Upload folder using huggingface_hub

config.json

@@ -0,0 +1,101 @@
+{
+  "vocab_size": 49152,
+  "n_positions": 8192,
+  "n_embd": 6144,
+  "n_layer": 40,
+  "n_head": 48,
+  "n_inner": 24576,
+  "activation_function": "gelu",
+  "resid_pdrop": 0.1,
+  "embd_pdrop": 0.1,
+  "attn_pdrop": 0.1,
+  "layer_norm_epsilon": 1e-05,
+  "initializer_range": 0.02,
+  "scale_attn_weights": true,
+  "use_cache": true,
+  "attention_softmax_in_fp32": true,
+  "scale_attention_softmax_in_fp32": true,
+  "multi_query": true,
+  "bos_token_id": 0,
+  "eos_token_id": 0,
+  "return_dict": true,
+  "output_hidden_states": false,
+  "output_attentions": false,
+  "torchscript": false,
+  "torch_dtype": "float32",
+  "use_bfloat16": false,
+  "tf_legacy_loss": false,
+  "pruned_heads": {},
+  "tie_word_embeddings": true,
+  "is_encoder_decoder": false,
+  "is_decoder": false,
+  "cross_attention_hidden_size": null,
+  "add_cross_attention": false,
+  "tie_encoder_decoder": false,
+  "max_length": 20,
+  "min_length": 0,
+  "do_sample": false,
+  "early_stopping": false,
+  "num_beams": 1,
+  "num_beam_groups": 1,
+  "diversity_penalty": 0.0,
+  "temperature": 1.0,
+  "top_k": 50,
+  "top_p": 1.0,
+  "typical_p": 1.0,
+  "repetition_penalty": 1.0,
+  "length_penalty": 1.0,
+  "no_repeat_ngram_size": 0,
+  "encoder_no_repeat_ngram_size": 0,
+  "bad_words_ids": null,
+  "num_return_sequences": 1,
+  "chunk_size_feed_forward": 0,
+  "output_scores": false,
+  "return_dict_in_generate": false,
+  "forced_bos_token_id": null,
+  "forced_eos_token_id": null,
+  "remove_invalid_values": false,
+  "exponential_decay_length_penalty": null,
+  "suppress_tokens": null,
+  "begin_suppress_tokens": null,
+  "architectures": [
+    "LoRDCoderForCausalLM"
+  ],
+  "finetuning_task": null,
+  "id2label": {
+    "0": "LABEL_0",
+    "1": "LABEL_1"
+  },
+  "label2id": {
+    "LABEL_0": 0,
+    "LABEL_1": 1
+  },
+  "gate_dim": 2304,
+  "auto_map": {
+    "AutoConfig": "configuration_lordcoder_v0.LoRDCoderConfig",
+    "AutoModelForCausalLM": "modeling_lordcoder_v0.LoRDCoderForCausalLM"
+  },
+  "tokenizer_class": null,
+  "prefix": null,
+  "pad_token_id": null,
+  "sep_token_id": null,
+  "decoder_start_token_id": null,
+  "task_specific_params": null,
+  "problem_type": null,
+  "transformers_version": "4.32.0.dev0",
+  "inference_runner": 0,
+  "max_batch_size": null,
+  "max_sequence_length": null,
+  "model_type": "lordcoder",
+  "pad_key_length": true,
+  "pre_allocate_kv_cache": false,
+  "summary_activation": null,
+  "summary_first_dropout": 0.1,
+  "summary_proj_to_labels": true,
+  "summary_type": "cls_index",
+  "summary_use_proj": true,
+  "validate_runner_input": true,
+  "lord_config": {
+    "mlp.c_proj": "directLoRD_double_bias"
+  }
+}

configuration_lordcoder_v0.py

@@ -0,0 +1,124 @@
+"""LoRDCoder configuration class, based on GPT configuration class.
+
+License: Apache-2.0
+"""
+from transformers.configuration_utils import PretrainedConfig
+
+class LoRDCoderConfig(PretrainedConfig):
+    """
+    This is the configuration class to store the configuration of a [`LoRDCoderModel`]. It is used to instantiate a
+    LoRDCoder model according to the specified arguments, defining the model architecture.
+
+    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 50257):
+            Vocabulary size of the GPT-2 model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`LoRDCoderModel`].
+        n_positions (`int`, *optional*, defaults to 1024):
+            The maximum sequence length that this model might ever be used with. Typically set this to something large
+            just in case (e.g., 512 or 1024 or 2048).
+        n_embd (`int`, *optional*, defaults to 768):
+            Dimensionality of the embeddings and hidden states.
+        n_layer (`int`, *optional*, defaults to 12):
+            Number of hidden layers in the Transformer encoder.
+        n_head (`int`, *optional*, defaults to 12):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        n_inner (`int`, *optional*, defaults to None):
+            Dimensionality of the inner feed-forward layers. `None` will set it to 4 times n_embd
+        activation_function (`str`, *optional*, defaults to `"gelu_pytorch_tanh"`):
+            Activation function, to be selected in the list `["relu", "silu", "gelu", "tanh", "gelu_new",
+            "gelu_pytorch_tanh"]`.
+        resid_pdrop (`float`, *optional*, defaults to 0.1):
+            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        embd_pdrop (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for the embeddings.
+        attn_pdrop (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for the attention.
+        layer_norm_epsilon (`float`, *optional*, defaults to 1e-5):
+            The epsilon to use in the layer normalization layers.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        scale_attn_weights (`bool`, *optional*, defaults to `True`):
+            Scale attention weights by dividing by sqrt(hidden_size)..
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models).
+        attention_softmax_in_fp32 (`bool`, *optional*, defaults to `True`):
+            Whether to call the fused softmax in float32.
+        scale_attention_softmax_in_fp32 (`bool`, *optional*, defaults to `True`):
+            Whether to scale the attention softmax in float32.
+        attention_type (`bool`, *optional*, defaults to `True`):
+            Whether to use Multi-Query Attion (`True`) or Multi-Head Attention (`False`).
+    Example:
+
+    ```python
+    >>> from transformers import LoRDCoderConfig, LoRDCoderModel
+
+    >>> # Initializing a LoRDCoder configuration
+    >>> configuration = LoRDCoderConfig()
+
+    >>> # Initializing a model (with random weights) from the configuration
+    >>> model = LoRDCoderModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "lordcoder"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    attribute_map = {
+        "hidden_size": "n_embd",
+        "max_position_embeddings": "n_positions",
+        "num_attention_heads": "n_head",
+        "num_hidden_layers": "n_layer",
+    }
+
+    def __init__(
+        self,
+        vocab_size=50257,
+        n_positions=1024,
+        n_embd=768,
+        n_layer=12,
+        n_head=12,
+        n_inner=None,
+        activation_function="gelu_pytorch_tanh",
+        resid_pdrop=0.1,
+        embd_pdrop=0.1,
+        attn_pdrop=0.1,
+        layer_norm_epsilon=1e-5,
+        initializer_range=0.02,
+        scale_attn_weights=True,
+        use_cache=True,
+        bos_token_id=50256,
+        eos_token_id=50256,
+        attention_softmax_in_fp32=True,
+        scale_attention_softmax_in_fp32=True,
+        multi_query=True,
+        gate_dim=4096,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.n_positions = n_positions
+        self.n_embd = n_embd
+        self.n_layer = n_layer
+        self.n_head = n_head
+        self.n_inner = n_inner
+        self.activation_function = activation_function
+        self.resid_pdrop = resid_pdrop
+        self.embd_pdrop = embd_pdrop
+        self.attn_pdrop = attn_pdrop
+        self.layer_norm_epsilon = layer_norm_epsilon
+        self.initializer_range = initializer_range
+        self.scale_attn_weights = scale_attn_weights
+        self.use_cache = use_cache
+        self.attention_softmax_in_fp32 = attention_softmax_in_fp32
+        self.scale_attention_softmax_in_fp32 = scale_attention_softmax_in_fp32
+        self.multi_query = multi_query
+        self.gate_dim = gate_dim
+
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+
+        super().__init__(bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "eos_token_id": 0,
+  "transformers_version": "4.32.0.dev0"
+}

modeling_lordcoder_v0.py

@@ -0,0 +1,770 @@
+"""LoRDCoder model class, based on GPT model.
+
+License: Apache-2.0
+"""
+import math
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
+
+from transformers.activations import ACT2FN
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPastAndCrossAttentions,
+    CausalLMOutputWithCrossAttentions,
+    SequenceClassifierOutputWithPast,
+    TokenClassifierOutput,
+)
+from transformers.modeling_utils import PreTrainedModel
+from .configuration_lordcoder import LoRDCoderConfig
+
+
+# Fused kernels
+# Use separate functions for each case because conditionals prevent kernel fusion.
+@torch.jit.script
+def upcast_masked_softmax(
+    x: torch.Tensor, mask: torch.Tensor, mask_value: torch.Tensor, scale: float, softmax_dtype: torch.dtype
+):
+    input_dtype = x.dtype
+    x = x.to(softmax_dtype) * scale
+    x = torch.where(mask, x, mask_value)
+    x = torch.nn.functional.softmax(x, dim=-1).to(input_dtype)
+    return x
+
+
+@torch.jit.script
+def upcast_softmax(x: torch.Tensor, scale: float, softmax_dtype: torch.dtype):
+    input_dtype = x.dtype
+    x = x.to(softmax_dtype) * scale
+    x = torch.nn.functional.softmax(x, dim=-1).to(input_dtype)
+    return x
+
+
+@torch.jit.script
+def masked_softmax(x: torch.Tensor, mask: torch.Tensor, mask_value: torch.Tensor):
+    x = torch.where(mask, x, mask_value)
+    x = torch.nn.functional.softmax(x, dim=-1)
+    return x
+
+
+class LoRDCoderAttention(nn.Module):
+    def __init__(self, config, is_cross_attention=False, layer_idx=None):
+        super().__init__()
+        self.mask_value = None
+
+        self.multi_query = config.multi_query
+        self.embed_dim = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.embed_dim // self.num_heads
+        self.kv_heads = 1 if self.multi_query else self.num_heads
+        self.kv_dim = self.kv_heads * self.head_dim
+        self.split_size = self.embed_dim
+        if self.head_dim * self.num_heads != self.embed_dim:
+            raise ValueError(
+                f"`embed_dim` must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:"
+                f" {self.num_heads})."
+            )
+
+        self.scale_attn_weights = config.scale_attn_weights
+        self.is_cross_attention = is_cross_attention
+
+        self.layer_idx = layer_idx
+        self.attention_softmax_in_fp32 = config.attention_softmax_in_fp32
+        self.scale_attention_softmax_in_fp32 = (
+            config.scale_attention_softmax_in_fp32 and config.attention_softmax_in_fp32
+        )
+
+        if self.is_cross_attention:
+            raise NotImplementedError("Cross Attention not supported.")
+            if self.multi_query:
+                raise NotImplementedError("Multi-Query Attention not supported for cross_attention")
+
+            self.c_attn = nn.Linear(self.embed_dim, 2 * self.embed_dim)
+            self.q_attn = nn.Linear(self.embed_dim, self.embed_dim)
+        else:
+            self.c_attn = nn.Linear(self.embed_dim, self.embed_dim + 2 * self.kv_dim)
+
+        self.c_proj = nn.Linear(self.embed_dim, self.embed_dim)
+
+        self.attn_dropout = nn.Dropout(config.attn_pdrop)
+        self.resid_dropout = nn.Dropout(config.resid_pdrop)
+
+    def _get_mask_value(self, device, dtype):
+        # torch.where expects a tensor. We use a cache to avoid recreating it every time.
+        if self.mask_value is None or self.mask_value.dtype != dtype or self.mask_value.device != device:
+            self.mask_value = torch.full([], torch.finfo(dtype).min, dtype=dtype, device=device)
+        return self.mask_value
+
+    def _attn(self, query, key, value, attention_mask=None, head_mask=None):
+        dtype = query.dtype
+        softmax_dtype = torch.float32 if self.attention_softmax_in_fp32 else dtype
+        upcast = dtype != softmax_dtype
+
+        unscale = self.layer_idx + 1 if self.scale_attention_softmax_in_fp32 and upcast else 1
+        scale_factor = unscale**-1
+        if self.scale_attn_weights:
+            scale_factor /= self.head_dim**0.5
+
+        # MQA models: (batch_size, query_length, num_heads * head_dim)
+        # MHA models: (batch_size, num_heads, query_length, head_dim)
+        query_shape = query.shape
+        batch_size = query_shape[0]
+        key_length = key.size(-1)
+        if self.multi_query:
+            # (batch_size, query_length, num_heads, head_dim) x (batch_size, head_dim, key_length)
+            # -> (batch_size, query_length, num_heads, key_length)
+            query_length = query_shape[1]
+            attn_shape = (batch_size, query_length, self.num_heads, key_length)
+            attn_view = (batch_size, query_length * self.num_heads, key_length)
+            # No copy needed for MQA 2, or when layer_past is provided.
+            query = query.reshape(batch_size, query_length * self.num_heads, self.head_dim)
+        else:
+            # (batch_size, num_heads, query_length, head_dim) x (batch_size, num_heads, head_dim, key_length)
+            # -> (batch_size, num_heads, query_length, key_length)
+            query_length = query_shape[2]
+            attn_shape = (batch_size, self.num_heads, query_length, key_length)
+            attn_view = (batch_size * self.num_heads, query_length, key_length)
+            # Always copies
+            query = query.reshape(batch_size * self.num_heads, query_length, self.head_dim)
+            # No copy when layer_past is provided.
+            key = key.reshape(batch_size * self.num_heads, self.head_dim, key_length)
+
+        attn_weights = torch.empty(attn_view, device=query.device, dtype=query.dtype)
+        if query.device.type == "cpu":
+            # This is needed because of a bug in pytorch https://github.com/pytorch/pytorch/issues/80588.
+            # The bug was fixed in https://github.com/pytorch/pytorch/pull/96086,
+            # but the fix has not been released as of pytorch version 2.0.0.
+            attn_weights = torch.zeros_like(attn_weights)
+            beta = 1
+        else:
+            beta = 0
+        attn_weights = torch.baddbmm(attn_weights, query, key, beta=beta, alpha=scale_factor).view(attn_shape)
+
+        if upcast:
+            # Use a fused kernel to prevent a large overhead from casting and scaling.
+            # Sub-optimal when the key length is not a multiple of 8.
+            if attention_mask is None:
+                attn_weights = upcast_softmax(attn_weights, unscale, softmax_dtype)
+            else:
+                mask_value = self._get_mask_value(attn_weights.device, softmax_dtype)
+                # print(attn_weights.device, attention_mask.device, mask_value.device, unscale.device, softmax_dtype)
+                attn_weights = upcast_masked_softmax(attn_weights, attention_mask, mask_value, unscale, softmax_dtype)
+        else:
+            if attention_mask is not None:
+                mask_value = self._get_mask_value(attn_weights.device, softmax_dtype)
+
+                # The fused kernel is very slow when the key length is not a multiple of 8, so we skip fusion.
+                attn_weights = torch.where(attention_mask, attn_weights, mask_value)
+
+            attn_weights = torch.nn.functional.softmax(attn_weights, dim=-1)
+
+        attn_weights = self.attn_dropout(attn_weights)
+
+        # Mask heads if we want to
+        if head_mask is not None:
+            if self.multi_query:
+                head_mask = head_mask.transpose(1, 2)
+            attn_weights = attn_weights * head_mask
+
+        if self.multi_query:
+            attn_output = torch.bmm(attn_weights.view(attn_view), value).view(query_shape)
+        else:
+            attn_output = torch.matmul(attn_weights, value)
+
+        return attn_output, attn_weights
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        layer_past: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        encoder_attention_mask: Optional[torch.Tensor] = None,
+        use_cache: Optional[bool] = False,
+        output_attentions: Optional[bool] = False,
+    ) -> Union[
+        Tuple[torch.Tensor, Optional[torch.Tensor]],
+        Tuple[torch.Tensor, Optional[torch.Tensor], Tuple[torch.Tensor, ...]],
+    ]:
+        if encoder_hidden_states is not None:
+            if not hasattr(self, "q_attn") or not self.is_cross_attention:
+                raise ValueError(
+                    "If class is used as cross attention, the weights `q_attn` have to be defined. "
+                    "Please make sure to instantiate class with `LoRDCoderAttention(..., is_cross_attention=True)`."
+                )
+
+            query = self.q_attn(hidden_states)
+            key_value = self.c_attn(encoder_hidden_states)
+            attention_mask = encoder_attention_mask
+        elif self.multi_query:
+            query, key_value = self.c_attn(hidden_states).split((self.embed_dim, 2 * self.kv_dim), dim=2)
+        else:
+            # Note: We split as (self.num_heads, 3, self.head_dim) instead of (3, self.num_heads, self.head_dim),
+            # i.e., the memory layout is not the same as GPT2.
+            # This makes the concatenation with past_key_value more efficient.
+            query, key_value = (
+                self.c_attn(hidden_states)
+                .view(*hidden_states.shape[:2], self.num_heads, 3 * self.head_dim)
+                .transpose(1, 2)
+                .split((self.head_dim, 2 * self.head_dim), dim=3)
+            )
+
+        if layer_past is not None:
+            key_value = torch.cat((layer_past, key_value), dim=-2)
+        present = key_value if use_cache else None
+
+        key, value = key_value.split((self.head_dim, self.head_dim), dim=-1)
+
+        attn_output, attn_weights = self._attn(query, key.transpose(-1, -2), value, attention_mask, head_mask)
+
+        if not self.multi_query:
+            attn_output = attn_output.transpose(1, 2).reshape(hidden_states.shape)
+        attn_output = self.c_proj(attn_output)
+        attn_output = self.resid_dropout(attn_output)
+
+        outputs = (attn_output, present)
+        if output_attentions:
+            if self.multi_query:
+                # Transpose to return weights in the usual format (batch_size, num_heads, query_length, key_length)
+                attn_weights = attn_weights.transpose(1, 2)
+            outputs += (attn_weights,)
+
+        return outputs  # a, present, (attentions)
+
+
+class LoRDCoderMLP(nn.Module):
+    def __init__(self, intermediate_size, config):
+        super().__init__()
+        embed_dim = config.hidden_size
+        self.gate_dim = config.gate_dim
+
+        self.c_fc = torch.nn.Linear(in_features=embed_dim, out_features=intermediate_size, bias=True)
+        self.c_gate = torch.nn.Linear(in_features=intermediate_size, out_features=self.gate_dim, bias=True)
+        self.c_proj = torch.nn.Linear(in_features=self.gate_dim, out_features=embed_dim, bias=True)
+
+        self.act = ACT2FN[config.activation_function]
+        self.dropout = nn.Dropout(config.resid_pdrop)
+
+    def forward(self, hidden_states: Optional[Tuple[torch.FloatTensor]]) -> torch.FloatTensor:
+        hidden_states = self.c_fc(hidden_states)
+        hidden_states = self.c_gate(self.act(hidden_states))
+        hidden_states = self.c_proj(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        return hidden_states
+
+
+class LoRDCoderBlock(nn.Module):
+    def __init__(self, config, layer_idx=None):
+        super().__init__()
+        hidden_size = config.hidden_size
+        self.inner_dim = config.n_inner if config.n_inner is not None else 4 * hidden_size
+
+        self.ln_1 = nn.LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
+        self.attn = LoRDCoderAttention(config, layer_idx=layer_idx)
+        self.ln_2 = nn.LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
+
+        if config.add_cross_attention:
+            if config.multi_query:
+                raise NotImplementedError("Cross-attention not implemented for MQA")
+            self.crossattention = LoRDCoderAttention(config, is_cross_attention=True, layer_idx=layer_idx)
+            self.ln_cross_attn = nn.LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
+
+        self.mlp = LoRDCoderMLP(self.inner_dim, config)
+
+    def forward(
+        self,
+        hidden_states: Optional[Tuple[torch.Tensor]],
+        layer_past: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        encoder_attention_mask: Optional[torch.Tensor] = None,
+        use_cache: Optional[bool] = False,
+        output_attentions: Optional[bool] = False,
+    ) -> Union[
+        Tuple[torch.Tensor], Tuple[torch.Tensor, torch.Tensor], Tuple[torch.Tensor, torch.Tensor, torch.Tensor]
+    ]:
+        residual = hidden_states
+        hidden_states = self.ln_1(hidden_states)
+        attn_outputs = self.attn(
+            hidden_states,
+            layer_past=layer_past,
+            attention_mask=attention_mask,
+            head_mask=head_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+        )
+        attn_output = attn_outputs[0]  # output_attn: a, present, (attentions)
+        outputs = attn_outputs[1:]
+        # residual connection
+        hidden_states = attn_output + residual
+
+        if encoder_hidden_states is not None:
+            # add one self-attention block for cross-attention
+            if not hasattr(self, "crossattention"):
+                raise ValueError(
+                    f"If `encoder_hidden_states` are passed, {self} has to be instantiated with "
+                    "cross-attention layers by setting `config.add_cross_attention=True`"
+                )
+            residual = hidden_states
+            hidden_states = self.ln_cross_attn(hidden_states)
+            cross_attn_outputs = self.crossattention(
+                hidden_states,
+                attention_mask=attention_mask,
+                head_mask=head_mask,
+                encoder_hidden_states=encoder_hidden_states,
+                encoder_attention_mask=encoder_attention_mask,
+                output_attentions=output_attentions,
+            )
+            attn_output = cross_attn_outputs[0]
+            # residual connection
+            hidden_states = residual + attn_output
+            outputs = outputs + cross_attn_outputs[2:]  # add cross attentions if we output attention weights
+
+        residual = hidden_states
+        hidden_states = self.ln_2(hidden_states)
+        feed_forward_hidden_states = self.mlp(hidden_states)
+        # residual connection
+        hidden_states = residual + feed_forward_hidden_states
+
+        if use_cache:
+            outputs = (hidden_states,) + outputs
+        else:
+            outputs = (hidden_states,) + outputs[1:]
+
+        return outputs  # hidden_states, present, (attentions, cross_attentions)
+
+
+class LoRDCoderPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = LoRDCoderConfig
+    base_model_prefix = "transformer"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["LoRDCoderBlock"]
+    _skip_keys_device_placement = "past_key_values"
+
+    def __init__(self, *inputs, **kwargs):
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(self, module):
+        """Initialize the weights."""
+        if isinstance(module, LoRDCoderMLP):
+            # Reinitialize selected weights subject to the OpenAI GPT-2 Paper Scheme:
+            #   > A modified initialization which accounts for the accumulation on the residual path with model depth. Scale
+            #   > the weights of residual layers at initialization by a factor of 1/√N where N is the # of residual layers.
+            #   >   -- GPT-2 :: https://openai.com/blog/better-language-models/
+            #
+            # Reference (Megatron-LM): https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/model/gpt_model.py
+            module.c_proj.weight.data.normal_(
+                mean=0.0, std=(self.config.initializer_range / math.sqrt(2 * self.config.n_layer))
+            )
+            module.c_proj._is_hf_initialized = True
+        elif isinstance(module, LoRDCoderAttention):
+            # Reinitialize selected weights subject to the OpenAI GPT-2 Paper Scheme:
+            #   > A modified initialization which accounts for the accumulation on the residual path with model depth. Scale
+            #   > the weights of residual layers at initialization by a factor of 1/√N where N is the # of residual layers.
+            #   >   -- GPT-2 :: https://openai.com/blog/better-language-models/
+            #
+            # Reference (Megatron-LM): https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/model/gpt_model.py
+            module.c_proj.weight.data.normal_(
+                mean=0.0, std=(self.config.initializer_range / math.sqrt(2 * self.config.n_layer))
+            )
+            module.c_proj.weight.data.normal_(
+                mean=0.0, std=(self.config.initializer_range / math.sqrt(2 * self.config.n_layer))
+            )
+            module.c_proj._is_hf_initialized = True
+        elif isinstance(module, nn.Linear):
+            # Slightly different from the TF version which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+        elif isinstance(module, nn.LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, LoRDCoderModel):
+            module.gradient_checkpointing = value
+
+
+class LoRDCoderModel(LoRDCoderPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.multi_query = config.multi_query
+        self.embed_dim = config.hidden_size
+
+        self.wte = nn.Embedding(config.vocab_size, self.embed_dim)
+        self.wpe = nn.Embedding(config.max_position_embeddings, self.embed_dim)
+
+        self.drop = nn.Dropout(config.embd_pdrop)
+        self.h = nn.ModuleList([LoRDCoderBlock(config, layer_idx=i) for i in range(config.num_hidden_layers)])
+        self.ln_f = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_epsilon)
+
+        max_positions = config.max_position_embeddings
+        self.register_buffer(
+            "bias", torch.tril(torch.ones((max_positions, max_positions), dtype=torch.bool)), persistent=False
+        )
+
+        self.gradient_checkpointing = False
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.wte
+
+    def set_input_embeddings(self, new_embeddings):
+        self.wte = new_embeddings
+
+    def warn_if_padding_and_no_attention_mask(self, input_ids, attention_mask):
+        """
+        Shows a one-time warning if the input_ids appear to contain padding and no attention mask was given.
+        """
+
+        if (attention_mask is not None) or (self.config.pad_token_id is None):
+            return
+
+        # Check only the first and last input IDs to reduce overhead.
+        if self.config.pad_token_id in input_ids[:, [-1, 0]]:
+            warn_string = (
+                "We strongly recommend passing in an `attention_mask` since your input_ids may be padded. See "
+                "https://huggingface.co/docs/transformers/troubleshooting"
+                "#incorrect-output-when-padding-tokens-arent-masked."
+            )
+
+            # If the pad token is equal to either BOS, EOS, or SEP, we do not know whether the user should use an
+            # attention_mask or not. In this case, we should still show a warning because this is a rare case.
+            if (
+                (self.config.bos_token_id is not None and self.config.bos_token_id == self.config.pad_token_id)
+                or (self.config.eos_token_id is not None and self.config.eos_token_id == self.config.pad_token_id)
+                or (self.config.sep_token_id is not None and self.config.sep_token_id == self.config.pad_token_id)
+            ):
+                warn_string += (
+                    f"\nYou may ignore this warning if your `pad_token_id` ({self.config.pad_token_id}) is identical "
+                    f"to the `bos_token_id` ({self.config.bos_token_id}), `eos_token_id` ({self.config.eos_token_id}), "
+                    f"or the `sep_token_id` ({self.config.sep_token_id}), and your input is not padded."
+                )
+
+            print("Warning:", warn_string)
+
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        past_key_values: Optional[List[torch.Tensor]] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        encoder_attention_mask: 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,
+    ) -> Union[Tuple, BaseModelOutputWithPastAndCrossAttentions]:
+        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
+
+        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:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
+            input_shape = input_ids.size()
+            input_ids = input_ids.view(-1, input_shape[-1])
+            batch_size = input_ids.shape[0]
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+            batch_size = inputs_embeds.shape[0]
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        if batch_size <= 0:
+            raise ValueError("batch_size has to be defined and > 0")
+
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+
+        if token_type_ids is not None:
+            token_type_ids = token_type_ids.view(-1, input_shape[-1])
+        if position_ids is not None:
+            position_ids = position_ids.view(-1, input_shape[-1])
+
+        if past_key_values is None:
+            past_length = 0
+            past_key_values = tuple([None] * len(self.h))
+        else:
+            past_length = past_key_values[0].size(-2)
+
+        if attention_mask is not None and len(attention_mask.shape) == 2 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_length > 0:
+                position_ids = position_ids[:, past_length : input_shape[-1] + past_length :]
+        elif position_ids is None:
+            position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device)
+            position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
+
+        # Self-attention mask.
+        query_length = input_shape[-1]
+        key_length = past_length + query_length
+        self_attention_mask = self.bias[None, key_length - query_length : key_length, :key_length]
+
+        if attention_mask is not None:
+            self_attention_mask = self_attention_mask * attention_mask.view(batch_size, 1, -1).to(
+                dtype=torch.bool, device=self_attention_mask.device
+            )
+
+        # MQA models: (batch_size, query_length, n_heads, key_length)
+        # MHA models: (batch_size, n_heads, query_length, key_length)
+        attention_mask = self_attention_mask.unsqueeze(2 if self.multi_query else 1)
+
+        # If a 2D or 3D attention mask is provided for the cross-attention
+        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
+        if (
+            self.config.add_cross_attention
+            and encoder_hidden_states is not None
+            and encoder_attention_mask is not None
+        ):
+            if encoder_attention_mask.dim() == 2:
+                encoder_attention_mask.unsqueeze(1)
+            assert encoder_attention_mask.dim() == 3
+            encoder_attention_mask = encoder_attention_mask.bool().unsqueeze(2 if self.multi_query else 1)
+        else:
+            encoder_attention_mask = None
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x n_heads x N x N
+        # head_mask has shape n_layer x batch x n_heads x N x N
+        head_mask = self.get_head_mask(head_mask, self.config.n_layer)
+
+        if inputs_embeds is None:
+            inputs_embeds = self.wte(input_ids)
+        position_embeds = self.wpe(position_ids)
+        hidden_states = inputs_embeds + position_embeds
+
+        if token_type_ids is not None:
+            token_type_embeds = self.wte(token_type_ids)
+            hidden_states = hidden_states + token_type_embeds
+
+        hidden_states = self.drop(hidden_states)
+
+        output_shape = input_shape + (hidden_states.size(-1),)
+
+        presents = [] if use_cache else None
+        all_self_attentions = () if output_attentions else None
+        all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
+        all_hidden_states = () if output_hidden_states else None
+        for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        # None for past_key_value
+                        return module(*inputs, use_cache, output_attentions)
+
+                    return custom_forward
+
+                outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(block),
+                    hidden_states,
+                    None,
+                    attention_mask,
+                    head_mask[i],
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                )
+            else:
+                outputs = block(
+                    hidden_states,
+                    layer_past=layer_past,
+                    attention_mask=attention_mask,
+                    head_mask=head_mask[i],
+                    encoder_hidden_states=encoder_hidden_states,
+                    encoder_attention_mask=encoder_attention_mask,
+                    use_cache=use_cache,
+                    output_attentions=output_attentions,
+                )
+
+            hidden_states = outputs[0]
+            if use_cache:
+                presents.append(outputs[1])
+
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (outputs[2 if use_cache else 1],)
+                if self.config.add_cross_attention:
+                    all_cross_attentions = all_cross_attentions + (outputs[3 if use_cache else 2],)
+
+        hidden_states = self.ln_f(hidden_states)
+
+        hidden_states = hidden_states.view(output_shape)
+        # Add last hidden state
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(
+                v
+                for v in [hidden_states, presents, all_hidden_states, all_self_attentions, all_cross_attentions]
+                if v is not None
+            )
+
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=presents,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+            cross_attentions=all_cross_attentions,
+        )
+
+
+class LoRDCoderForCausalLM(LoRDCoderPreTrainedModel):
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.transformer = LoRDCoderModel(config)
+        self.lm_head = lambda x: torch.matmul(x, self.transformer.wte.weight.T)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        raise NotImplementedError("Cannot resize the embeddings of LoRDCoderForCausalLM.")
+
+    def prepare_inputs_for_generation(self, input_ids, past_key_values=None, inputs_embeds=None, **kwargs):
+        token_type_ids = kwargs.get("token_type_ids", None)
+        # only last token for inputs_ids if past is defined in kwargs
+        if past_key_values:
+            input_ids = input_ids[:, -1].unsqueeze(-1)
+            if token_type_ids is not None:
+                token_type_ids = token_type_ids[:, -1].unsqueeze(-1)
+
+        attention_mask = kwargs.get("attention_mask", None)
+        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[:, -1].unsqueeze(-1)
+        else:
+            position_ids = None
+
+        # 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(
+            {
+                "past_key_values": past_key_values,
+                "use_cache": kwargs.get("use_cache"),
+                "position_ids": position_ids,
+                "attention_mask": attention_mask,
+                "token_type_ids": token_type_ids,
+            }
+        )
+        return model_inputs
+
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        encoder_attention_mask: 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,
+    ) -> Union[Tuple, CausalLMOutputWithCrossAttentions]:
+        r"""
+        labels (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
+            `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
+            are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        transformer_outputs = self.transformer(
+            input_ids,
+            past_key_values=past_key_values,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        hidden_states = transformer_outputs[0]
+
+        lm_logits = self.lm_head(hidden_states)
+
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            shift_logits = lm_logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous().to(shift_logits.device)
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
+
+        if not return_dict:
+            output = (lm_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return CausalLMOutputWithCrossAttentions(
+            loss=loss,
+            logits=lm_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+            cross_attentions=transformer_outputs.cross_attentions,
+        )
+
+    @staticmethod
+    def _reorder_cache(
+        past_key_values: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor
+    ) -> Tuple[Tuple[torch.Tensor]]:
+        """
+        This function is used to re-order the `past_key_values` cache if [`~PreTrainedModel.beam_search`] or
+        [`~PreTrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
+        beam_idx at every generation step.
+        """
+        return tuple(layer_past.index_select(0, beam_idx.to(layer_past.device)) for layer_past in past_key_values)

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7814a14add602d7880a0d92e5b27b218cb3e5a02a24621e15639d6bd25d64d4a
+size 24617994449

special_tokens_map.json

@@ -0,0 +1,26 @@
+{
+  "additional_special_tokens": [
+    "<|endoftext|>",
+    "<fim_prefix>",
+    "<fim_middle>",
+    "<fim_suffix>",
+    "<fim_pad>",
+    "<filename>",
+    "<gh_stars>",
+    "<issue_start>",
+    "<issue_comment>",
+    "<issue_closed>",
+    "<jupyter_start>",
+    "<jupyter_text>",
+    "<jupyter_code>",
+    "<jupyter_output>",
+    "<empty_output>",
+    "<commit_before>",
+    "<commit_msg>",
+    "<commit_after>",
+    "<reponame>"
+  ],
+  "bos_token": "<|endoftext|>",
+  "eos_token": "<|endoftext|>",
+  "unk_token": "<|endoftext|>"
+}

tokenizer_config.json

@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "additional_special_tokens": [
+    "<|endoftext|>",
+    "<fim_prefix>",
+    "<fim_middle>",
+    "<fim_suffix>",
+    "<fim_pad>",
+    "<filename>",
+    "<gh_stars>",
+    "<issue_start>",
+    "<issue_comment>",
+    "<issue_closed>",
+    "<jupyter_start>",
+    "<jupyter_text>",
+    "<jupyter_code>",
+    "<jupyter_output>",
+    "<empty_output>",
+    "<commit_before>",
+    "<commit_msg>",
+    "<commit_after>",
+    "<reponame>"
+  ],
+  "bos_token": "<|endoftext|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|endoftext|>",
+  "model_max_length": 1000000000000000019884624838656,
+  "tokenizer_class": "GPT2Tokenizer",
+  "unk_token": "<|endoftext|>",
+  "vocab_size": 49152
+}