Upload folder using huggingface_hub

README.md

@@ -7,11 +7,10 @@ base_model: ai21labs/Jamba-v0.1
 model-index:
 - name: out
   results: []
-datasets:
-- chargoddard/Open-Platypus-Chat
 ---
 
-# Jambatypus
+<!-- 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. -->
 
 [<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
 <details><summary>See axolotl config</summary>
@@ -33,7 +32,6 @@ chat_template: chatml
 dataset_prepared_path:
 val_set_size: 0.01
 output_dir: ./out
-hub_strategy: every_save
 
 sequence_len: 4096
 sample_packing: true
@@ -80,9 +78,6 @@ logging_steps: 1
 xformers_attention:
 flash_attention: true
 
-loss_watchdog_threshold: 5.0
-loss_watchdog_patience: 3
-
 warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 4
@@ -96,9 +91,23 @@ special_tokens:
 
 </details><br>
 
-This model is a fine-tuned version of [ai21labs/Jamba-v0.1](https://huggingface.co/ai21labs/Jamba-v0.1) on the [chargoddard/Open-Platypus-Chat](https://huggingface.co/datasets/chargoddard/Open-Platypus-Chat) dataset.
+# out
+
+This model is a fine-tuned version of [ai21labs/Jamba-v0.1](https://huggingface.co/ai21labs/Jamba-v0.1) on the None dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.9573
+- Loss: 0.9651
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
 
 ## Training procedure
 
@@ -109,8 +118,11 @@ The following hyperparameters were used during training:
 - train_batch_size: 1
 - eval_batch_size: 1
 - seed: 42
+- distributed_type: multi-GPU
+- num_devices: 2
 - gradient_accumulation_steps: 8
-- total_train_batch_size: 8
+- total_train_batch_size: 16
+- total_eval_batch_size: 2
 - optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-05
 - lr_scheduler_type: cosine
 - lr_scheduler_warmup_steps: 10
@@ -120,11 +132,11 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch | Step | Validation Loss |
 |:-------------:|:-----:|:----:|:---------------:|
-| 0.5915        | 0.0   | 1    | 1.0279          |
-| 0.4249        | 0.25  | 84   | 0.9711          |
-| 0.3911        | 0.5   | 168  | 0.9590          |
-| 0.4655        | 0.75  | 252  | 0.9628          |
-| 0.4569        | 1.0   | 336  | 0.9573          |
+| 0.6274        | 0.01  | 1    | 1.0298          |
+| 0.4403        | 0.25  | 42   | 0.9768          |
+| 0.4417        | 0.5   | 84   | 0.9675          |
+| 0.4451        | 0.75  | 126  | 0.9652          |
+| 0.4616        | 1.0   | 168  | 0.9651          |
 
 
 ### Framework versions

adapter_config.json

@@ -20,18 +20,18 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "q_proj",
+    "x_proj",
+    "o_proj",
+    "up_proj",
+    "v_proj",
+    "down_proj",
+    "dt_proj",
     "k_proj",
+    "q_proj",
     "in_proj",
+    "gate_proj",
     "out_proj",
-    "dt_proj",
-    "down_proj",
-    "router",
-    "up_proj",
-    "o_proj",
-    "v_proj",
-    "x_proj"
+    "router"
   ],
   "task_type": "CAUSAL_LM",
   "use_dora": false,

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:08b546e88fba6a21cb58cb344345b20a7fe1c09f4e948ca61b90164deb1efdd3
-size 1063383458
+oid sha256:485eddbdd38763c9850251f824a3877a9ddc0db9c893fae77c6ad1f9eb0d379c
+size 531653306

checkpoint-126/README.md

@@ -0,0 +1,202 @@
+---
+library_name: peft
+base_model: ai21labs/Jamba-v0.1
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.10.0

checkpoint-126/adapter_config.json

@@ -0,0 +1,39 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "ai21labs/Jamba-v0.1",
+  "bias": "none",
+  "fan_in_fan_out": null,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "x_proj",
+    "o_proj",
+    "up_proj",
+    "v_proj",
+    "down_proj",
+    "dt_proj",
+    "k_proj",
+    "q_proj",
+    "in_proj",
+    "gate_proj",
+    "out_proj",
+    "router"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-126/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be949fb50dbe53d64e9f4375ed327e36ae2368bf683cfdaf813667de418fcbb8
+size 531611600

checkpoint-126/global_step126/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6998ecb2f6d60f27d30371a8ce207805cf8a48dd8f296036fff813d22b65593
+size 797643792

checkpoint-126/global_step126/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f504e0638f11f71fa63e7587596b481c876e9ec17a2e4703833ab1713a855b1
+size 797644432

checkpoint-126/global_step126/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:271bfc43b751ef756538a0f6b19ee11530e5d5c9f68f5e3c3b086877dc1d3a6a
+size 1345660121

checkpoint-126/latest

@@ -0,0 +1 @@
+global_step126

checkpoint-126/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:601237d8f7ccca9157ec802c2d8dbf105a33ca51949abaf17e8caac019fa582f
+size 14512

checkpoint-126/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b453660e9304c04b047fceaa4cad80dfb1b50b01880351c794e0268dc4d5d75
+size 14512

checkpoint-126/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb0b78cebb6cf02731e93f8b5296f19d99fc5821bb80f3f750f5f7f3c9de25e2
+size 1064

checkpoint-126/special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<|startoftext|>",
+    "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": "<|pad|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<|unk|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoint-126/tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02fd6530b8ede0eedd8e509fcab32da7b1dd04c8119f8498c787100f13112713
+size 1124742

checkpoint-126/tokenizer_config.json

@@ -0,0 +1,48 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<|pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<|startoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "<|unk|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<|startoftext|>",
+  "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "<|pad|>",
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<|unk|>",
+  "use_default_system_prompt": false
+}

checkpoint-126/trainer_state.json

@@ -0,0 +1,935 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.7488855869242199,
+  "eval_steps": 42,
+  "global_step": 126,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.30475762367257353,
+      "learning_rate": 2e-05,
+      "loss": 0.6274,
+      "step": 1
+    },
+    {
+      "epoch": 0.01,
+      "eval_loss": 1.0297596454620361,
+      "eval_runtime": 150.6014,
+      "eval_samples_per_second": 1.653,
+      "eval_steps_per_second": 0.83,
+      "step": 1
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.2798137150913395,
+      "learning_rate": 4e-05,
+      "loss": 0.6362,
+      "step": 2
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.3178684451319545,
+      "learning_rate": 6e-05,
+      "loss": 0.6299,
+      "step": 3
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.33651284916847835,
+      "learning_rate": 8e-05,
+      "loss": 0.6391,
+      "step": 4
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.23984915388648712,
+      "learning_rate": 0.0001,
+      "loss": 0.6071,
+      "step": 5
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.20514040410017348,
+      "learning_rate": 0.00012,
+      "loss": 0.5996,
+      "step": 6
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.1950018128286362,
+      "learning_rate": 0.00014,
+      "loss": 0.6298,
+      "step": 7
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.14246019947393238,
+      "learning_rate": 0.00016,
+      "loss": 0.5108,
+      "step": 8
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.15792014279750227,
+      "learning_rate": 0.00018,
+      "loss": 0.5529,
+      "step": 9
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.1517889177511264,
+      "learning_rate": 0.0002,
+      "loss": 0.5433,
+      "step": 10
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.1372121219437277,
+      "learning_rate": 0.00019998023297700658,
+      "loss": 0.5856,
+      "step": 11
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.16740807394942855,
+      "learning_rate": 0.00019992093972273018,
+      "loss": 0.5546,
+      "step": 12
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.13512320693394078,
+      "learning_rate": 0.00019982214367819328,
+      "loss": 0.6193,
+      "step": 13
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.16169796294070152,
+      "learning_rate": 0.0001996838839014696,
+      "loss": 0.5495,
+      "step": 14
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.16796913812281988,
+      "learning_rate": 0.00019950621505224273,
+      "loss": 0.5035,
+      "step": 15
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.1800514764162192,
+      "learning_rate": 0.00019928920737019733,
+      "loss": 0.5083,
+      "step": 16
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.170432124866908,
+      "learning_rate": 0.0001990329466472502,
+      "loss": 0.632,
+      "step": 17
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.19129325489749488,
+      "learning_rate": 0.00019873753419363336,
+      "loss": 0.4813,
+      "step": 18
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.1459357988760762,
+      "learning_rate": 0.00019840308679784207,
+      "loss": 0.4973,
+      "step": 19
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.192594730984382,
+      "learning_rate": 0.00019802973668046363,
+      "loss": 0.5291,
+      "step": 20
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.859025467969139,
+      "learning_rate": 0.0001976176314419051,
+      "loss": 0.5296,
+      "step": 21
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.13366297885670222,
+      "learning_rate": 0.000197166934004041,
+      "loss": 0.4819,
+      "step": 22
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.15698714419747645,
+      "learning_rate": 0.00019667782254580374,
+      "loss": 0.5409,
+      "step": 23
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.10995943735837355,
+      "learning_rate": 0.00019615049043274205,
+      "loss": 0.5108,
+      "step": 24
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.10796742192788925,
+      "learning_rate": 0.00019558514614057609,
+      "loss": 0.5215,
+      "step": 25
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.11641740089490231,
+      "learning_rate": 0.00019498201317277828,
+      "loss": 0.5012,
+      "step": 26
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.1120175962893241,
+      "learning_rate": 0.00019434132997221345,
+      "loss": 0.474,
+      "step": 27
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.1218171278782483,
+      "learning_rate": 0.0001936633498268728,
+      "loss": 0.5216,
+      "step": 28
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.11718521115928844,
+      "learning_rate": 0.0001929483407697387,
+      "loss": 0.4856,
+      "step": 29
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.12611471038571026,
+      "learning_rate": 0.00019219658547282067,
+      "loss": 0.4823,
+      "step": 30
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.11106871615269753,
+      "learning_rate": 0.00019140838113540346,
+      "loss": 0.4869,
+      "step": 31
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.1416503230360699,
+      "learning_rate": 0.00019058403936655233,
+      "loss": 0.5341,
+      "step": 32
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.10761396399791698,
+      "learning_rate": 0.00018972388606192125,
+      "loss": 0.4304,
+      "step": 33
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.10975376180434356,
+      "learning_rate": 0.0001888282612749132,
+      "loss": 0.4646,
+      "step": 34
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.12848879670359908,
+      "learning_rate": 0.00018789751908224338,
+      "loss": 0.4972,
+      "step": 35
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.11904721819683833,
+      "learning_rate": 0.00018693202744395827,
+      "loss": 0.505,
+      "step": 36
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.12249852034224981,
+      "learning_rate": 0.00018593216805796612,
+      "loss": 0.5396,
+      "step": 37
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.12453395046646995,
+      "learning_rate": 0.00018489833620913642,
+      "loss": 0.4917,
+      "step": 38
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.12585770374422164,
+      "learning_rate": 0.00018383094061302766,
+      "loss": 0.5079,
+      "step": 39
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.11095271476322731,
+      "learning_rate": 0.00018273040325430574,
+      "loss": 0.4812,
+      "step": 40
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.12968115101635422,
+      "learning_rate": 0.00018159715921991612,
+      "loss": 0.5106,
+      "step": 41
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.10933018515590627,
+      "learning_rate": 0.00018043165652707649,
+      "loss": 0.4403,
+      "step": 42
+    },
+    {
+      "epoch": 0.25,
+      "eval_loss": 0.9767947196960449,
+      "eval_runtime": 152.0343,
+      "eval_samples_per_second": 1.638,
+      "eval_steps_per_second": 0.822,
+      "step": 42
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.11525719626792096,
+      "learning_rate": 0.00017923435594615744,
+      "loss": 0.482,
+      "step": 43
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.12962154411778218,
+      "learning_rate": 0.00017800573081852122,
+      "loss": 0.5452,
+      "step": 44
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.12555700120045588,
+      "learning_rate": 0.0001767462668693908,
+      "loss": 0.5084,
+      "step": 45
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.11427565378293324,
+      "learning_rate": 0.00017545646201582303,
+      "loss": 0.5191,
+      "step": 46
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.10974901402857151,
+      "learning_rate": 0.00017413682616986185,
+      "loss": 0.4703,
+      "step": 47
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.11781465084480325,
+      "learning_rate": 0.00017278788103694943,
+      "loss": 0.4548,
+      "step": 48
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.10781807228559999,
+      "learning_rate": 0.000171410159909675,
+      "loss": 0.476,
+      "step": 49
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.12502639462035098,
+      "learning_rate": 0.00017000420745694254,
+      "loss": 0.5084,
+      "step": 50
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.10718920826593327,
+      "learning_rate": 0.00016857057950864132,
+      "loss": 0.5093,
+      "step": 51
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.10040549880547282,
+      "learning_rate": 0.0001671098428359037,
+      "loss": 0.4644,
+      "step": 52
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.11778478994740472,
+      "learning_rate": 0.00016562257492703757,
+      "loss": 0.4725,
+      "step": 53
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.1008386031049932,
+      "learning_rate": 0.000164109363759222,
+      "loss": 0.5121,
+      "step": 54
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.1170302528140235,
+      "learning_rate": 0.000162570807566056,
+      "loss": 0.4766,
+      "step": 55
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.1104526773884303,
+      "learning_rate": 0.00016100751460105243,
+      "loss": 0.4886,
+      "step": 56
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.10467920768691032,
+      "learning_rate": 0.00015942010289717105,
+      "loss": 0.4703,
+      "step": 57
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.11551406829220555,
+      "learning_rate": 0.00015780920002248484,
+      "loss": 0.4837,
+      "step": 58
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.11133818831887894,
+      "learning_rate": 0.0001561754428320771,
+      "loss": 0.5148,
+      "step": 59
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.11281448423273216,
+      "learning_rate": 0.00015451947721626676,
+      "loss": 0.4561,
+      "step": 60
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.13934126997471205,
+      "learning_rate": 0.00015284195784526195,
+      "loss": 0.5069,
+      "step": 61
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.11851655387640142,
+      "learning_rate": 0.00015114354791034225,
+      "loss": 0.5094,
+      "step": 62
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.12909148374566123,
+      "learning_rate": 0.0001494249188616723,
+      "loss": 0.581,
+      "step": 63
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.11070161341925377,
+      "learning_rate": 0.00014768675014285062,
+      "loss": 0.4585,
+      "step": 64
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.13308674882888374,
+      "learning_rate": 0.00014592972892229778,
+      "loss": 0.4974,
+      "step": 65
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.12124588853708144,
+      "learning_rate": 0.0001441545498215912,
+      "loss": 0.4463,
+      "step": 66
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.1183570515369953,
+      "learning_rate": 0.00014236191464085286,
+      "loss": 0.447,
+      "step": 67
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.13520024884417237,
+      "learning_rate": 0.00014055253208129938,
+      "loss": 0.5309,
+      "step": 68
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.12184981458813801,
+      "learning_rate": 0.00013872711746506413,
+      "loss": 0.4532,
+      "step": 69
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.12449299540645078,
+      "learning_rate": 0.00013688639245240078,
+      "loss": 0.5198,
+      "step": 70
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.1383134750490429,
+      "learning_rate": 0.00013503108475638244,
+      "loss": 0.5629,
+      "step": 71
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.246237001656926,
+      "learning_rate": 0.0001331619278552068,
+      "loss": 0.4869,
+      "step": 72
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.13337703940933632,
+      "learning_rate": 0.00013127966070222274,
+      "loss": 0.4792,
+      "step": 73
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.12428922033806454,
+      "learning_rate": 0.00012938502743379212,
+      "loss": 0.4825,
+      "step": 74
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.13290774912900208,
+      "learning_rate": 0.00012747877707510252,
+      "loss": 0.5138,
+      "step": 75
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.11185975046756892,
+      "learning_rate": 0.0001255616632440475,
+      "loss": 0.4815,
+      "step": 76
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.1130592868215497,
+      "learning_rate": 0.0001236344438532905,
+      "loss": 0.5046,
+      "step": 77
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.12882943465594857,
+      "learning_rate": 0.0001216978808106318,
+      "loss": 0.5091,
+      "step": 78
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.14837896297082676,
+      "learning_rate": 0.00011975273971779528,
+      "loss": 0.5158,
+      "step": 79
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.1265223309856292,
+      "learning_rate": 0.00011779978956775506,
+      "loss": 0.5068,
+      "step": 80
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.14042502330520407,
+      "learning_rate": 0.0001158398024407215,
+      "loss": 0.5061,
+      "step": 81
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.1261526695491767,
+      "learning_rate": 0.00011387355319890685,
+      "loss": 0.4691,
+      "step": 82
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.12007305451001854,
+      "learning_rate": 0.00011190181918019049,
+      "loss": 0.4753,
+      "step": 83
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.12809956897166885,
+      "learning_rate": 0.00010992537989080618,
+      "loss": 0.4417,
+      "step": 84
+    },
+    {
+      "epoch": 0.5,
+      "eval_loss": 0.9675251841545105,
+      "eval_runtime": 152.4793,
+      "eval_samples_per_second": 1.633,
+      "eval_steps_per_second": 0.82,
+      "step": 84
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.11858329804793687,
+      "learning_rate": 0.00010794501669717145,
+      "loss": 0.4868,
+      "step": 85
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.10984649953887334,
+      "learning_rate": 0.00010596151251698199,
+      "loss": 0.4598,
+      "step": 86
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.10927203986256682,
+      "learning_rate": 0.0001039756515096926,
+      "loss": 0.4693,
+      "step": 87
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.11205046531522328,
+      "learning_rate": 0.00010198821876650701,
+      "loss": 0.4921,
+      "step": 88
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.13232347270009215,
+      "learning_rate": 0.0001,
+      "loss": 0.4695,
+      "step": 89
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.12136881873560385,
+      "learning_rate": 9.801178123349298e-05,
+      "loss": 0.4859,
+      "step": 90
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.14347476421156694,
+      "learning_rate": 9.602434849030745e-05,
+      "loss": 0.4796,
+      "step": 91
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.13956845267055204,
+      "learning_rate": 9.403848748301802e-05,
+      "loss": 0.5339,
+      "step": 92
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.12814010903196785,
+      "learning_rate": 9.205498330282856e-05,
+      "loss": 0.5267,
+      "step": 93
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.12798850330908082,
+      "learning_rate": 9.007462010919386e-05,
+      "loss": 0.4604,
+      "step": 94
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.13673366056605873,
+      "learning_rate": 8.809818081980953e-05,
+      "loss": 0.49,
+      "step": 95
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.12607483394599764,
+      "learning_rate": 8.612644680109319e-05,
+      "loss": 0.4774,
+      "step": 96
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.1365629261848207,
+      "learning_rate": 8.416019755927851e-05,
+      "loss": 0.4827,
+      "step": 97
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.12122559291940836,
+      "learning_rate": 8.2200210432245e-05,
+      "loss": 0.5044,
+      "step": 98
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.11655390642565265,
+      "learning_rate": 8.024726028220474e-05,
+      "loss": 0.503,
+      "step": 99
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.12394574502796742,
+      "learning_rate": 7.83021191893682e-05,
+      "loss": 0.491,
+      "step": 100
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.14922841699852962,
+      "learning_rate": 7.636555614670953e-05,
+      "loss": 0.457,
+      "step": 101
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.11076924096187928,
+      "learning_rate": 7.443833675595255e-05,
+      "loss": 0.4603,
+      "step": 102
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.1263594611752413,
+      "learning_rate": 7.252122292489747e-05,
+      "loss": 0.4859,
+      "step": 103
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.11432575178505003,
+      "learning_rate": 7.061497256620793e-05,
+      "loss": 0.4627,
+      "step": 104
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.1078119391965793,
+      "learning_rate": 6.87203392977773e-05,
+      "loss": 0.4829,
+      "step": 105
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.12752089816514908,
+      "learning_rate": 6.683807214479323e-05,
+      "loss": 0.46,
+      "step": 106
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.11421623043902956,
+      "learning_rate": 6.496891524361757e-05,
+      "loss": 0.4429,
+      "step": 107
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.10432253193399477,
+      "learning_rate": 6.311360754759923e-05,
+      "loss": 0.402,
+      "step": 108
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.12155248673662734,
+      "learning_rate": 6.127288253493591e-05,
+      "loss": 0.5126,
+      "step": 109
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.14266947863559803,
+      "learning_rate": 5.9447467918700614e-05,
+      "loss": 0.4821,
+      "step": 110
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.14851250761112514,
+      "learning_rate": 5.763808535914723e-05,
+      "loss": 0.4891,
+      "step": 111
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.14264023747361737,
+      "learning_rate": 5.584545017840885e-05,
+      "loss": 0.5181,
+      "step": 112
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.12837168363458795,
+      "learning_rate": 5.407027107770219e-05,
+      "loss": 0.5599,
+      "step": 113
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.11874709251257598,
+      "learning_rate": 5.2313249857149414e-05,
+      "loss": 0.4536,
+      "step": 114
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.12010754957532713,
+      "learning_rate": 5.0575081138327715e-05,
+      "loss": 0.5004,
+      "step": 115
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.13464124440677885,
+      "learning_rate": 4.885645208965779e-05,
+      "loss": 0.4985,
+      "step": 116
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.13701854261941088,
+      "learning_rate": 4.715804215473809e-05,
+      "loss": 0.4709,
+      "step": 117
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.1335483738873249,
+      "learning_rate": 4.548052278373327e-05,
+      "loss": 0.4735,
+      "step": 118
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.13603172024059101,
+      "learning_rate": 4.382455716792291e-05,
+      "loss": 0.4721,
+      "step": 119
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.13843339239058639,
+      "learning_rate": 4.219079997751515e-05,
+      "loss": 0.4954,
+      "step": 120
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.15011169526780793,
+      "learning_rate": 4.0579897102828966e-05,
+      "loss": 0.4648,
+      "step": 121
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.13061595453081623,
+      "learning_rate": 3.899248539894757e-05,
+      "loss": 0.4801,
+      "step": 122
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.14067787924603412,
+      "learning_rate": 3.7429192433944014e-05,
+      "loss": 0.4805,
+      "step": 123
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.13420057703295998,
+      "learning_rate": 3.589063624077802e-05,
+      "loss": 0.4446,
+      "step": 124
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.14083737654873127,
+      "learning_rate": 3.4377425072962465e-05,
+      "loss": 0.46,
+      "step": 125
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.13231889777376862,
+      "learning_rate": 3.289015716409631e-05,
+      "loss": 0.4451,
+      "step": 126
+    },
+    {
+      "epoch": 0.75,
+      "eval_loss": 0.9651579260826111,
+      "eval_runtime": 155.5959,
+      "eval_samples_per_second": 1.6,
+      "eval_steps_per_second": 0.803,
+      "step": 126
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 168,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 42,
+  "total_flos": 2.5549258322041897e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-126/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cbfaffe231fe183758cc2353296944fef75710807fa8ea7663978d9e906d00f7
+size 6968

checkpoint-126/zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

checkpoint-168/README.md

@@ -0,0 +1,202 @@
+---
+library_name: peft
+base_model: ai21labs/Jamba-v0.1
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.10.0

checkpoint-168/adapter_config.json

@@ -0,0 +1,39 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "ai21labs/Jamba-v0.1",
+  "bias": "none",
+  "fan_in_fan_out": null,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "x_proj",
+    "o_proj",
+    "up_proj",
+    "v_proj",
+    "down_proj",
+    "dt_proj",
+    "k_proj",
+    "q_proj",
+    "in_proj",
+    "gate_proj",
+    "out_proj",
+    "router"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-168/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e965e9e667252ca3c021de51e081f6a3831a14d282903458d11e44f878a73597
+size 531611600

checkpoint-168/global_step168/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b66d8b3841b807a595a50e795a52d67de68eb94015081f06580ca93cd533b645
+size 797643792

checkpoint-168/global_step168/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b94caa551e67c01642f6b55c16bf11b71089c2f0fbd09d4215bd26e34ed7a530
+size 797644432

checkpoint-168/global_step168/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8130cd37e9434fdc6e150b642bbebca3eeebac77da37779b2183486786db25a5
+size 1345660121

checkpoint-168/latest

@@ -0,0 +1 @@
+global_step168

checkpoint-168/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:753bf22647e1e0f236cfa8df696cc134d9e1c0684712d00bf72503af01afa161
+size 14512

checkpoint-168/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c15347a58866f49cdc458b966fb754bbfbe4244548a2a26de0617bb3969cd63
+size 14512

checkpoint-168/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35ea2e1439244ec6e596a083a4657a3e879cc60c5a80ec3ad3f705a12a8fa66a
+size 1064

checkpoint-168/special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<|startoftext|>",
+    "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": "<|pad|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<|unk|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoint-168/tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02fd6530b8ede0eedd8e509fcab32da7b1dd04c8119f8498c787100f13112713
+size 1124742

checkpoint-168/tokenizer_config.json

@@ -0,0 +1,48 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<|pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<|startoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "<|unk|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<|startoftext|>",
+  "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "<|pad|>",
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<|unk|>",
+  "use_default_system_prompt": false
+}

checkpoint-168/trainer_state.json

@@ -0,0 +1,1237 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9985141158989599,
+  "eval_steps": 42,
+  "global_step": 168,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.30475762367257353,
+      "learning_rate": 2e-05,
+      "loss": 0.6274,
+      "step": 1
+    },
+    {
+      "epoch": 0.01,
+      "eval_loss": 1.0297596454620361,
+      "eval_runtime": 150.6014,
+      "eval_samples_per_second": 1.653,
+      "eval_steps_per_second": 0.83,
+      "step": 1
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.2798137150913395,
+      "learning_rate": 4e-05,
+      "loss": 0.6362,
+      "step": 2
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.3178684451319545,
+      "learning_rate": 6e-05,
+      "loss": 0.6299,
+      "step": 3
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.33651284916847835,
+      "learning_rate": 8e-05,
+      "loss": 0.6391,
+      "step": 4
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.23984915388648712,
+      "learning_rate": 0.0001,
+      "loss": 0.6071,
+      "step": 5
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.20514040410017348,
+      "learning_rate": 0.00012,
+      "loss": 0.5996,
+      "step": 6
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.1950018128286362,
+      "learning_rate": 0.00014,
+      "loss": 0.6298,
+      "step": 7
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.14246019947393238,
+      "learning_rate": 0.00016,
+      "loss": 0.5108,
+      "step": 8
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.15792014279750227,
+      "learning_rate": 0.00018,
+      "loss": 0.5529,
+      "step": 9
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.1517889177511264,
+      "learning_rate": 0.0002,
+      "loss": 0.5433,
+      "step": 10
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.1372121219437277,
+      "learning_rate": 0.00019998023297700658,
+      "loss": 0.5856,
+      "step": 11
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.16740807394942855,
+      "learning_rate": 0.00019992093972273018,
+      "loss": 0.5546,
+      "step": 12
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.13512320693394078,
+      "learning_rate": 0.00019982214367819328,
+      "loss": 0.6193,
+      "step": 13
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.16169796294070152,
+      "learning_rate": 0.0001996838839014696,
+      "loss": 0.5495,
+      "step": 14
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.16796913812281988,
+      "learning_rate": 0.00019950621505224273,
+      "loss": 0.5035,
+      "step": 15
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.1800514764162192,
+      "learning_rate": 0.00019928920737019733,
+      "loss": 0.5083,
+      "step": 16
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.170432124866908,
+      "learning_rate": 0.0001990329466472502,
+      "loss": 0.632,
+      "step": 17
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.19129325489749488,
+      "learning_rate": 0.00019873753419363336,
+      "loss": 0.4813,
+      "step": 18
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.1459357988760762,
+      "learning_rate": 0.00019840308679784207,
+      "loss": 0.4973,
+      "step": 19
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.192594730984382,
+      "learning_rate": 0.00019802973668046363,
+      "loss": 0.5291,
+      "step": 20
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.859025467969139,
+      "learning_rate": 0.0001976176314419051,
+      "loss": 0.5296,
+      "step": 21
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.13366297885670222,
+      "learning_rate": 0.000197166934004041,
+      "loss": 0.4819,
+      "step": 22
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.15698714419747645,
+      "learning_rate": 0.00019667782254580374,
+      "loss": 0.5409,
+      "step": 23
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.10995943735837355,
+      "learning_rate": 0.00019615049043274205,
+      "loss": 0.5108,
+      "step": 24
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.10796742192788925,
+      "learning_rate": 0.00019558514614057609,
+      "loss": 0.5215,
+      "step": 25
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.11641740089490231,
+      "learning_rate": 0.00019498201317277828,
+      "loss": 0.5012,
+      "step": 26
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.1120175962893241,
+      "learning_rate": 0.00019434132997221345,
+      "loss": 0.474,
+      "step": 27
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.1218171278782483,
+      "learning_rate": 0.0001936633498268728,
+      "loss": 0.5216,
+      "step": 28
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.11718521115928844,
+      "learning_rate": 0.0001929483407697387,
+      "loss": 0.4856,
+      "step": 29
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.12611471038571026,
+      "learning_rate": 0.00019219658547282067,
+      "loss": 0.4823,
+      "step": 30
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.11106871615269753,
+      "learning_rate": 0.00019140838113540346,
+      "loss": 0.4869,
+      "step": 31
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.1416503230360699,
+      "learning_rate": 0.00019058403936655233,
+      "loss": 0.5341,
+      "step": 32
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.10761396399791698,
+      "learning_rate": 0.00018972388606192125,
+      "loss": 0.4304,
+      "step": 33
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.10975376180434356,
+      "learning_rate": 0.0001888282612749132,
+      "loss": 0.4646,
+      "step": 34
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.12848879670359908,
+      "learning_rate": 0.00018789751908224338,
+      "loss": 0.4972,
+      "step": 35
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.11904721819683833,
+      "learning_rate": 0.00018693202744395827,
+      "loss": 0.505,
+      "step": 36
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.12249852034224981,
+      "learning_rate": 0.00018593216805796612,
+      "loss": 0.5396,
+      "step": 37
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.12453395046646995,
+      "learning_rate": 0.00018489833620913642,
+      "loss": 0.4917,
+      "step": 38
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.12585770374422164,
+      "learning_rate": 0.00018383094061302766,
+      "loss": 0.5079,
+      "step": 39
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.11095271476322731,
+      "learning_rate": 0.00018273040325430574,
+      "loss": 0.4812,
+      "step": 40
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.12968115101635422,
+      "learning_rate": 0.00018159715921991612,
+      "loss": 0.5106,
+      "step": 41
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.10933018515590627,
+      "learning_rate": 0.00018043165652707649,
+      "loss": 0.4403,
+      "step": 42
+    },
+    {
+      "epoch": 0.25,
+      "eval_loss": 0.9767947196960449,
+      "eval_runtime": 152.0343,
+      "eval_samples_per_second": 1.638,
+      "eval_steps_per_second": 0.822,
+      "step": 42
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.11525719626792096,
+      "learning_rate": 0.00017923435594615744,
+      "loss": 0.482,
+      "step": 43
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.12962154411778218,
+      "learning_rate": 0.00017800573081852122,
+      "loss": 0.5452,
+      "step": 44
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.12555700120045588,
+      "learning_rate": 0.0001767462668693908,
+      "loss": 0.5084,
+      "step": 45
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.11427565378293324,
+      "learning_rate": 0.00017545646201582303,
+      "loss": 0.5191,
+      "step": 46
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.10974901402857151,
+      "learning_rate": 0.00017413682616986185,
+      "loss": 0.4703,
+      "step": 47
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.11781465084480325,
+      "learning_rate": 0.00017278788103694943,
+      "loss": 0.4548,
+      "step": 48
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.10781807228559999,
+      "learning_rate": 0.000171410159909675,
+      "loss": 0.476,
+      "step": 49
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.12502639462035098,
+      "learning_rate": 0.00017000420745694254,
+      "loss": 0.5084,
+      "step": 50
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.10718920826593327,
+      "learning_rate": 0.00016857057950864132,
+      "loss": 0.5093,
+      "step": 51
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.10040549880547282,
+      "learning_rate": 0.0001671098428359037,
+      "loss": 0.4644,
+      "step": 52
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.11778478994740472,
+      "learning_rate": 0.00016562257492703757,
+      "loss": 0.4725,
+      "step": 53
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.1008386031049932,
+      "learning_rate": 0.000164109363759222,
+      "loss": 0.5121,
+      "step": 54
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.1170302528140235,
+      "learning_rate": 0.000162570807566056,
+      "loss": 0.4766,
+      "step": 55
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.1104526773884303,
+      "learning_rate": 0.00016100751460105243,
+      "loss": 0.4886,
+      "step": 56
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.10467920768691032,
+      "learning_rate": 0.00015942010289717105,
+      "loss": 0.4703,
+      "step": 57
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.11551406829220555,
+      "learning_rate": 0.00015780920002248484,
+      "loss": 0.4837,
+      "step": 58
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.11133818831887894,
+      "learning_rate": 0.0001561754428320771,
+      "loss": 0.5148,
+      "step": 59
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.11281448423273216,
+      "learning_rate": 0.00015451947721626676,
+      "loss": 0.4561,
+      "step": 60
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.13934126997471205,
+      "learning_rate": 0.00015284195784526195,
+      "loss": 0.5069,
+      "step": 61
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.11851655387640142,
+      "learning_rate": 0.00015114354791034225,
+      "loss": 0.5094,
+      "step": 62
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.12909148374566123,
+      "learning_rate": 0.0001494249188616723,
+      "loss": 0.581,
+      "step": 63
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.11070161341925377,
+      "learning_rate": 0.00014768675014285062,
+      "loss": 0.4585,
+      "step": 64
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.13308674882888374,
+      "learning_rate": 0.00014592972892229778,
+      "loss": 0.4974,
+      "step": 65
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.12124588853708144,
+      "learning_rate": 0.0001441545498215912,
+      "loss": 0.4463,
+      "step": 66
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.1183570515369953,
+      "learning_rate": 0.00014236191464085286,
+      "loss": 0.447,
+      "step": 67
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.13520024884417237,
+      "learning_rate": 0.00014055253208129938,
+      "loss": 0.5309,
+      "step": 68
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.12184981458813801,
+      "learning_rate": 0.00013872711746506413,
+      "loss": 0.4532,
+      "step": 69
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.12449299540645078,
+      "learning_rate": 0.00013688639245240078,
+      "loss": 0.5198,
+      "step": 70
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.1383134750490429,
+      "learning_rate": 0.00013503108475638244,
+      "loss": 0.5629,
+      "step": 71
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.246237001656926,
+      "learning_rate": 0.0001331619278552068,
+      "loss": 0.4869,
+      "step": 72
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.13337703940933632,
+      "learning_rate": 0.00013127966070222274,
+      "loss": 0.4792,
+      "step": 73
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.12428922033806454,
+      "learning_rate": 0.00012938502743379212,
+      "loss": 0.4825,
+      "step": 74
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.13290774912900208,
+      "learning_rate": 0.00012747877707510252,
+      "loss": 0.5138,
+      "step": 75
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.11185975046756892,
+      "learning_rate": 0.0001255616632440475,
+      "loss": 0.4815,
+      "step": 76
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.1130592868215497,
+      "learning_rate": 0.0001236344438532905,
+      "loss": 0.5046,
+      "step": 77
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.12882943465594857,
+      "learning_rate": 0.0001216978808106318,
+      "loss": 0.5091,
+      "step": 78
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.14837896297082676,
+      "learning_rate": 0.00011975273971779528,
+      "loss": 0.5158,
+      "step": 79
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.1265223309856292,
+      "learning_rate": 0.00011779978956775506,
+      "loss": 0.5068,
+      "step": 80
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.14042502330520407,
+      "learning_rate": 0.0001158398024407215,
+      "loss": 0.5061,
+      "step": 81
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.1261526695491767,
+      "learning_rate": 0.00011387355319890685,
+      "loss": 0.4691,
+      "step": 82
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.12007305451001854,
+      "learning_rate": 0.00011190181918019049,
+      "loss": 0.4753,
+      "step": 83
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.12809956897166885,
+      "learning_rate": 0.00010992537989080618,
+      "loss": 0.4417,
+      "step": 84
+    },
+    {
+      "epoch": 0.5,
+      "eval_loss": 0.9675251841545105,
+      "eval_runtime": 152.4793,
+      "eval_samples_per_second": 1.633,
+      "eval_steps_per_second": 0.82,
+      "step": 84
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.11858329804793687,
+      "learning_rate": 0.00010794501669717145,
+      "loss": 0.4868,
+      "step": 85
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.10984649953887334,
+      "learning_rate": 0.00010596151251698199,
+      "loss": 0.4598,
+      "step": 86
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.10927203986256682,
+      "learning_rate": 0.0001039756515096926,
+      "loss": 0.4693,
+      "step": 87
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.11205046531522328,
+      "learning_rate": 0.00010198821876650701,
+      "loss": 0.4921,
+      "step": 88
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.13232347270009215,
+      "learning_rate": 0.0001,
+      "loss": 0.4695,
+      "step": 89
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.12136881873560385,
+      "learning_rate": 9.801178123349298e-05,
+      "loss": 0.4859,
+      "step": 90
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.14347476421156694,
+      "learning_rate": 9.602434849030745e-05,
+      "loss": 0.4796,
+      "step": 91
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.13956845267055204,
+      "learning_rate": 9.403848748301802e-05,
+      "loss": 0.5339,
+      "step": 92
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.12814010903196785,
+      "learning_rate": 9.205498330282856e-05,
+      "loss": 0.5267,
+      "step": 93
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.12798850330908082,
+      "learning_rate": 9.007462010919386e-05,
+      "loss": 0.4604,
+      "step": 94
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.13673366056605873,
+      "learning_rate": 8.809818081980953e-05,
+      "loss": 0.49,
+      "step": 95
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.12607483394599764,
+      "learning_rate": 8.612644680109319e-05,
+      "loss": 0.4774,
+      "step": 96
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.1365629261848207,
+      "learning_rate": 8.416019755927851e-05,
+      "loss": 0.4827,
+      "step": 97
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.12122559291940836,
+      "learning_rate": 8.2200210432245e-05,
+      "loss": 0.5044,
+      "step": 98
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.11655390642565265,
+      "learning_rate": 8.024726028220474e-05,
+      "loss": 0.503,
+      "step": 99
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.12394574502796742,
+      "learning_rate": 7.83021191893682e-05,
+      "loss": 0.491,
+      "step": 100
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.14922841699852962,
+      "learning_rate": 7.636555614670953e-05,
+      "loss": 0.457,
+      "step": 101
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.11076924096187928,
+      "learning_rate": 7.443833675595255e-05,
+      "loss": 0.4603,
+      "step": 102
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.1263594611752413,
+      "learning_rate": 7.252122292489747e-05,
+      "loss": 0.4859,
+      "step": 103
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.11432575178505003,
+      "learning_rate": 7.061497256620793e-05,
+      "loss": 0.4627,
+      "step": 104
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.1078119391965793,
+      "learning_rate": 6.87203392977773e-05,
+      "loss": 0.4829,
+      "step": 105
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.12752089816514908,
+      "learning_rate": 6.683807214479323e-05,
+      "loss": 0.46,
+      "step": 106
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.11421623043902956,
+      "learning_rate": 6.496891524361757e-05,
+      "loss": 0.4429,
+      "step": 107
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.10432253193399477,
+      "learning_rate": 6.311360754759923e-05,
+      "loss": 0.402,
+      "step": 108
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.12155248673662734,
+      "learning_rate": 6.127288253493591e-05,
+      "loss": 0.5126,
+      "step": 109
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.14266947863559803,
+      "learning_rate": 5.9447467918700614e-05,
+      "loss": 0.4821,
+      "step": 110
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.14851250761112514,
+      "learning_rate": 5.763808535914723e-05,
+      "loss": 0.4891,
+      "step": 111
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.14264023747361737,
+      "learning_rate": 5.584545017840885e-05,
+      "loss": 0.5181,
+      "step": 112
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.12837168363458795,
+      "learning_rate": 5.407027107770219e-05,
+      "loss": 0.5599,
+      "step": 113
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.11874709251257598,
+      "learning_rate": 5.2313249857149414e-05,
+      "loss": 0.4536,
+      "step": 114
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.12010754957532713,
+      "learning_rate": 5.0575081138327715e-05,
+      "loss": 0.5004,
+      "step": 115
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.13464124440677885,
+      "learning_rate": 4.885645208965779e-05,
+      "loss": 0.4985,
+      "step": 116
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.13701854261941088,
+      "learning_rate": 4.715804215473809e-05,
+      "loss": 0.4709,
+      "step": 117
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.1335483738873249,
+      "learning_rate": 4.548052278373327e-05,
+      "loss": 0.4735,
+      "step": 118
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.13603172024059101,
+      "learning_rate": 4.382455716792291e-05,
+      "loss": 0.4721,
+      "step": 119
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.13843339239058639,
+      "learning_rate": 4.219079997751515e-05,
+      "loss": 0.4954,
+      "step": 120
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.15011169526780793,
+      "learning_rate": 4.0579897102828966e-05,
+      "loss": 0.4648,
+      "step": 121
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.13061595453081623,
+      "learning_rate": 3.899248539894757e-05,
+      "loss": 0.4801,
+      "step": 122
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.14067787924603412,
+      "learning_rate": 3.7429192433944014e-05,
+      "loss": 0.4805,
+      "step": 123
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.13420057703295998,
+      "learning_rate": 3.589063624077802e-05,
+      "loss": 0.4446,
+      "step": 124
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.14083737654873127,
+      "learning_rate": 3.4377425072962465e-05,
+      "loss": 0.46,
+      "step": 125
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.13231889777376862,
+      "learning_rate": 3.289015716409631e-05,
+      "loss": 0.4451,
+      "step": 126
+    },
+    {
+      "epoch": 0.75,
+      "eval_loss": 0.9651579260826111,
+      "eval_runtime": 155.5959,
+      "eval_samples_per_second": 1.6,
+      "eval_steps_per_second": 0.803,
+      "step": 126
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.1119782513388846,
+      "learning_rate": 3.14294204913587e-05,
+      "loss": 0.4728,
+      "step": 127
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.1258610262132018,
+      "learning_rate": 2.9995792543057478e-05,
+      "loss": 0.4793,
+      "step": 128
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.1332956040603169,
+      "learning_rate": 2.8589840090325027e-05,
+      "loss": 0.4919,
+      "step": 129
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.13669237836174272,
+      "learning_rate": 2.7212118963050592e-05,
+      "loss": 0.4863,
+      "step": 130
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.13718073990096244,
+      "learning_rate": 2.586317383013821e-05,
+      "loss": 0.4652,
+      "step": 131
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.1328715725965393,
+      "learning_rate": 2.4543537984176978e-05,
+      "loss": 0.4464,
+      "step": 132
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.12474829320490133,
+      "learning_rate": 2.325373313060919e-05,
+      "loss": 0.501,
+      "step": 133
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.13832785688421897,
+      "learning_rate": 2.19942691814788e-05,
+      "loss": 0.5113,
+      "step": 134
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.13446374612957607,
+      "learning_rate": 2.076564405384258e-05,
+      "loss": 0.452,
+      "step": 135
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.15087962227561943,
+      "learning_rate": 1.9568343472923524e-05,
+      "loss": 0.5114,
+      "step": 136
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.13471199433579678,
+      "learning_rate": 1.840284078008393e-05,
+      "loss": 0.5074,
+      "step": 137
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.14471602871304384,
+      "learning_rate": 1.7269596745694295e-05,
+      "loss": 0.4812,
+      "step": 138
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.12791095480012135,
+      "learning_rate": 1.616905938697234e-05,
+      "loss": 0.439,
+      "step": 139
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.13395748535894433,
+      "learning_rate": 1.5101663790863596e-05,
+      "loss": 0.4942,
+      "step": 140
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.1216157843375959,
+      "learning_rate": 1.4067831942033904e-05,
+      "loss": 0.4904,
+      "step": 141
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.12513029490165878,
+      "learning_rate": 1.3067972556041752e-05,
+      "loss": 0.4583,
+      "step": 142
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.11835922329329535,
+      "learning_rate": 1.210248091775663e-05,
+      "loss": 0.5281,
+      "step": 143
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.14513302866752326,
+      "learning_rate": 1.1171738725086833e-05,
+      "loss": 0.4432,
+      "step": 144
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.11934584390756656,
+      "learning_rate": 1.0276113938078769e-05,
+      "loss": 0.4664,
+      "step": 145
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.11746932949614314,
+      "learning_rate": 9.415960633447674e-06,
+      "loss": 0.4452,
+      "step": 146
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.1183312470782195,
+      "learning_rate": 8.59161886459654e-06,
+      "loss": 0.4858,
+      "step": 147
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.1226342260426643,
+      "learning_rate": 7.803414527179343e-06,
+      "loss": 0.4609,
+      "step": 148
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.12849903086517628,
+      "learning_rate": 7.051659230261298e-06,
+      "loss": 0.4992,
+      "step": 149
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.12558798780888547,
+      "learning_rate": 6.336650173127223e-06,
+      "loss": 0.5055,
+      "step": 150
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.11991457404362381,
+      "learning_rate": 5.658670027786561e-06,
+      "loss": 0.4711,
+      "step": 151
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.14486189343828018,
+      "learning_rate": 5.017986827221733e-06,
+      "loss": 0.4984,
+      "step": 152
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.14292272311516058,
+      "learning_rate": 4.4148538594239174e-06,
+      "loss": 0.518,
+      "step": 153
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.12669351883956026,
+      "learning_rate": 3.849509567257959e-06,
+      "loss": 0.509,
+      "step": 154
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.14745027617256945,
+      "learning_rate": 3.3221774541962845e-06,
+      "loss": 0.4655,
+      "step": 155
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.13904171638337365,
+      "learning_rate": 2.8330659959589946e-06,
+      "loss": 0.445,
+      "step": 156
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.12971202836694892,
+      "learning_rate": 2.3823685580949273e-06,
+      "loss": 0.4534,
+      "step": 157
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.13437185949159883,
+      "learning_rate": 1.9702633195363917e-06,
+      "loss": 0.4785,
+      "step": 158
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.12319276879857373,
+      "learning_rate": 1.5969132021579347e-06,
+      "loss": 0.5011,
+      "step": 159
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.12918661435233847,
+      "learning_rate": 1.2624658063666639e-06,
+      "loss": 0.5236,
+      "step": 160
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.12874241011432666,
+      "learning_rate": 9.670533527498137e-07,
+      "loss": 0.4439,
+      "step": 161
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.1315361924610259,
+      "learning_rate": 7.10792629802659e-07,
+      "loss": 0.4929,
+      "step": 162
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.1336404295007477,
+      "learning_rate": 4.937849477572587e-07,
+      "loss": 0.433,
+      "step": 163
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.11865993758025514,
+      "learning_rate": 3.161160985304168e-07,
+      "loss": 0.4266,
+      "step": 164
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.12559739797694514,
+      "learning_rate": 1.7785632180670198e-07,
+      "loss": 0.436,
+      "step": 165
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.12409278477769135,
+      "learning_rate": 7.906027726981568e-08,
+      "loss": 0.5017,
+      "step": 166
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.1398337808177678,
+      "learning_rate": 1.976702299344435e-08,
+      "loss": 0.5112,
+      "step": 167
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.11434132222175684,
+      "learning_rate": 0.0,
+      "loss": 0.4616,
+      "step": 168
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 0.9651336073875427,
+      "eval_runtime": 154.7455,
+      "eval_samples_per_second": 1.609,
+      "eval_steps_per_second": 0.808,
+      "step": 168
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 168,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 42,
+  "total_flos": 3.406567776272253e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-168/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cbfaffe231fe183758cc2353296944fef75710807fa8ea7663978d9e906d00f7
+size 6968

checkpoint-168/zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

config.json

@@ -40,15 +40,13 @@
   "quantization_config": {
     "_load_in_4bit": true,
     "_load_in_8bit": false,
-    "bnb_4bit_compute_dtype": "float32",
-    "bnb_4bit_quant_storage": "uint8",
-    "bnb_4bit_quant_type": "fp4",
-    "bnb_4bit_use_double_quant": false,
+    "bnb_4bit_compute_dtype": "bfloat16",
+    "bnb_4bit_quant_storage": "bfloat16",
+    "bnb_4bit_quant_type": "nf4",
+    "bnb_4bit_use_double_quant": true,
     "llm_int8_enable_fp32_cpu_offload": false,
     "llm_int8_has_fp16_weight": false,
-    "llm_int8_skip_modules": [
-      "mamba"
-    ],
+    "llm_int8_skip_modules": null,
     "llm_int8_threshold": 6.0,
     "load_in_4bit": true,
     "load_in_8bit": false,
@@ -58,9 +56,9 @@
   "router_aux_loss_coef": 0.001,
   "sliding_window": null,
   "tie_word_embeddings": false,
-  "torch_dtype": "float16",
+  "torch_dtype": "bfloat16",
   "transformers_version": "4.40.0.dev0",
-  "use_cache": true,
+  "use_cache": false,
   "use_mamba_kernels": true,
   "vocab_size": 65536
 }

tokenizer_config.json

@@ -36,6 +36,7 @@
     }
   },
   "bos_token": "<|startoftext|>",
+  "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
   "clean_up_tokenization_spaces": false,
   "eos_token": "<|endoftext|>",
   "model_max_length": 1000000000000000019884624838656,