_steps_per_second": 0.025
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 1248,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 100.0,
+  "total_flos": 132590267662336.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a5252fd1d5c3ae6e8eedab656003266bd9a9302edb91e20004f9582cf004a79
+size 7032