Training in progress, step 550, checkpoint

checkpoint-550/README.md

@@ -0,0 +1,202 @@
+---
+base_model: microsoft/Phi-3-mini-4k-instruct
+library_name: peft
+---
+
+# 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 0.14.0

checkpoint-550/adapter_config.json

@@ -0,0 +1,34 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "microsoft/Phi-3-mini-4k-instruct",
+  "bias": "none",
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_up_proj",
+    "o_proj",
+    "qkv_proj",
+    "down_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-550/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb73060bfe624dcaa74a58d902285026cf8ad811290a1da97b54741c05d24f99
+size 25200088

checkpoint-550/global_step550/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4149ee00ed4a38138385a86cb6d150e219173ff7b9dde184030be3354867ef02
+size 18881328

checkpoint-550/global_step550/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0afe1aeac6cc3dd35706f2490bcff5753a771b3bdf37251adba10edbe8beacd0
+size 18881328

checkpoint-550/global_step550/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48d8f090171b0450c102e8ce1edd1cfb44db16063d632a81870ff1694e3c831f
+size 18881328

checkpoint-550/global_step550/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e4abb843128472fdc81f491e33f74e5e33e5edc754986c682dfeafe31e65ca46
+size 18881392

checkpoint-550/global_step550/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd1a38e257f958b50f2a8750beaf83ae4e825d82f1b3d28ae84d691672c35574
+size 18881392

checkpoint-550/global_step550/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99245b485718c4bdcefc1e489869e28a6a86b614f432cddf35db2cddcc47c6a8
+size 18881392

checkpoint-550/global_step550/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1cf52b2f90e5a7e5aa5476abefa93def254eca84dbe34f054d2a16abc5d5be0b
+size 18881392

checkpoint-550/global_step550/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6de41e53c0ddbe4b64e2cb2d2986114423c30440a73627431cf5b4a2e37014a9
+size 18881392

checkpoint-550/global_step550/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e3a121bf008790217a00e591cd5b8252e589ff8c48ac9c7aeed9aff032296a2
+size 25379244

checkpoint-550/latest

@@ -0,0 +1 @@
+global_step550

checkpoint-550/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e5cd4ad571350abe2eb98424dd5c5dd650f79de5be8de2b9ff4da9d030d723b
+size 15984

checkpoint-550/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dcff98006be86afc3f75b37d6113fdf5b62db51c94b6f68b33f555f4ac346822
+size 15984

checkpoint-550/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f220fd74a6757e167d014f721e96b7e5710e8f5c97f48c9fe6d72e19ebbbd65c
+size 15984

checkpoint-550/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21565575b5db0aa139865ffb0d9df6ceb55078dc7b218f601419cc3d7b873134
+size 15984

checkpoint-550/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:487a03a3b6c36091572b8fbb74add1eb3c753efe5ab0eee791c8d03f495e5c98
+size 15984

checkpoint-550/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98f8c6e22cfd0b3668705becc42fb2c443ef5e4cfe38d4ba5e3dfdc565094143
+size 15984

checkpoint-550/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:379eebc7ccebea3c24281c6604242d09589a64d4774ea37b6d5cf6e7bbece645
+size 15984

checkpoint-550/rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3668b553f323a1aa5806c5d8feff7c926f6116dc2b7f961e9746634c8e825c0
+size 15984

checkpoint-550/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0a8efee4112d2dd8949a8275b8c557402eda5213e3f06fa3f59f89e1c62ee85
+size 1064

checkpoint-550/special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoint-550/tokenizer_config.json

@@ -0,0 +1,133 @@
+{
+  "add_bos_token": false,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": false
+    },
+    "32000": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32001": {
+      "content": "<|assistant|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32002": {
+      "content": "<|placeholder1|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32003": {
+      "content": "<|placeholder2|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32004": {
+      "content": "<|placeholder3|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32005": {
+      "content": "<|placeholder4|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32006": {
+      "content": "<|system|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32007": {
+      "content": "<|end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32008": {
+      "content": "<|placeholder5|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32009": {
+      "content": "<|placeholder6|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32010": {
+      "content": "<|user|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "chat_template": "{% set system_message = 'You are a helpful AI assistant.' %}{% if messages[0]['role'] == 'system' %}{% set system_message = messages[0]['content'] %}{% endif %}{% if system_message is defined %}{{ '<s>' + '<|system|>\n' + system_message + '<|end|>\n' }}{% endif %}{% for message in messages %}{% set content = message['content'] %}{% if message['role'] == 'user' %}{{ '<|user|>\n' + content + '<|end|>\n<|assistant|>\n' }}{% elif message['role'] == 'assistant' %}{{ content + '<|end|>' + '\n' }}{% endif %}{% endfor %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|end|>",
+  "extra_special_tokens": {},
+  "legacy": false,
+  "model_max_length": 4096,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "split_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

checkpoint-550/trainer_state.json

@@ -0,0 +1,1034 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.6984126984126984,
+  "eval_steps": 50,
+  "global_step": 550,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.012698412698412698,
+      "grad_norm": 0.04716634005308151,
+      "learning_rate": 4.999451708687114e-06,
+      "logits/chosen": 14.883691787719727,
+      "logits/rejected": 15.016583442687988,
+      "logps/chosen": -0.29512909054756165,
+      "logps/rejected": -0.30033987760543823,
+      "loss": 1.0007,
+      "rewards/accuracies": 0.36250001192092896,
+      "rewards/chosen": -0.4426936209201813,
+      "rewards/margins": 0.007816222496330738,
+      "rewards/rejected": -0.45050984621047974,
+      "step": 10
+    },
+    {
+      "epoch": 0.025396825396825397,
+      "grad_norm": 0.04392225295305252,
+      "learning_rate": 4.997807075247147e-06,
+      "logits/chosen": 14.705224990844727,
+      "logits/rejected": 14.737253189086914,
+      "logps/chosen": -0.3201700747013092,
+      "logps/rejected": -0.280050128698349,
+      "loss": 0.9984,
+      "rewards/accuracies": 0.26249998807907104,
+      "rewards/chosen": -0.4802550673484802,
+      "rewards/margins": -0.060179851949214935,
+      "rewards/rejected": -0.4200752377510071,
+      "step": 20
+    },
+    {
+      "epoch": 0.0380952380952381,
+      "grad_norm": 0.04989476501941681,
+      "learning_rate": 4.9950668210706795e-06,
+      "logits/chosen": 15.147977828979492,
+      "logits/rejected": 15.080279350280762,
+      "logps/chosen": -0.3157380223274231,
+      "logps/rejected": -0.2997627556324005,
+      "loss": 0.9991,
+      "rewards/accuracies": 0.2874999940395355,
+      "rewards/chosen": -0.47360706329345703,
+      "rewards/margins": -0.023962898179888725,
+      "rewards/rejected": -0.44964417815208435,
+      "step": 30
+    },
+    {
+      "epoch": 0.050793650793650794,
+      "grad_norm": 0.054729390889406204,
+      "learning_rate": 4.9912321481237616e-06,
+      "logits/chosen": 15.301165580749512,
+      "logits/rejected": 15.276555061340332,
+      "logps/chosen": -0.30470195412635803,
+      "logps/rejected": -0.29767152667045593,
+      "loss": 0.9849,
+      "rewards/accuracies": 0.36250001192092896,
+      "rewards/chosen": -0.45705294609069824,
+      "rewards/margins": -0.010545584373176098,
+      "rewards/rejected": -0.44650736451148987,
+      "step": 40
+    },
+    {
+      "epoch": 0.06349206349206349,
+      "grad_norm": 0.06035450100898743,
+      "learning_rate": 4.986304738420684e-06,
+      "logits/chosen": 14.600168228149414,
+      "logits/rejected": 14.7944917678833,
+      "logps/chosen": -0.32278841733932495,
+      "logps/rejected": -0.3014402687549591,
+      "loss": 0.9991,
+      "rewards/accuracies": 0.3499999940395355,
+      "rewards/chosen": -0.48418259620666504,
+      "rewards/margins": -0.03202226758003235,
+      "rewards/rejected": -0.4521603584289551,
+      "step": 50
+    },
+    {
+      "epoch": 0.06349206349206349,
+      "eval_logits/chosen": 15.261337280273438,
+      "eval_logits/rejected": 15.51547908782959,
+      "eval_logps/chosen": -0.3022651970386505,
+      "eval_logps/rejected": -0.3061661124229431,
+      "eval_loss": 0.9846106171607971,
+      "eval_rewards/accuracies": 0.40625,
+      "eval_rewards/chosen": -0.4533977508544922,
+      "eval_rewards/margins": 0.005851435009390116,
+      "eval_rewards/rejected": -0.45924919843673706,
+      "eval_runtime": 18.4033,
+      "eval_samples_per_second": 27.712,
+      "eval_steps_per_second": 3.478,
+      "step": 50
+    },
+    {
+      "epoch": 0.0761904761904762,
+      "grad_norm": 0.052623528987169266,
+      "learning_rate": 4.980286753286196e-06,
+      "logits/chosen": 15.2726469039917,
+      "logits/rejected": 15.245884895324707,
+      "logps/chosen": -0.31394433975219727,
+      "logps/rejected": -0.2805258333683014,
+      "loss": 1.0063,
+      "rewards/accuracies": 0.22499999403953552,
+      "rewards/chosen": -0.4709165096282959,
+      "rewards/margins": -0.05012776702642441,
+      "rewards/rejected": -0.4207887053489685,
+      "step": 60
+    },
+    {
+      "epoch": 0.08888888888888889,
+      "grad_norm": 0.3512271046638489,
+      "learning_rate": 4.973180832407471e-06,
+      "logits/chosen": 15.302217483520508,
+      "logits/rejected": 15.387911796569824,
+      "logps/chosen": -0.30377697944641113,
+      "logps/rejected": -0.29276782274246216,
+      "loss": 0.9888,
+      "rewards/accuracies": 0.3499999940395355,
+      "rewards/chosen": -0.4556654393672943,
+      "rewards/margins": -0.016513748094439507,
+      "rewards/rejected": -0.43915170431137085,
+      "step": 70
+    },
+    {
+      "epoch": 0.10158730158730159,
+      "grad_norm": 0.056646134704351425,
+      "learning_rate": 4.964990092676263e-06,
+      "logits/chosen": 15.843823432922363,
+      "logits/rejected": 15.895855903625488,
+      "logps/chosen": -0.3091468811035156,
+      "logps/rejected": -0.27621737122535706,
+      "loss": 0.9926,
+      "rewards/accuracies": 0.3375000059604645,
+      "rewards/chosen": -0.46372032165527344,
+      "rewards/margins": -0.04939427971839905,
+      "rewards/rejected": -0.4143260419368744,
+      "step": 80
+    },
+    {
+      "epoch": 0.11428571428571428,
+      "grad_norm": 0.062348950654268265,
+      "learning_rate": 4.9557181268217225e-06,
+      "logits/chosen": 15.522564888000488,
+      "logits/rejected": 15.576242446899414,
+      "logps/chosen": -0.3092329502105713,
+      "logps/rejected": -0.27670228481292725,
+      "loss": 0.9812,
+      "rewards/accuracies": 0.22499999403953552,
+      "rewards/chosen": -0.4638494551181793,
+      "rewards/margins": -0.048796020448207855,
+      "rewards/rejected": -0.41505345702171326,
+      "step": 90
+    },
+    {
+      "epoch": 0.12698412698412698,
+      "grad_norm": 0.06779928505420685,
+      "learning_rate": 4.9453690018345144e-06,
+      "logits/chosen": 15.999185562133789,
+      "logits/rejected": 15.802484512329102,
+      "logps/chosen": -0.3173079490661621,
+      "logps/rejected": -0.283602774143219,
+      "loss": 0.981,
+      "rewards/accuracies": 0.2750000059604645,
+      "rewards/chosen": -0.4759618639945984,
+      "rewards/margins": -0.050557754933834076,
+      "rewards/rejected": -0.4254041314125061,
+      "step": 100
+    },
+    {
+      "epoch": 0.12698412698412698,
+      "eval_logits/chosen": 15.745854377746582,
+      "eval_logits/rejected": 16.0076961517334,
+      "eval_logps/chosen": -0.29866182804107666,
+      "eval_logps/rejected": -0.3121251165866852,
+      "eval_loss": 0.9764130115509033,
+      "eval_rewards/accuracies": 0.453125,
+      "eval_rewards/chosen": -0.4479926824569702,
+      "eval_rewards/margins": 0.02019493840634823,
+      "eval_rewards/rejected": -0.4681876599788666,
+      "eval_runtime": 18.3383,
+      "eval_samples_per_second": 27.811,
+      "eval_steps_per_second": 3.49,
+      "step": 100
+    },
+    {
+      "epoch": 0.13968253968253969,
+      "grad_norm": 0.07346926629543304,
+      "learning_rate": 4.933947257182901e-06,
+      "logits/chosen": 15.723541259765625,
+      "logits/rejected": 15.764880180358887,
+      "logps/chosen": -0.3212783634662628,
+      "logps/rejected": -0.3159862160682678,
+      "loss": 0.9859,
+      "rewards/accuracies": 0.4124999940395355,
+      "rewards/chosen": -0.4819175601005554,
+      "rewards/margins": -0.00793826300650835,
+      "rewards/rejected": -0.47397929430007935,
+      "step": 110
+    },
+    {
+      "epoch": 0.1523809523809524,
+      "grad_norm": 0.06748441606760025,
+      "learning_rate": 4.921457902821578e-06,
+      "logits/chosen": 15.817390441894531,
+      "logits/rejected": 16.023067474365234,
+      "logps/chosen": -0.29811763763427734,
+      "logps/rejected": -0.30654552578926086,
+      "loss": 0.9827,
+      "rewards/accuracies": 0.375,
+      "rewards/chosen": -0.447176456451416,
+      "rewards/margins": 0.01264181174337864,
+      "rewards/rejected": -0.4598182737827301,
+      "step": 120
+    },
+    {
+      "epoch": 0.16507936507936508,
+      "grad_norm": 0.08042758703231812,
+      "learning_rate": 4.907906416994146e-06,
+      "logits/chosen": 15.45033073425293,
+      "logits/rejected": 15.865328788757324,
+      "logps/chosen": -0.27259278297424316,
+      "logps/rejected": -0.31338179111480713,
+      "loss": 0.9743,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": -0.40888920426368713,
+      "rewards/margins": 0.06118353083729744,
+      "rewards/rejected": -0.4700726866722107,
+      "step": 130
+    },
+    {
+      "epoch": 0.17777777777777778,
+      "grad_norm": 0.15669430792331696,
+      "learning_rate": 4.893298743830168e-06,
+      "logits/chosen": 16.17765235900879,
+      "logits/rejected": 15.951130867004395,
+      "logps/chosen": -0.3015133738517761,
+      "logps/rejected": -0.2970424294471741,
+      "loss": 0.9824,
+      "rewards/accuracies": 0.36250001192092896,
+      "rewards/chosen": -0.4522700905799866,
+      "rewards/margins": -0.006706444080919027,
+      "rewards/rejected": -0.44556355476379395,
+      "step": 140
+    },
+    {
+      "epoch": 0.19047619047619047,
+      "grad_norm": 0.07115967571735382,
+      "learning_rate": 4.8776412907378845e-06,
+      "logits/chosen": 15.552728652954102,
+      "logits/rejected": 15.788042068481445,
+      "logps/chosen": -0.28624147176742554,
+      "logps/rejected": -0.3222900331020355,
+      "loss": 0.9727,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.4293622076511383,
+      "rewards/margins": 0.05407290905714035,
+      "rewards/rejected": -0.48343515396118164,
+      "step": 150
+    },
+    {
+      "epoch": 0.19047619047619047,
+      "eval_logits/chosen": 15.72143268585205,
+      "eval_logits/rejected": 15.974818229675293,
+      "eval_logps/chosen": -0.29440543055534363,
+      "eval_logps/rejected": -0.33279579877853394,
+      "eval_loss": 0.9540849328041077,
+      "eval_rewards/accuracies": 0.546875,
+      "eval_rewards/chosen": -0.44160816073417664,
+      "eval_rewards/margins": 0.05758553743362427,
+      "eval_rewards/rejected": -0.4991936683654785,
+      "eval_runtime": 18.335,
+      "eval_samples_per_second": 27.816,
+      "eval_steps_per_second": 3.491,
+      "step": 150
+    },
+    {
+      "epoch": 0.20317460317460317,
+      "grad_norm": 0.1144075095653534,
+      "learning_rate": 4.860940925593703e-06,
+      "logits/chosen": 15.63622760772705,
+      "logits/rejected": 15.758941650390625,
+      "logps/chosen": -0.310846745967865,
+      "logps/rejected": -0.35185256600379944,
+      "loss": 0.9517,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.4662701487541199,
+      "rewards/margins": 0.061508744955062866,
+      "rewards/rejected": -0.5277789235115051,
+      "step": 160
+    },
+    {
+      "epoch": 0.21587301587301588,
+      "grad_norm": 0.1173442155122757,
+      "learning_rate": 4.84320497372973e-06,
+      "logits/chosen": 15.8660306930542,
+      "logits/rejected": 15.83642292022705,
+      "logps/chosen": -0.279310017824173,
+      "logps/rejected": -0.3301998972892761,
+      "loss": 0.9439,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.41896501183509827,
+      "rewards/margins": 0.07633484899997711,
+      "rewards/rejected": -0.49529990553855896,
+      "step": 170
+    },
+    {
+      "epoch": 0.22857142857142856,
+      "grad_norm": 0.204214945435524,
+      "learning_rate": 4.824441214720629e-06,
+      "logits/chosen": 15.557527542114258,
+      "logits/rejected": 15.321504592895508,
+      "logps/chosen": -0.3029389977455139,
+      "logps/rejected": -0.3163699507713318,
+      "loss": 0.9369,
+      "rewards/accuracies": 0.38749998807907104,
+      "rewards/chosen": -0.4544084668159485,
+      "rewards/margins": 0.020146463066339493,
+      "rewards/rejected": -0.47455495595932007,
+      "step": 180
+    },
+    {
+      "epoch": 0.24126984126984127,
+      "grad_norm": 0.14012843370437622,
+      "learning_rate": 4.804657878971252e-06,
+      "logits/chosen": 15.567469596862793,
+      "logits/rejected": 15.717155456542969,
+      "logps/chosen": -0.3400501310825348,
+      "logps/rejected": -0.3475819230079651,
+      "loss": 0.9443,
+      "rewards/accuracies": 0.38749998807907104,
+      "rewards/chosen": -0.5100752115249634,
+      "rewards/margins": 0.011297682300209999,
+      "rewards/rejected": -0.5213728547096252,
+      "step": 190
+    },
+    {
+      "epoch": 0.25396825396825395,
+      "grad_norm": 0.12908300757408142,
+      "learning_rate": 4.783863644106502e-06,
+      "logits/chosen": 14.985162734985352,
+      "logits/rejected": 15.223039627075195,
+      "logps/chosen": -0.2876646816730499,
+      "logps/rejected": -0.3777172267436981,
+      "loss": 0.9034,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": -0.4314970076084137,
+      "rewards/margins": 0.13507887721061707,
+      "rewards/rejected": -0.5665758848190308,
+      "step": 200
+    },
+    {
+      "epoch": 0.25396825396825395,
+      "eval_logits/chosen": 15.242753028869629,
+      "eval_logits/rejected": 15.523069381713867,
+      "eval_logps/chosen": -0.2961229085922241,
+      "eval_logps/rejected": -0.3966684341430664,
+      "eval_loss": 0.9075753092765808,
+      "eval_rewards/accuracies": 0.59375,
+      "eval_rewards/chosen": -0.4441843330860138,
+      "eval_rewards/margins": 0.1508183479309082,
+      "eval_rewards/rejected": -0.5950026512145996,
+      "eval_runtime": 18.341,
+      "eval_samples_per_second": 27.807,
+      "eval_steps_per_second": 3.489,
+      "step": 200
+    },
+    {
+      "epoch": 0.26666666666666666,
+      "grad_norm": 0.13152211904525757,
+      "learning_rate": 4.762067631165049e-06,
+      "logits/chosen": 15.283185958862305,
+      "logits/rejected": 15.545951843261719,
+      "logps/chosen": -0.30045825242996216,
+      "logps/rejected": -0.4519944190979004,
+      "loss": 0.9085,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.45068734884262085,
+      "rewards/margins": 0.22730426490306854,
+      "rewards/rejected": -0.6779916286468506,
+      "step": 210
+    },
+    {
+      "epoch": 0.27936507936507937,
+      "grad_norm": 0.14594660699367523,
+      "learning_rate": 4.7392794005985324e-06,
+      "logits/chosen": 15.53101921081543,
+      "logits/rejected": 15.322232246398926,
+      "logps/chosen": -0.3336530327796936,
+      "logps/rejected": -0.36337023973464966,
+      "loss": 0.9188,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.5004795789718628,
+      "rewards/margins": 0.04457578808069229,
+      "rewards/rejected": -0.5450553894042969,
+      "step": 220
+    },
+    {
+      "epoch": 0.2920634920634921,
+      "grad_norm": 0.14854153990745544,
+      "learning_rate": 4.715508948078037e-06,
+      "logits/chosen": 14.157003402709961,
+      "logits/rejected": 14.34089183807373,
+      "logps/chosen": -0.28324443101882935,
+      "logps/rejected": -0.41318759322166443,
+      "loss": 0.9015,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.4248666763305664,
+      "rewards/margins": 0.19491472840309143,
+      "rewards/rejected": -0.6197813749313354,
+      "step": 230
+    },
+    {
+      "epoch": 0.3047619047619048,
+      "grad_norm": 0.19532087445259094,
+      "learning_rate": 4.690766700109659e-06,
+      "logits/chosen": 14.729398727416992,
+      "logits/rejected": 14.97203254699707,
+      "logps/chosen": -0.3013172149658203,
+      "logps/rejected": -0.41750937700271606,
+      "loss": 0.8814,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.45197582244873047,
+      "rewards/margins": 0.17428824305534363,
+      "rewards/rejected": -0.6262640357017517,
+      "step": 240
+    },
+    {
+      "epoch": 0.31746031746031744,
+      "grad_norm": 0.5795227885246277,
+      "learning_rate": 4.665063509461098e-06,
+      "logits/chosen": 14.328412055969238,
+      "logits/rejected": 14.645428657531738,
+      "logps/chosen": -0.3055194914340973,
+      "logps/rejected": -0.509621262550354,
+      "loss": 0.8478,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.45827922224998474,
+      "rewards/margins": 0.30615273118019104,
+      "rewards/rejected": -0.764431893825531,
+      "step": 250
+    },
+    {
+      "epoch": 0.31746031746031744,
+      "eval_logits/chosen": 14.310781478881836,
+      "eval_logits/rejected": 14.5652494430542,
+      "eval_logps/chosen": -0.3155084252357483,
+      "eval_logps/rejected": -0.8301784992218018,
+      "eval_loss": 0.7700436115264893,
+      "eval_rewards/accuracies": 0.640625,
+      "eval_rewards/chosen": -0.47326260805130005,
+      "eval_rewards/margins": 0.7720052003860474,
+      "eval_rewards/rejected": -1.2452678680419922,
+      "eval_runtime": 18.3413,
+      "eval_samples_per_second": 27.806,
+      "eval_steps_per_second": 3.489,
+      "step": 250
+    },
+    {
+      "epoch": 0.33015873015873015,
+      "grad_norm": 0.24876689910888672,
+      "learning_rate": 4.638410650401267e-06,
+      "logits/chosen": 14.115270614624023,
+      "logits/rejected": 14.167577743530273,
+      "logps/chosen": -0.32328444719314575,
+      "logps/rejected": -0.8564422726631165,
+      "loss": 0.795,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.484926700592041,
+      "rewards/margins": 0.7997367978096008,
+      "rewards/rejected": -1.284663438796997,
+      "step": 260
+    },
+    {
+      "epoch": 0.34285714285714286,
+      "grad_norm": 0.5779634118080139,
+      "learning_rate": 4.610819813755038e-06,
+      "logits/chosen": 14.019407272338867,
+      "logits/rejected": 13.600751876831055,
+      "logps/chosen": -0.35225847363471985,
+      "logps/rejected": -1.1924464702606201,
+      "loss": 0.7644,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.5283876657485962,
+      "rewards/margins": 1.2602821588516235,
+      "rewards/rejected": -1.7886695861816406,
+      "step": 270
+    },
+    {
+      "epoch": 0.35555555555555557,
+      "grad_norm": 0.2085695117712021,
+      "learning_rate": 4.582303101775249e-06,
+      "logits/chosen": 13.448820114135742,
+      "logits/rejected": 13.464624404907227,
+      "logps/chosen": -0.3442023992538452,
+      "logps/rejected": -1.4783053398132324,
+      "loss": 0.7671,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": -0.516303539276123,
+      "rewards/margins": 1.7011544704437256,
+      "rewards/rejected": -2.2174580097198486,
+      "step": 280
+    },
+    {
+      "epoch": 0.3682539682539683,
+      "grad_norm": 0.2825860381126404,
+      "learning_rate": 4.55287302283426e-06,
+      "logits/chosen": 13.038851737976074,
+      "logits/rejected": 13.013893127441406,
+      "logps/chosen": -0.3524690568447113,
+      "logps/rejected": -1.2354861497879028,
+      "loss": 0.7922,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.5287035703659058,
+      "rewards/margins": 1.3245255947113037,
+      "rewards/rejected": -1.8532291650772095,
+      "step": 290
+    },
+    {
+      "epoch": 0.38095238095238093,
+      "grad_norm": 0.30771639943122864,
+      "learning_rate": 4.522542485937369e-06,
+      "logits/chosen": 13.67170524597168,
+      "logits/rejected": 13.185449600219727,
+      "logps/chosen": -0.3915162682533264,
+      "logps/rejected": -1.3473880290985107,
+      "loss": 0.7506,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": -0.587274432182312,
+      "rewards/margins": 1.433807611465454,
+      "rewards/rejected": -2.0210821628570557,
+      "step": 300
+    },
+    {
+      "epoch": 0.38095238095238093,
+      "eval_logits/chosen": 13.198552131652832,
+      "eval_logits/rejected": 13.458879470825195,
+      "eval_logps/chosen": -0.37132135033607483,
+      "eval_logps/rejected": -1.4595866203308105,
+      "eval_loss": 0.7391816973686218,
+      "eval_rewards/accuracies": 0.640625,
+      "eval_rewards/chosen": -0.5569820404052734,
+      "eval_rewards/margins": 1.6323981285095215,
+      "eval_rewards/rejected": -2.189380168914795,
+      "eval_runtime": 18.3117,
+      "eval_samples_per_second": 27.851,
+      "eval_steps_per_second": 3.495,
+      "step": 300
+    },
+    {
+      "epoch": 0.39365079365079364,
+      "grad_norm": 0.3021286129951477,
+      "learning_rate": 4.491324795060491e-06,
+      "logits/chosen": 13.347516059875488,
+      "logits/rejected": 13.285726547241211,
+      "logps/chosen": -0.4108191132545471,
+      "logps/rejected": -1.436702013015747,
+      "loss": 0.7134,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.6162286400794983,
+      "rewards/margins": 1.5388243198394775,
+      "rewards/rejected": -2.15505313873291,
+      "step": 310
+    },
+    {
+      "epoch": 0.40634920634920635,
+      "grad_norm": 0.40123897790908813,
+      "learning_rate": 4.4592336433146e-06,
+      "logits/chosen": 12.949071884155273,
+      "logits/rejected": 12.86626148223877,
+      "logps/chosen": -0.42898029088974,
+      "logps/rejected": -1.1741807460784912,
+      "loss": 0.74,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.6434704065322876,
+      "rewards/margins": 1.1178009510040283,
+      "rewards/rejected": -1.7612712383270264,
+      "step": 320
+    },
+    {
+      "epoch": 0.41904761904761906,
+      "grad_norm": 0.4613189697265625,
+      "learning_rate": 4.426283106939474e-06,
+      "logits/chosen": 12.796838760375977,
+      "logits/rejected": 12.524165153503418,
+      "logps/chosen": -0.48975634574890137,
+      "logps/rejected": -1.2766364812850952,
+      "loss": 0.7408,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.7346345782279968,
+      "rewards/margins": 1.180320143699646,
+      "rewards/rejected": -1.9149547815322876,
+      "step": 330
+    },
+    {
+      "epoch": 0.43174603174603177,
+      "grad_norm": 0.49141305685043335,
+      "learning_rate": 4.3924876391293915e-06,
+      "logits/chosen": 12.58080005645752,
+      "logits/rejected": 12.5891695022583,
+      "logps/chosen": -0.5303409099578857,
+      "logps/rejected": -1.6262977123260498,
+      "loss": 0.7501,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.7955113649368286,
+      "rewards/margins": 1.643935203552246,
+      "rewards/rejected": -2.4394466876983643,
+      "step": 340
+    },
+    {
+      "epoch": 0.4444444444444444,
+      "grad_norm": 0.7533912062644958,
+      "learning_rate": 4.357862063693486e-06,
+      "logits/chosen": 12.15321159362793,
+      "logits/rejected": 12.38386344909668,
+      "logps/chosen": -0.5782582759857178,
+      "logps/rejected": -1.580635905265808,
+      "loss": 0.7405,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.8673874139785767,
+      "rewards/margins": 1.5035666227340698,
+      "rewards/rejected": -2.3709540367126465,
+      "step": 350
+    },
+    {
+      "epoch": 0.4444444444444444,
+      "eval_logits/chosen": 12.111034393310547,
+      "eval_logits/rejected": 12.33246898651123,
+      "eval_logps/chosen": -0.5941734910011292,
+      "eval_logps/rejected": -1.6774102449417114,
+      "eval_loss": 0.6953701972961426,
+      "eval_rewards/accuracies": 0.765625,
+      "eval_rewards/chosen": -0.8912601470947266,
+      "eval_rewards/margins": 1.6248550415039062,
+      "eval_rewards/rejected": -2.516115188598633,
+      "eval_runtime": 18.3366,
+      "eval_samples_per_second": 27.813,
+      "eval_steps_per_second": 3.49,
+      "step": 350
+    },
+    {
+      "epoch": 0.45714285714285713,
+      "grad_norm": 1.168564796447754,
+      "learning_rate": 4.322421568553529e-06,
+      "logits/chosen": 11.706710815429688,
+      "logits/rejected": 11.451011657714844,
+      "logps/chosen": -0.758882462978363,
+      "logps/rejected": -1.842167615890503,
+      "loss": 0.702,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.1383236646652222,
+      "rewards/margins": 1.6249277591705322,
+      "rewards/rejected": -2.763251304626465,
+      "step": 360
+    },
+    {
+      "epoch": 0.46984126984126984,
+      "grad_norm": 1.8962547779083252,
+      "learning_rate": 4.286181699082008e-06,
+      "logits/chosen": 11.41061019897461,
+      "logits/rejected": 11.534398078918457,
+      "logps/chosen": -1.1785552501678467,
+      "logps/rejected": -2.235652446746826,
+      "loss": 0.6781,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -1.7678327560424805,
+      "rewards/margins": 1.5856454372406006,
+      "rewards/rejected": -3.3534786701202393,
+      "step": 370
+    },
+    {
+      "epoch": 0.48253968253968255,
+      "grad_norm": 2.9152028560638428,
+      "learning_rate": 4.249158351283414e-06,
+      "logits/chosen": 11.582561492919922,
+      "logits/rejected": 11.439531326293945,
+      "logps/chosen": -1.9169971942901611,
+      "logps/rejected": -2.788179397583008,
+      "loss": 0.6033,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -2.875495672225952,
+      "rewards/margins": 1.306773066520691,
+      "rewards/rejected": -4.182269096374512,
+      "step": 380
+    },
+    {
+      "epoch": 0.49523809523809526,
+      "grad_norm": 1.5070641040802002,
+      "learning_rate": 4.211367764821722e-06,
+      "logits/chosen": 11.340807914733887,
+      "logits/rejected": 10.927321434020996,
+      "logps/chosen": -2.1075730323791504,
+      "logps/rejected": -3.4216556549072266,
+      "loss": 0.5698,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -3.1613595485687256,
+      "rewards/margins": 1.9711239337921143,
+      "rewards/rejected": -5.13248348236084,
+      "step": 390
+    },
+    {
+      "epoch": 0.5079365079365079,
+      "grad_norm": 2.6913516521453857,
+      "learning_rate": 4.172826515897146e-06,
+      "logits/chosen": 10.273612976074219,
+      "logits/rejected": 10.116204261779785,
+      "logps/chosen": -3.0047056674957275,
+      "logps/rejected": -4.053646564483643,
+      "loss": 0.5869,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.507058143615723,
+      "rewards/margins": 1.5734113454818726,
+      "rewards/rejected": -6.080469608306885,
+      "step": 400
+    },
+    {
+      "epoch": 0.5079365079365079,
+      "eval_logits/chosen": 9.58273696899414,
+      "eval_logits/rejected": 9.765192031860352,
+      "eval_logps/chosen": -2.4895575046539307,
+      "eval_logps/rejected": -3.8314545154571533,
+      "eval_loss": 0.5232856273651123,
+      "eval_rewards/accuracies": 0.875,
+      "eval_rewards/chosen": -3.7343361377716064,
+      "eval_rewards/margins": 2.012845754623413,
+      "eval_rewards/rejected": -5.7471818923950195,
+      "eval_runtime": 18.3437,
+      "eval_samples_per_second": 27.802,
+      "eval_steps_per_second": 3.489,
+      "step": 400
+    },
+    {
+      "epoch": 0.5206349206349207,
+      "grad_norm": 3.7197885513305664,
+      "learning_rate": 4.133551509975264e-06,
+      "logits/chosen": 9.081324577331543,
+      "logits/rejected": 9.309822082519531,
+      "logps/chosen": -2.2554564476013184,
+      "logps/rejected": -3.2624363899230957,
+      "loss": 0.5774,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -3.3831849098205566,
+      "rewards/margins": 1.5104694366455078,
+      "rewards/rejected": -4.893653869628906,
+      "step": 410
+    },
+    {
+      "epoch": 0.5333333333333333,
+      "grad_norm": 2.757873773574829,
+      "learning_rate": 4.093559974371725e-06,
+      "logits/chosen": 8.145927429199219,
+      "logits/rejected": 8.402036666870117,
+      "logps/chosen": -3.1110777854919434,
+      "logps/rejected": -4.202320098876953,
+      "loss": 0.5084,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -4.666616916656494,
+      "rewards/margins": 1.636863350868225,
+      "rewards/rejected": -6.30348014831543,
+      "step": 420
+    },
+    {
+      "epoch": 0.546031746031746,
+      "grad_norm": 3.3764102458953857,
+      "learning_rate": 4.052869450695776e-06,
+      "logits/chosen": 8.132574081420898,
+      "logits/rejected": 8.386279106140137,
+      "logps/chosen": -3.0871124267578125,
+      "logps/rejected": -4.399088382720947,
+      "loss": 0.5233,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.630668640136719,
+      "rewards/margins": 1.967963457107544,
+      "rewards/rejected": -6.598631858825684,
+      "step": 430
+    },
+    {
+      "epoch": 0.5587301587301587,
+      "grad_norm": 1.8188632726669312,
+      "learning_rate": 4.011497787155938e-06,
+      "logits/chosen": 8.104793548583984,
+      "logits/rejected": 8.25890064239502,
+      "logps/chosen": -3.598665237426758,
+      "logps/rejected": -5.2809929847717285,
+      "loss": 0.5272,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -5.397997856140137,
+      "rewards/margins": 2.5234923362731934,
+      "rewards/rejected": -7.921489715576172,
+      "step": 440
+    },
+    {
+      "epoch": 0.5714285714285714,
+      "grad_norm": 2.528883934020996,
+      "learning_rate": 3.969463130731183e-06,
+      "logits/chosen": 8.675338745117188,
+      "logits/rejected": 8.550267219543457,
+      "logps/chosen": -3.582223892211914,
+      "logps/rejected": -5.077345848083496,
+      "loss": 0.5717,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -5.373335838317871,
+      "rewards/margins": 2.2426836490631104,
+      "rewards/rejected": -7.616019248962402,
+      "step": 450
+    },
+    {
+      "epoch": 0.5714285714285714,
+      "eval_logits/chosen": 8.125189781188965,
+      "eval_logits/rejected": 8.239119529724121,
+      "eval_logps/chosen": -3.107839584350586,
+      "eval_logps/rejected": -4.726868629455566,
+      "eval_loss": 0.4568406939506531,
+      "eval_rewards/accuracies": 0.875,
+      "eval_rewards/chosen": -4.661759376525879,
+      "eval_rewards/margins": 2.4285435676574707,
+      "eval_rewards/rejected": -7.090303421020508,
+      "eval_runtime": 18.3222,
+      "eval_samples_per_second": 27.835,
+      "eval_steps_per_second": 3.493,
+      "step": 450
+    },
+    {
+      "epoch": 0.5841269841269842,
+      "grad_norm": 2.5079057216644287,
+      "learning_rate": 3.92678391921108e-06,
+      "logits/chosen": 8.503251075744629,
+      "logits/rejected": 8.457250595092773,
+      "logps/chosen": -3.499116897583008,
+      "logps/rejected": -5.12787389755249,
+      "loss": 0.4803,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.2486748695373535,
+      "rewards/margins": 2.4431352615356445,
+      "rewards/rejected": -7.69180965423584,
+      "step": 460
+    },
+    {
+      "epoch": 0.5968253968253968,
+      "grad_norm": 1.869667887687683,
+      "learning_rate": 3.88347887310836e-06,
+      "logits/chosen": 7.972044467926025,
+      "logits/rejected": 7.825163841247559,
+      "logps/chosen": -3.6900856494903564,
+      "logps/rejected": -5.166090488433838,
+      "loss": 0.4733,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.535128593444824,
+      "rewards/margins": 2.2140071392059326,
+      "rewards/rejected": -7.749135494232178,
+      "step": 470
+    },
+    {
+      "epoch": 0.6095238095238096,
+      "grad_norm": 2.8777735233306885,
+      "learning_rate": 3.839566987447492e-06,
+      "logits/chosen": 8.073511123657227,
+      "logits/rejected": 8.345193862915039,
+      "logps/chosen": -3.7006676197052,
+      "logps/rejected": -5.271113395690918,
+      "loss": 0.5048,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.551001071929932,
+      "rewards/margins": 2.3556694984436035,
+      "rewards/rejected": -7.906670570373535,
+      "step": 480
+    },
+    {
+      "epoch": 0.6222222222222222,
+      "grad_norm": 2.7042839527130127,
+      "learning_rate": 3.795067523432826e-06,
+      "logits/chosen": 8.602261543273926,
+      "logits/rejected": 8.32430362701416,
+      "logps/chosen": -3.9612879753112793,
+      "logps/rejected": -6.123431205749512,
+      "loss": 0.4728,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.94193172454834,
+      "rewards/margins": 3.2432150840759277,
+      "rewards/rejected": -9.185147285461426,
+      "step": 490
+    },
+    {
+      "epoch": 0.6349206349206349,
+      "grad_norm": 2.1625821590423584,
+      "learning_rate": 3.7500000000000005e-06,
+      "logits/chosen": 7.857954978942871,
+      "logits/rejected": 7.460604190826416,
+      "logps/chosen": -3.581023693084717,
+      "logps/rejected": -5.477323532104492,
+      "loss": 0.4296,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.371535778045654,
+      "rewards/margins": 2.844449520111084,
+      "rewards/rejected": -8.215986251831055,
+      "step": 500
+    },
+    {
+      "epoch": 0.6349206349206349,
+      "eval_logits/chosen": 7.520671844482422,
+      "eval_logits/rejected": 7.586723327636719,
+      "eval_logps/chosen": -3.1659929752349854,
+      "eval_logps/rejected": -4.955198764801025,
+      "eval_loss": 0.4266711175441742,
+      "eval_rewards/accuracies": 0.890625,
+      "eval_rewards/chosen": -4.748989105224609,
+      "eval_rewards/margins": 2.683809518814087,
+      "eval_rewards/rejected": -7.432799339294434,
+      "eval_runtime": 18.3456,
+      "eval_samples_per_second": 27.8,
+      "eval_steps_per_second": 3.489,
+      "step": 500
+    },
+    {
+      "epoch": 0.6476190476190476,
+      "grad_norm": 4.2689104080200195,
+      "learning_rate": 3.7043841852542884e-06,
+      "logits/chosen": 7.948687553405762,
+      "logits/rejected": 7.656899929046631,
+      "logps/chosen": -2.8974719047546387,
+      "logps/rejected": -4.501969337463379,
+      "loss": 0.4981,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.346207618713379,
+      "rewards/margins": 2.4067459106445312,
+      "rewards/rejected": -6.752954006195068,
+      "step": 510
+    },
+    {
+      "epoch": 0.6603174603174603,
+      "grad_norm": 3.299290895462036,
+      "learning_rate": 3.658240087799655e-06,
+      "logits/chosen": 7.676476955413818,
+      "logits/rejected": 7.332755088806152,
+      "logps/chosen": -3.9289379119873047,
+      "logps/rejected": -5.498073577880859,
+      "loss": 0.4736,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -5.893406867980957,
+      "rewards/margins": 2.353703498840332,
+      "rewards/rejected": -8.247110366821289,
+      "step": 520
+    },
+    {
+      "epoch": 0.6730158730158731,
+      "grad_norm": 3.1058828830718994,
+      "learning_rate": 3.611587947962319e-06,
+      "logits/chosen": 7.516615867614746,
+      "logits/rejected": 7.344383239746094,
+      "logps/chosen": -3.57861328125,
+      "logps/rejected": -5.48795223236084,
+      "loss": 0.502,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.367920398712158,
+      "rewards/margins": 2.8640081882476807,
+      "rewards/rejected": -8.231928825378418,
+      "step": 530
+    },
+    {
+      "epoch": 0.6857142857142857,
+      "grad_norm": 3.874868869781494,
+      "learning_rate": 3.564448228912682e-06,
+      "logits/chosen": 7.619947910308838,
+      "logits/rejected": 7.048402309417725,
+      "logps/chosen": -4.164804935455322,
+      "logps/rejected": -5.771735191345215,
+      "loss": 0.44,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.247206687927246,
+      "rewards/margins": 2.4103951454162598,
+      "rewards/rejected": -8.657602310180664,
+      "step": 540
+    },
+    {
+      "epoch": 0.6984126984126984,
+      "grad_norm": 1.7583000659942627,
+      "learning_rate": 3.516841607689501e-06,
+      "logits/chosen": 7.068112850189209,
+      "logits/rejected": 7.042977333068848,
+      "logps/chosen": -3.760105609893799,
+      "logps/rejected": -5.3382248878479,
+      "loss": 0.4751,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.640158653259277,
+      "rewards/margins": 2.3671793937683105,
+      "rewards/rejected": -8.00733757019043,
+      "step": 550
+    },
+    {
+      "epoch": 0.6984126984126984,
+      "eval_logits/chosen": 7.213573455810547,
+      "eval_logits/rejected": 7.25667667388916,
+      "eval_logps/chosen": -3.8712658882141113,
+      "eval_logps/rejected": -5.882909774780273,
+      "eval_loss": 0.3922204077243805,
+      "eval_rewards/accuracies": 0.90625,
+      "eval_rewards/chosen": -5.806899070739746,
+      "eval_rewards/margins": 3.017465591430664,
+      "eval_rewards/rejected": -8.824363708496094,
+      "eval_runtime": 18.3231,
+      "eval_samples_per_second": 27.834,
+      "eval_steps_per_second": 3.493,
+      "step": 550
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 1500,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 50,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.3248943036836086e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-550/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6104f1a61db1df2c606ca77691d14a708e1aba41ef959437c3789e25b88c33b
+size 7224

checkpoint-550/zero_to_fp32.py

@@ -0,0 +1,674 @@
+#!/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 . output_dir/
+#   or
+#   python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import json
+from tqdm import tqdm
+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, exclude_frozen_parameters):
+    """
+    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,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+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,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _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 tqdm(param_shapes.items(), desc='Gathering Sharded Weights'):
+        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,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _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, exclude_frozen_parameters=False):
+    """
+    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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+
+    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, exclude_frozen_parameters)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+                                               output_dir,
+                                               max_shard_size="5GB",
+                                               safe_serialization=False,
+                                               tag=None,
+                                               exclude_frozen_parameters=False):
+    """
+    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_dir``: directory to the pytorch fp32 state_dict output files
+        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+        - ``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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+    # Dependency pre-check
+    if safe_serialization:
+        try:
+            from safetensors.torch import save_file
+        except ImportError:
+            print('If you want to use `safe_serialization`, please `pip install safetensors`')
+            raise
+    if max_shard_size is not None:
+        try:
+            from huggingface_hub import split_torch_state_dict_into_shards
+        except ImportError:
+            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+            raise
+
+    # Convert zero checkpoint to state_dict
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
+
+    # Shard the model if it is too big.
+    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+    if max_shard_size is not None:
+        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+        state_dict_split = split_torch_state_dict_into_shards(state_dict,
+                                                              filename_pattern=filename_pattern,
+                                                              max_shard_size=max_shard_size)
+    else:
+        from collections import namedtuple
+        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+        state_dict_split = StateDictSplit(is_sharded=False,
+                                          filename_to_tensors={weights_name: list(state_dict.keys())})
+
+    # Save the model
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard = {tensor: state_dict[tensor].contiguous() for tensor in tensors}
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard, output_path)
+
+    # Save index if sharded
+    if state_dict_split.is_sharded:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
+        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+        save_index_file = os.path.join(output_dir, save_index_file)
+        with open(save_index_file, "w", encoding="utf-8") as f:
+            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+            f.write(content)
+
+
+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_dir",
+                        type=str,
+                        help="directory to the pytorch fp32 state_dict output files"
+                        "(e.g. path/checkpoint-12-output/)")
+    parser.add_argument(
+        "--max_shard_size",
+        type=str,
+        default="5GB",
+        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+        "without CPU OOM issues.")
+    parser.add_argument(
+        "--safe_serialization",
+        default=False,
+        action='store_true',
+        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+    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("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    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_dir,
+                                               max_shard_size=args.max_shard_size,
+                                               safe_serialization=args.safe_serialization,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)