Training in progress, step 296, checkpoint

checkpoint-296/README.md

@@ -0,0 +1,202 @@
+---
+base_model: unsloth/Mistral-Small-Instruct-2409
+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-296/adapter_config.json

@@ -0,0 +1,37 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "unsloth/Mistral-Small-Instruct-2409",
+  "bias": "none",
+  "eva_config": null,
+  "exclude_modules": null,
+  "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": 64,
+  "lora_bias": false,
+  "lora_dropout": 0.25,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 32,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "up_proj",
+    "q_proj",
+    "k_proj",
+    "down_proj",
+    "v_proj",
+    "o_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-296/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:052d9fedaf0fa6a741029ff6a0685fb89390397bc3cc317b1ef6c135bb628718
+size 381788248

checkpoint-296/global_step296/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:cb98fc1364da3c1c33f20462b187b0cd2ef40cbdd068cb60d176e1ad6f2df7f5
+size 337530064

checkpoint-296/global_step296/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:c1394758f4f02ce1a761b558e0edd1d1331ed7b938055de65bab53e77ed8d218
+size 337530064

checkpoint-296/global_step296/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:a79abebd6461818d4090d4b1416c9a3f878374fd5b1aee5b3f5f8a9cef94ad9b
+size 337530064

checkpoint-296/global_step296/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16d798065d5444afd41c03aea0c44d6da60c0dd912d4bab36c312ea5a60d7882
+size 337530064

checkpoint-296/global_step296/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fcf4b62edd46f064c027c7d8381013a57c737503839803ee4fa0b8536c3b5704
+size 348711702

checkpoint-296/global_step296/zero_pp_rank_1_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40540f91660eb8675322d9d2c2541786e760c19744ad225bfca1140706026d99
+size 348711702

checkpoint-296/global_step296/zero_pp_rank_2_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92fb36f9f7b2451001d4180de258372f464d89ca9285da9720da3edc60dd0228
+size 348711702

checkpoint-296/global_step296/zero_pp_rank_3_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f92a657666174e5b9aface7200c3aca7a2c02d4f343b7f3bc0127a8996c58d6f
+size 348711702

checkpoint-296/latest

@@ -0,0 +1 @@
+global_step296

checkpoint-296/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:029651baaee62ba13d0fefefdd856356c4b1d62bd431e2f912bc801aaceffa2d
+size 14960

checkpoint-296/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d2e11a8f992ade71a4a3f9278a11ad46f52c38226dd6bccd059f4cec511de2f
+size 14960

checkpoint-296/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3cacbfa33ee9e24f3a00ba0a3bd02d2abfeb4b0db4756080f14e6e7d34f1aa3a
+size 14960

checkpoint-296/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d365ab3c8428ab021faf6545a150416f07211c3ea57155495abdd155bdc72a99
+size 14960

checkpoint-296/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5730af9c278a41d02afecd2348635d0a61707085122dd2733b7e78eddf91c693
+size 1064

checkpoint-296/special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "[control_748]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoint-296/tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:59f95e28944c062244741268596badc900df86c7f5ded05088d2da22a7379e06
+size 587583

checkpoint-296/trainer_state.json

@@ -0,0 +1,2185 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 30,
+  "global_step": 296,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0033783783783783786,
+      "grad_norm": 1.4937225174709319,
+      "learning_rate": 2.5e-06,
+      "loss": 2.1185,
+      "step": 1
+    },
+    {
+      "epoch": 0.0033783783783783786,
+      "eval_loss": 2.0952866077423096,
+      "eval_runtime": 187.5733,
+      "eval_samples_per_second": 0.267,
+      "eval_steps_per_second": 0.069,
+      "step": 1
+    },
+    {
+      "epoch": 0.006756756756756757,
+      "grad_norm": 1.442277502174201,
+      "learning_rate": 5e-06,
+      "loss": 2.1679,
+      "step": 2
+    },
+    {
+      "epoch": 0.010135135135135136,
+      "grad_norm": 1.4361264107689526,
+      "learning_rate": 7.5e-06,
+      "loss": 2.1475,
+      "step": 3
+    },
+    {
+      "epoch": 0.013513513513513514,
+      "grad_norm": 1.5925067561400204,
+      "learning_rate": 1e-05,
+      "loss": 2.1111,
+      "step": 4
+    },
+    {
+      "epoch": 0.016891891891891893,
+      "grad_norm": 1.5729343116389016,
+      "learning_rate": 1.25e-05,
+      "loss": 2.1772,
+      "step": 5
+    },
+    {
+      "epoch": 0.02027027027027027,
+      "grad_norm": 0.77386625695677,
+      "learning_rate": 1.5e-05,
+      "loss": 2.1445,
+      "step": 6
+    },
+    {
+      "epoch": 0.02364864864864865,
+      "grad_norm": 0.9687092422157498,
+      "learning_rate": 1.75e-05,
+      "loss": 2.0929,
+      "step": 7
+    },
+    {
+      "epoch": 0.02702702702702703,
+      "grad_norm": 0.5745308246234537,
+      "learning_rate": 2e-05,
+      "loss": 2.1085,
+      "step": 8
+    },
+    {
+      "epoch": 0.030405405405405407,
+      "grad_norm": 0.7880273258575979,
+      "learning_rate": 2.25e-05,
+      "loss": 2.1717,
+      "step": 9
+    },
+    {
+      "epoch": 0.033783783783783786,
+      "grad_norm": 0.5987395071382063,
+      "learning_rate": 2.5e-05,
+      "loss": 2.074,
+      "step": 10
+    },
+    {
+      "epoch": 0.037162162162162164,
+      "grad_norm": 0.8312307017045393,
+      "learning_rate": 2.7500000000000004e-05,
+      "loss": 2.2338,
+      "step": 11
+    },
+    {
+      "epoch": 0.04054054054054054,
+      "grad_norm": 0.7733440491356615,
+      "learning_rate": 3e-05,
+      "loss": 2.2148,
+      "step": 12
+    },
+    {
+      "epoch": 0.04391891891891892,
+      "grad_norm": 0.7510878886656168,
+      "learning_rate": 3.2500000000000004e-05,
+      "loss": 2.243,
+      "step": 13
+    },
+    {
+      "epoch": 0.0472972972972973,
+      "grad_norm": 0.6210806018904155,
+      "learning_rate": 3.5e-05,
+      "loss": 1.9863,
+      "step": 14
+    },
+    {
+      "epoch": 0.05067567567567568,
+      "grad_norm": 1.018373738736693,
+      "learning_rate": 3.7500000000000003e-05,
+      "loss": 2.0294,
+      "step": 15
+    },
+    {
+      "epoch": 0.05405405405405406,
+      "grad_norm": 0.771759792529477,
+      "learning_rate": 4e-05,
+      "loss": 2.0353,
+      "step": 16
+    },
+    {
+      "epoch": 0.057432432432432436,
+      "grad_norm": 0.5625682320002908,
+      "learning_rate": 4.25e-05,
+      "loss": 2.0775,
+      "step": 17
+    },
+    {
+      "epoch": 0.060810810810810814,
+      "grad_norm": 0.5157565411105806,
+      "learning_rate": 4.5e-05,
+      "loss": 2.1791,
+      "step": 18
+    },
+    {
+      "epoch": 0.06418918918918919,
+      "grad_norm": 0.8173537204599863,
+      "learning_rate": 4.75e-05,
+      "loss": 2.0549,
+      "step": 19
+    },
+    {
+      "epoch": 0.06756756756756757,
+      "grad_norm": 2.5851976259061664,
+      "learning_rate": 5e-05,
+      "loss": 2.0757,
+      "step": 20
+    },
+    {
+      "epoch": 0.07094594594594594,
+      "grad_norm": 0.5130578194248145,
+      "learning_rate": 4.999854243002125e-05,
+      "loss": 2.2505,
+      "step": 21
+    },
+    {
+      "epoch": 0.07432432432432433,
+      "grad_norm": 0.5786770218696188,
+      "learning_rate": 4.999416990893036e-05,
+      "loss": 2.2533,
+      "step": 22
+    },
+    {
+      "epoch": 0.0777027027027027,
+      "grad_norm": 0.8219446198677083,
+      "learning_rate": 4.998688300323891e-05,
+      "loss": 2.1092,
+      "step": 23
+    },
+    {
+      "epoch": 0.08108108108108109,
+      "grad_norm": 0.7123342305131811,
+      "learning_rate": 4.997668265705137e-05,
+      "loss": 2.3369,
+      "step": 24
+    },
+    {
+      "epoch": 0.08445945945945946,
+      "grad_norm": 0.44584577341815756,
+      "learning_rate": 4.9963570191942696e-05,
+      "loss": 2.0125,
+      "step": 25
+    },
+    {
+      "epoch": 0.08783783783783784,
+      "grad_norm": 0.5308641065159894,
+      "learning_rate": 4.994754730678713e-05,
+      "loss": 2.0653,
+      "step": 26
+    },
+    {
+      "epoch": 0.09121621621621621,
+      "grad_norm": 0.4833891468070717,
+      "learning_rate": 4.992861607753817e-05,
+      "loss": 2.0177,
+      "step": 27
+    },
+    {
+      "epoch": 0.0945945945945946,
+      "grad_norm": 0.4487741255853019,
+      "learning_rate": 4.9906778956959454e-05,
+      "loss": 1.9773,
+      "step": 28
+    },
+    {
+      "epoch": 0.09797297297297297,
+      "grad_norm": 0.7070413556606907,
+      "learning_rate": 4.988203877430713e-05,
+      "loss": 2.0163,
+      "step": 29
+    },
+    {
+      "epoch": 0.10135135135135136,
+      "grad_norm": 0.6433406570980086,
+      "learning_rate": 4.985439873496321e-05,
+      "loss": 2.055,
+      "step": 30
+    },
+    {
+      "epoch": 0.10135135135135136,
+      "eval_loss": 2.022883653640747,
+      "eval_runtime": 194.2976,
+      "eval_samples_per_second": 0.257,
+      "eval_steps_per_second": 0.067,
+      "step": 30
+    },
+    {
+      "epoch": 0.10472972972972973,
+      "grad_norm": 1.9573381693992367,
+      "learning_rate": 4.982386242002024e-05,
+      "loss": 2.3371,
+      "step": 31
+    },
+    {
+      "epoch": 0.10810810810810811,
+      "grad_norm": 1.6578265269127788,
+      "learning_rate": 4.979043378581744e-05,
+      "loss": 2.1288,
+      "step": 32
+    },
+    {
+      "epoch": 0.11148648648648649,
+      "grad_norm": 0.5544836623908992,
+      "learning_rate": 4.975411716342802e-05,
+      "loss": 2.1887,
+      "step": 33
+    },
+    {
+      "epoch": 0.11486486486486487,
+      "grad_norm": 0.5455040506142,
+      "learning_rate": 4.971491725809807e-05,
+      "loss": 2.1214,
+      "step": 34
+    },
+    {
+      "epoch": 0.11824324324324324,
+      "grad_norm": 0.5065209751020103,
+      "learning_rate": 4.967283914863693e-05,
+      "loss": 2.1692,
+      "step": 35
+    },
+    {
+      "epoch": 0.12162162162162163,
+      "grad_norm": 0.6453232517165945,
+      "learning_rate": 4.96278882867592e-05,
+      "loss": 2.0598,
+      "step": 36
+    },
+    {
+      "epoch": 0.125,
+      "grad_norm": 1.1731862208697716,
+      "learning_rate": 4.9580070496378364e-05,
+      "loss": 2.2156,
+      "step": 37
+    },
+    {
+      "epoch": 0.12837837837837837,
+      "grad_norm": 0.7667101172051888,
+      "learning_rate": 4.952939197285227e-05,
+      "loss": 2.1392,
+      "step": 38
+    },
+    {
+      "epoch": 0.13175675675675674,
+      "grad_norm": 0.41781856442196297,
+      "learning_rate": 4.947585928218041e-05,
+      "loss": 2.1534,
+      "step": 39
+    },
+    {
+      "epoch": 0.13513513513513514,
+      "grad_norm": 0.5945736341225079,
+      "learning_rate": 4.9419479360153286e-05,
+      "loss": 1.9795,
+      "step": 40
+    },
+    {
+      "epoch": 0.13851351351351351,
+      "grad_norm": 0.6467707595603454,
+      "learning_rate": 4.936025951145368e-05,
+      "loss": 2.0017,
+      "step": 41
+    },
+    {
+      "epoch": 0.14189189189189189,
+      "grad_norm": 0.5109600024808127,
+      "learning_rate": 4.929820740871039e-05,
+      "loss": 2.2144,
+      "step": 42
+    },
+    {
+      "epoch": 0.14527027027027026,
+      "grad_norm": 1.0548277894819795,
+      "learning_rate": 4.9233331091504034e-05,
+      "loss": 2.0657,
+      "step": 43
+    },
+    {
+      "epoch": 0.14864864864864866,
+      "grad_norm": 0.6580975841596574,
+      "learning_rate": 4.916563896532549e-05,
+      "loss": 2.1431,
+      "step": 44
+    },
+    {
+      "epoch": 0.15202702702702703,
+      "grad_norm": 0.3988619886160755,
+      "learning_rate": 4.9095139800486824e-05,
+      "loss": 2.0123,
+      "step": 45
+    },
+    {
+      "epoch": 0.1554054054054054,
+      "grad_norm": 0.5209858406284475,
+      "learning_rate": 4.9021842730985036e-05,
+      "loss": 2.2487,
+      "step": 46
+    },
+    {
+      "epoch": 0.15878378378378377,
+      "grad_norm": 0.5433634333987769,
+      "learning_rate": 4.894575725331862e-05,
+      "loss": 2.1736,
+      "step": 47
+    },
+    {
+      "epoch": 0.16216216216216217,
+      "grad_norm": 0.5109855249570185,
+      "learning_rate": 4.886689322525719e-05,
+      "loss": 2.0823,
+      "step": 48
+    },
+    {
+      "epoch": 0.16554054054054054,
+      "grad_norm": 0.6194335580813743,
+      "learning_rate": 4.878526086456426e-05,
+      "loss": 2.1036,
+      "step": 49
+    },
+    {
+      "epoch": 0.16891891891891891,
+      "grad_norm": 1.0780413583147488,
+      "learning_rate": 4.8700870747673466e-05,
+      "loss": 2.0302,
+      "step": 50
+    },
+    {
+      "epoch": 0.17229729729729729,
+      "grad_norm": 0.580340495769622,
+      "learning_rate": 4.8613733808318204e-05,
+      "loss": 2.1776,
+      "step": 51
+    },
+    {
+      "epoch": 0.17567567567567569,
+      "grad_norm": 0.8129429382764808,
+      "learning_rate": 4.85238613361151e-05,
+      "loss": 2.0965,
+      "step": 52
+    },
+    {
+      "epoch": 0.17905405405405406,
+      "grad_norm": 0.7779782864994534,
+      "learning_rate": 4.8431264975101245e-05,
+      "loss": 2.1582,
+      "step": 53
+    },
+    {
+      "epoch": 0.18243243243243243,
+      "grad_norm": 0.8528745581545691,
+      "learning_rate": 4.8335956722225616e-05,
+      "loss": 2.1511,
+      "step": 54
+    },
+    {
+      "epoch": 0.1858108108108108,
+      "grad_norm": 0.49188324413548323,
+      "learning_rate": 4.823794892579471e-05,
+      "loss": 2.1583,
+      "step": 55
+    },
+    {
+      "epoch": 0.1891891891891892,
+      "grad_norm": 0.724195478643155,
+      "learning_rate": 4.8137254283872696e-05,
+      "loss": 1.9438,
+      "step": 56
+    },
+    {
+      "epoch": 0.19256756756756757,
+      "grad_norm": 0.8754624301347438,
+      "learning_rate": 4.803388584263618e-05,
+      "loss": 2.1349,
+      "step": 57
+    },
+    {
+      "epoch": 0.19594594594594594,
+      "grad_norm": 0.4870044791836658,
+      "learning_rate": 4.7927856994684e-05,
+      "loss": 2.0239,
+      "step": 58
+    },
+    {
+      "epoch": 0.19932432432432431,
+      "grad_norm": 0.5041896039366629,
+      "learning_rate": 4.781918147730199e-05,
+      "loss": 2.0841,
+      "step": 59
+    },
+    {
+      "epoch": 0.20270270270270271,
+      "grad_norm": 0.4874068239264915,
+      "learning_rate": 4.7707873370683163e-05,
+      "loss": 2.1407,
+      "step": 60
+    },
+    {
+      "epoch": 0.20270270270270271,
+      "eval_loss": 2.015751838684082,
+      "eval_runtime": 196.268,
+      "eval_samples_per_second": 0.255,
+      "eval_steps_per_second": 0.066,
+      "step": 60
+    },
+    {
+      "epoch": 0.20608108108108109,
+      "grad_norm": 0.5424018707478311,
+      "learning_rate": 4.75939470961035e-05,
+      "loss": 2.186,
+      "step": 61
+    },
+    {
+      "epoch": 0.20945945945945946,
+      "grad_norm": 0.5115180976703219,
+      "learning_rate": 4.747741741405344e-05,
+      "loss": 2.2014,
+      "step": 62
+    },
+    {
+      "epoch": 0.21283783783783783,
+      "grad_norm": 0.5058558197015601,
+      "learning_rate": 4.735829942232555e-05,
+      "loss": 2.0927,
+      "step": 63
+    },
+    {
+      "epoch": 0.21621621621621623,
+      "grad_norm": 0.8021043767946636,
+      "learning_rate": 4.7236608554058375e-05,
+      "loss": 2.1884,
+      "step": 64
+    },
+    {
+      "epoch": 0.2195945945945946,
+      "grad_norm": 0.5293842363703689,
+      "learning_rate": 4.711236057573691e-05,
+      "loss": 2.0714,
+      "step": 65
+    },
+    {
+      "epoch": 0.22297297297297297,
+      "grad_norm": 0.48722486629288786,
+      "learning_rate": 4.6985571585149876e-05,
+      "loss": 2.0562,
+      "step": 66
+    },
+    {
+      "epoch": 0.22635135135135134,
+      "grad_norm": 0.4279248526312935,
+      "learning_rate": 4.685625800930406e-05,
+      "loss": 2.0847,
+      "step": 67
+    },
+    {
+      "epoch": 0.22972972972972974,
+      "grad_norm": 0.5130161800768928,
+      "learning_rate": 4.6724436602296e-05,
+      "loss": 2.0617,
+      "step": 68
+    },
+    {
+      "epoch": 0.23310810810810811,
+      "grad_norm": 0.7460272297512813,
+      "learning_rate": 4.659012444314128e-05,
+      "loss": 2.1029,
+      "step": 69
+    },
+    {
+      "epoch": 0.23648648648648649,
+      "grad_norm": 0.4943699960691202,
+      "learning_rate": 4.645333893356176e-05,
+      "loss": 1.9948,
+      "step": 70
+    },
+    {
+      "epoch": 0.23986486486486486,
+      "grad_norm": 0.6350670407835639,
+      "learning_rate": 4.6314097795731e-05,
+      "loss": 2.0935,
+      "step": 71
+    },
+    {
+      "epoch": 0.24324324324324326,
+      "grad_norm": 0.542940777570525,
+      "learning_rate": 4.6172419069978065e-05,
+      "loss": 2.0267,
+      "step": 72
+    },
+    {
+      "epoch": 0.24662162162162163,
+      "grad_norm": 0.4567422527911003,
+      "learning_rate": 4.602832111245029e-05,
+      "loss": 1.9971,
+      "step": 73
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.4699317313725709,
+      "learning_rate": 4.5881822592734946e-05,
+      "loss": 2.1758,
+      "step": 74
+    },
+    {
+      "epoch": 0.2533783783783784,
+      "grad_norm": 0.6913326755097541,
+      "learning_rate": 4.573294249144041e-05,
+      "loss": 2.1574,
+      "step": 75
+    },
+    {
+      "epoch": 0.25675675675675674,
+      "grad_norm": 0.4362574963180428,
+      "learning_rate": 4.5581700097737015e-05,
+      "loss": 2.0498,
+      "step": 76
+    },
+    {
+      "epoch": 0.26013513513513514,
+      "grad_norm": 0.6932798791839507,
+      "learning_rate": 4.542811500685785e-05,
+      "loss": 2.0777,
+      "step": 77
+    },
+    {
+      "epoch": 0.2635135135135135,
+      "grad_norm": 0.4469936094893796,
+      "learning_rate": 4.527220711756007e-05,
+      "loss": 2.1046,
+      "step": 78
+    },
+    {
+      "epoch": 0.2668918918918919,
+      "grad_norm": 0.5112163825070385,
+      "learning_rate": 4.511399662954667e-05,
+      "loss": 2.0955,
+      "step": 79
+    },
+    {
+      "epoch": 0.2702702702702703,
+      "grad_norm": 0.4272672573543929,
+      "learning_rate": 4.4953504040849445e-05,
+      "loss": 2.1927,
+      "step": 80
+    },
+    {
+      "epoch": 0.27364864864864863,
+      "grad_norm": 0.4203647309881028,
+      "learning_rate": 4.479075014517321e-05,
+      "loss": 2.0421,
+      "step": 81
+    },
+    {
+      "epoch": 0.27702702702702703,
+      "grad_norm": 0.5896256230009808,
+      "learning_rate": 4.462575602920171e-05,
+      "loss": 2.109,
+      "step": 82
+    },
+    {
+      "epoch": 0.28040540540540543,
+      "grad_norm": 0.7482696034746834,
+      "learning_rate": 4.445854306986563e-05,
+      "loss": 2.0851,
+      "step": 83
+    },
+    {
+      "epoch": 0.28378378378378377,
+      "grad_norm": 0.508289422303338,
+      "learning_rate": 4.428913293157293e-05,
+      "loss": 2.1759,
+      "step": 84
+    },
+    {
+      "epoch": 0.28716216216216217,
+      "grad_norm": 0.4749189097462539,
+      "learning_rate": 4.411754756340198e-05,
+      "loss": 2.0721,
+      "step": 85
+    },
+    {
+      "epoch": 0.2905405405405405,
+      "grad_norm": 0.493594282927666,
+      "learning_rate": 4.3943809196257794e-05,
+      "loss": 2.0351,
+      "step": 86
+    },
+    {
+      "epoch": 0.2939189189189189,
+      "grad_norm": 0.4836545743981663,
+      "learning_rate": 4.376794033999177e-05,
+      "loss": 2.0896,
+      "step": 87
+    },
+    {
+      "epoch": 0.2972972972972973,
+      "grad_norm": 0.47767898533003106,
+      "learning_rate": 4.358996378048524e-05,
+      "loss": 2.083,
+      "step": 88
+    },
+    {
+      "epoch": 0.30067567567567566,
+      "grad_norm": 0.4390064059005752,
+      "learning_rate": 4.340990257669732e-05,
+      "loss": 2.0627,
+      "step": 89
+    },
+    {
+      "epoch": 0.30405405405405406,
+      "grad_norm": 0.4601711355647231,
+      "learning_rate": 4.3227780057677345e-05,
+      "loss": 2.0997,
+      "step": 90
+    },
+    {
+      "epoch": 0.30405405405405406,
+      "eval_loss": 2.011120080947876,
+      "eval_runtime": 190.8365,
+      "eval_samples_per_second": 0.262,
+      "eval_steps_per_second": 0.068,
+      "step": 90
+    },
+    {
+      "epoch": 0.30743243243243246,
+      "grad_norm": 0.5654017461216198,
+      "learning_rate": 4.304361981954231e-05,
+      "loss": 2.2149,
+      "step": 91
+    },
+    {
+      "epoch": 0.3108108108108108,
+      "grad_norm": 0.5064873219371222,
+      "learning_rate": 4.285744572241972e-05,
+      "loss": 2.1093,
+      "step": 92
+    },
+    {
+      "epoch": 0.3141891891891892,
+      "grad_norm": 0.46470652155232134,
+      "learning_rate": 4.266928188735621e-05,
+      "loss": 2.1098,
+      "step": 93
+    },
+    {
+      "epoch": 0.31756756756756754,
+      "grad_norm": 0.4413493288478021,
+      "learning_rate": 4.247915269319241e-05,
+      "loss": 2.1431,
+      "step": 94
+    },
+    {
+      "epoch": 0.32094594594594594,
+      "grad_norm": 1.0020586721524896,
+      "learning_rate": 4.2287082773404386e-05,
+      "loss": 2.0877,
+      "step": 95
+    },
+    {
+      "epoch": 0.32432432432432434,
+      "grad_norm": 0.4848089444845991,
+      "learning_rate": 4.209309701291201e-05,
+      "loss": 2.1175,
+      "step": 96
+    },
+    {
+      "epoch": 0.3277027027027027,
+      "grad_norm": 0.5555159336979827,
+      "learning_rate": 4.189722054485492e-05,
+      "loss": 2.1563,
+      "step": 97
+    },
+    {
+      "epoch": 0.3310810810810811,
+      "grad_norm": 0.3802895433591588,
+      "learning_rate": 4.169947874733619e-05,
+      "loss": 2.0732,
+      "step": 98
+    },
+    {
+      "epoch": 0.3344594594594595,
+      "grad_norm": 0.7320601930371831,
+      "learning_rate": 4.149989724013425e-05,
+      "loss": 2.1452,
+      "step": 99
+    },
+    {
+      "epoch": 0.33783783783783783,
+      "grad_norm": 0.8602357197050848,
+      "learning_rate": 4.1298501881383624e-05,
+      "loss": 2.1766,
+      "step": 100
+    },
+    {
+      "epoch": 0.34121621621621623,
+      "grad_norm": 0.5831812051877154,
+      "learning_rate": 4.109531876422463e-05,
+      "loss": 2.1593,
+      "step": 101
+    },
+    {
+      "epoch": 0.34459459459459457,
+      "grad_norm": 0.7665489432929005,
+      "learning_rate": 4.089037421342277e-05,
+      "loss": 2.0295,
+      "step": 102
+    },
+    {
+      "epoch": 0.34797297297297297,
+      "grad_norm": 0.45337471468141477,
+      "learning_rate": 4.0683694781958e-05,
+      "loss": 2.1087,
+      "step": 103
+    },
+    {
+      "epoch": 0.35135135135135137,
+      "grad_norm": 0.4591955107711257,
+      "learning_rate": 4.047530724758451e-05,
+      "loss": 2.0354,
+      "step": 104
+    },
+    {
+      "epoch": 0.3547297297297297,
+      "grad_norm": 0.7505711753697653,
+      "learning_rate": 4.026523860936132e-05,
+      "loss": 2.0258,
+      "step": 105
+    },
+    {
+      "epoch": 0.3581081081081081,
+      "grad_norm": 0.589397231315979,
+      "learning_rate": 4.005351608415426e-05,
+      "loss": 2.0051,
+      "step": 106
+    },
+    {
+      "epoch": 0.3614864864864865,
+      "grad_norm": 0.6872824203092515,
+      "learning_rate": 3.9840167103109675e-05,
+      "loss": 1.9355,
+      "step": 107
+    },
+    {
+      "epoch": 0.36486486486486486,
+      "grad_norm": 0.45802674220232215,
+      "learning_rate": 3.9625219308100455e-05,
+      "loss": 2.0836,
+      "step": 108
+    },
+    {
+      "epoch": 0.36824324324324326,
+      "grad_norm": 0.5524787455898191,
+      "learning_rate": 3.940870054814462e-05,
+      "loss": 2.0474,
+      "step": 109
+    },
+    {
+      "epoch": 0.3716216216216216,
+      "grad_norm": 0.38416177227514725,
+      "learning_rate": 3.919063887579726e-05,
+      "loss": 2.0985,
+      "step": 110
+    },
+    {
+      "epoch": 0.375,
+      "grad_norm": 0.45960539321926364,
+      "learning_rate": 3.897106254351587e-05,
+      "loss": 2.0129,
+      "step": 111
+    },
+    {
+      "epoch": 0.3783783783783784,
+      "grad_norm": 0.37861955022931376,
+      "learning_rate": 3.875e-05,
+      "loss": 2.1427,
+      "step": 112
+    },
+    {
+      "epoch": 0.38175675675675674,
+      "grad_norm": 0.581523763120522,
+      "learning_rate": 3.852747988650539e-05,
+      "loss": 2.0318,
+      "step": 113
+    },
+    {
+      "epoch": 0.38513513513513514,
+      "grad_norm": 0.3504600135974039,
+      "learning_rate": 3.83035310331331e-05,
+      "loss": 2.033,
+      "step": 114
+    },
+    {
+      "epoch": 0.3885135135135135,
+      "grad_norm": 0.4242352094167052,
+      "learning_rate": 3.807818245509429e-05,
+      "loss": 2.034,
+      "step": 115
+    },
+    {
+      "epoch": 0.3918918918918919,
+      "grad_norm": 0.4250673616692747,
+      "learning_rate": 3.785146334895093e-05,
+      "loss": 2.1702,
+      "step": 116
+    },
+    {
+      "epoch": 0.3952702702702703,
+      "grad_norm": 3.2751318573643617,
+      "learning_rate": 3.762340308883302e-05,
+      "loss": 2.1026,
+      "step": 117
+    },
+    {
+      "epoch": 0.39864864864864863,
+      "grad_norm": 0.48584128738743915,
+      "learning_rate": 3.739403122263288e-05,
+      "loss": 2.22,
+      "step": 118
+    },
+    {
+      "epoch": 0.40202702702702703,
+      "grad_norm": 0.4187766943285619,
+      "learning_rate": 3.716337746817685e-05,
+      "loss": 2.1541,
+      "step": 119
+    },
+    {
+      "epoch": 0.40540540540540543,
+      "grad_norm": 0.5051183151449187,
+      "learning_rate": 3.6931471709374946e-05,
+      "loss": 2.1113,
+      "step": 120
+    },
+    {
+      "epoch": 0.40540540540540543,
+      "eval_loss": 2.009887218475342,
+      "eval_runtime": 193.8883,
+      "eval_samples_per_second": 0.258,
+      "eval_steps_per_second": 0.067,
+      "step": 120
+    },
+    {
+      "epoch": 0.40878378378378377,
+      "grad_norm": 0.40423483975901836,
+      "learning_rate": 3.669834399234913e-05,
+      "loss": 2.0209,
+      "step": 121
+    },
+    {
+      "epoch": 0.41216216216216217,
+      "grad_norm": 0.3825955755250103,
+      "learning_rate": 3.646402452154043e-05,
+      "loss": 2.074,
+      "step": 122
+    },
+    {
+      "epoch": 0.4155405405405405,
+      "grad_norm": 1.4013949742645142,
+      "learning_rate": 3.622854365579561e-05,
+      "loss": 2.1807,
+      "step": 123
+    },
+    {
+      "epoch": 0.4189189189189189,
+      "grad_norm": 0.43240866134882766,
+      "learning_rate": 3.5991931904433824e-05,
+      "loss": 1.9988,
+      "step": 124
+    },
+    {
+      "epoch": 0.4222972972972973,
+      "grad_norm": 0.4777714379483413,
+      "learning_rate": 3.575421992329377e-05,
+      "loss": 2.0801,
+      "step": 125
+    },
+    {
+      "epoch": 0.42567567567567566,
+      "grad_norm": 0.49357523405169307,
+      "learning_rate": 3.551543851076188e-05,
+      "loss": 2.0035,
+      "step": 126
+    },
+    {
+      "epoch": 0.42905405405405406,
+      "grad_norm": 0.3447821776586979,
+      "learning_rate": 3.5275618603782006e-05,
+      "loss": 2.0435,
+      "step": 127
+    },
+    {
+      "epoch": 0.43243243243243246,
+      "grad_norm": 0.3718901049854852,
+      "learning_rate": 3.503479127384719e-05,
+      "loss": 1.8828,
+      "step": 128
+    },
+    {
+      "epoch": 0.4358108108108108,
+      "grad_norm": 0.3888324394883177,
+      "learning_rate": 3.479298772297398e-05,
+      "loss": 2.1094,
+      "step": 129
+    },
+    {
+      "epoch": 0.4391891891891892,
+      "grad_norm": 0.5267690929558577,
+      "learning_rate": 3.4550239279659854e-05,
+      "loss": 2.1593,
+      "step": 130
+    },
+    {
+      "epoch": 0.44256756756756754,
+      "grad_norm": 0.4636315581971976,
+      "learning_rate": 3.4306577394824207e-05,
+      "loss": 2.1605,
+      "step": 131
+    },
+    {
+      "epoch": 0.44594594594594594,
+      "grad_norm": 0.3968487601273563,
+      "learning_rate": 3.406203363773356e-05,
+      "loss": 2.0023,
+      "step": 132
+    },
+    {
+      "epoch": 0.44932432432432434,
+      "grad_norm": 0.392460435962094,
+      "learning_rate": 3.381663969191137e-05,
+      "loss": 2.1517,
+      "step": 133
+    },
+    {
+      "epoch": 0.4527027027027027,
+      "grad_norm": 0.4031589419917055,
+      "learning_rate": 3.3570427351033046e-05,
+      "loss": 2.0701,
+      "step": 134
+    },
+    {
+      "epoch": 0.4560810810810811,
+      "grad_norm": 0.4253231691696294,
+      "learning_rate": 3.332342851480672e-05,
+      "loss": 2.0944,
+      "step": 135
+    },
+    {
+      "epoch": 0.4594594594594595,
+      "grad_norm": 1.089550042036897,
+      "learning_rate": 3.307567518484025e-05,
+      "loss": 2.1879,
+      "step": 136
+    },
+    {
+      "epoch": 0.46283783783783783,
+      "grad_norm": 0.5433174961880026,
+      "learning_rate": 3.282719946049505e-05,
+      "loss": 2.1142,
+      "step": 137
+    },
+    {
+      "epoch": 0.46621621621621623,
+      "grad_norm": 0.47522024279968983,
+      "learning_rate": 3.257803353472724e-05,
+      "loss": 2.0765,
+      "step": 138
+    },
+    {
+      "epoch": 0.46959459459459457,
+      "grad_norm": 0.6397031743657627,
+      "learning_rate": 3.232820968991664e-05,
+      "loss": 2.0565,
+      "step": 139
+    },
+    {
+      "epoch": 0.47297297297297297,
+      "grad_norm": 0.8231847588094275,
+      "learning_rate": 3.207776029368427e-05,
+      "loss": 2.1382,
+      "step": 140
+    },
+    {
+      "epoch": 0.47635135135135137,
+      "grad_norm": 0.4277815234715947,
+      "learning_rate": 3.1826717794698635e-05,
+      "loss": 2.1454,
+      "step": 141
+    },
+    {
+      "epoch": 0.4797297297297297,
+      "grad_norm": 0.395176536808308,
+      "learning_rate": 3.157511471847176e-05,
+      "loss": 2.0293,
+      "step": 142
+    },
+    {
+      "epoch": 0.4831081081081081,
+      "grad_norm": 0.42124877734068217,
+      "learning_rate": 3.1322983663145e-05,
+      "loss": 2.0417,
+      "step": 143
+    },
+    {
+      "epoch": 0.4864864864864865,
+      "grad_norm": 0.6103856586396839,
+      "learning_rate": 3.107035729526566e-05,
+      "loss": 2.0703,
+      "step": 144
+    },
+    {
+      "epoch": 0.48986486486486486,
+      "grad_norm": 0.40802626443551787,
+      "learning_rate": 3.081726834555458e-05,
+      "loss": 1.9202,
+      "step": 145
+    },
+    {
+      "epoch": 0.49324324324324326,
+      "grad_norm": 0.4038507240265364,
+      "learning_rate": 3.0563749604665556e-05,
+      "loss": 2.1105,
+      "step": 146
+    },
+    {
+      "epoch": 0.4966216216216216,
+      "grad_norm": 0.39745914058435033,
+      "learning_rate": 3.0309833918936865e-05,
+      "loss": 2.0854,
+      "step": 147
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.39094920659893395,
+      "learning_rate": 3.0055554186135688e-05,
+      "loss": 2.1638,
+      "step": 148
+    },
+    {
+      "epoch": 0.5033783783783784,
+      "grad_norm": 0.4684851166837971,
+      "learning_rate": 2.980094335119577e-05,
+      "loss": 2.0368,
+      "step": 149
+    },
+    {
+      "epoch": 0.5067567567567568,
+      "grad_norm": 0.8309495298034889,
+      "learning_rate": 2.9546034401949064e-05,
+      "loss": 2.1695,
+      "step": 150
+    },
+    {
+      "epoch": 0.5067567567567568,
+      "eval_loss": 2.0075740814208984,
+      "eval_runtime": 193.1213,
+      "eval_samples_per_second": 0.259,
+      "eval_steps_per_second": 0.067,
+      "step": 150
+    },
+    {
+      "epoch": 0.5101351351351351,
+      "grad_norm": 0.5583192813687627,
+      "learning_rate": 2.9290860364851702e-05,
+      "loss": 2.0582,
+      "step": 151
+    },
+    {
+      "epoch": 0.5135135135135135,
+      "grad_norm": 0.42473795943255627,
+      "learning_rate": 2.90354543007051e-05,
+      "loss": 2.0871,
+      "step": 152
+    },
+    {
+      "epoch": 0.5168918918918919,
+      "grad_norm": 0.4247509513157766,
+      "learning_rate": 2.877984930037251e-05,
+      "loss": 2.0964,
+      "step": 153
+    },
+    {
+      "epoch": 0.5202702702702703,
+      "grad_norm": 0.8778087000903096,
+      "learning_rate": 2.8524078480491684e-05,
+      "loss": 2.0506,
+      "step": 154
+    },
+    {
+      "epoch": 0.5236486486486487,
+      "grad_norm": 0.501109409670388,
+      "learning_rate": 2.826817497918428e-05,
+      "loss": 2.1435,
+      "step": 155
+    },
+    {
+      "epoch": 0.527027027027027,
+      "grad_norm": 0.7730432872928119,
+      "learning_rate": 2.8012171951762378e-05,
+      "loss": 2.0909,
+      "step": 156
+    },
+    {
+      "epoch": 0.5304054054054054,
+      "grad_norm": 0.4007388361090252,
+      "learning_rate": 2.7756102566432845e-05,
+      "loss": 2.1613,
+      "step": 157
+    },
+    {
+      "epoch": 0.5337837837837838,
+      "grad_norm": 0.4121213099878934,
+      "learning_rate": 2.7500000000000004e-05,
+      "loss": 2.2213,
+      "step": 158
+    },
+    {
+      "epoch": 0.5371621621621622,
+      "grad_norm": 0.4237936256648311,
+      "learning_rate": 2.7243897433567157e-05,
+      "loss": 2.1795,
+      "step": 159
+    },
+    {
+      "epoch": 0.5405405405405406,
+      "grad_norm": 0.41107097005560067,
+      "learning_rate": 2.6987828048237624e-05,
+      "loss": 2.0895,
+      "step": 160
+    },
+    {
+      "epoch": 0.543918918918919,
+      "grad_norm": 0.4463337134292773,
+      "learning_rate": 2.6731825020815725e-05,
+      "loss": 2.046,
+      "step": 161
+    },
+    {
+      "epoch": 0.5472972972972973,
+      "grad_norm": 0.5230967905508315,
+      "learning_rate": 2.6475921519508325e-05,
+      "loss": 2.0374,
+      "step": 162
+    },
+    {
+      "epoch": 0.5506756756756757,
+      "grad_norm": 0.414839716539812,
+      "learning_rate": 2.62201506996275e-05,
+      "loss": 2.2485,
+      "step": 163
+    },
+    {
+      "epoch": 0.5540540540540541,
+      "grad_norm": 0.38087462228816815,
+      "learning_rate": 2.5964545699294906e-05,
+      "loss": 2.0365,
+      "step": 164
+    },
+    {
+      "epoch": 0.5574324324324325,
+      "grad_norm": 0.5305482547516058,
+      "learning_rate": 2.570913963514831e-05,
+      "loss": 2.0701,
+      "step": 165
+    },
+    {
+      "epoch": 0.5608108108108109,
+      "grad_norm": 0.39988875104823635,
+      "learning_rate": 2.5453965598050944e-05,
+      "loss": 2.1985,
+      "step": 166
+    },
+    {
+      "epoch": 0.5641891891891891,
+      "grad_norm": 0.6054380331009478,
+      "learning_rate": 2.5199056648804233e-05,
+      "loss": 2.1136,
+      "step": 167
+    },
+    {
+      "epoch": 0.5675675675675675,
+      "grad_norm": 0.39275163056643564,
+      "learning_rate": 2.4944445813864314e-05,
+      "loss": 2.0735,
+      "step": 168
+    },
+    {
+      "epoch": 0.5709459459459459,
+      "grad_norm": 0.7078296605563206,
+      "learning_rate": 2.469016608106315e-05,
+      "loss": 2.1578,
+      "step": 169
+    },
+    {
+      "epoch": 0.5743243243243243,
+      "grad_norm": 0.439519908152304,
+      "learning_rate": 2.443625039533446e-05,
+      "loss": 2.0469,
+      "step": 170
+    },
+    {
+      "epoch": 0.5777027027027027,
+      "grad_norm": 0.47171431683346926,
+      "learning_rate": 2.4182731654445427e-05,
+      "loss": 2.0609,
+      "step": 171
+    },
+    {
+      "epoch": 0.581081081081081,
+      "grad_norm": 0.43725040423496425,
+      "learning_rate": 2.3929642704734347e-05,
+      "loss": 2.1686,
+      "step": 172
+    },
+    {
+      "epoch": 0.5844594594594594,
+      "grad_norm": 0.5156240873689197,
+      "learning_rate": 2.3677016336855002e-05,
+      "loss": 2.111,
+      "step": 173
+    },
+    {
+      "epoch": 0.5878378378378378,
+      "grad_norm": 0.478909789809296,
+      "learning_rate": 2.3424885281528248e-05,
+      "loss": 2.0993,
+      "step": 174
+    },
+    {
+      "epoch": 0.5912162162162162,
+      "grad_norm": 0.8822684350527039,
+      "learning_rate": 2.3173282205301367e-05,
+      "loss": 2.091,
+      "step": 175
+    },
+    {
+      "epoch": 0.5945945945945946,
+      "grad_norm": 0.5005747298379083,
+      "learning_rate": 2.2922239706315745e-05,
+      "loss": 2.0998,
+      "step": 176
+    },
+    {
+      "epoch": 0.597972972972973,
+      "grad_norm": 0.38147794073978136,
+      "learning_rate": 2.2671790310083364e-05,
+      "loss": 2.0433,
+      "step": 177
+    },
+    {
+      "epoch": 0.6013513513513513,
+      "grad_norm": 0.449696856682795,
+      "learning_rate": 2.2421966465272765e-05,
+      "loss": 1.9547,
+      "step": 178
+    },
+    {
+      "epoch": 0.6047297297297297,
+      "grad_norm": 0.4046647546809551,
+      "learning_rate": 2.217280053950495e-05,
+      "loss": 2.0109,
+      "step": 179
+    },
+    {
+      "epoch": 0.6081081081081081,
+      "grad_norm": 0.44319480569143777,
+      "learning_rate": 2.1924324815159757e-05,
+      "loss": 2.0782,
+      "step": 180
+    },
+    {
+      "epoch": 0.6081081081081081,
+      "eval_loss": 2.006296157836914,
+      "eval_runtime": 194.8007,
+      "eval_samples_per_second": 0.257,
+      "eval_steps_per_second": 0.067,
+      "step": 180
+    },
+    {
+      "epoch": 0.6114864864864865,
+      "grad_norm": 0.4245370001741025,
+      "learning_rate": 2.1676571485193282e-05,
+      "loss": 1.9819,
+      "step": 181
+    },
+    {
+      "epoch": 0.6148648648648649,
+      "grad_norm": 0.4123813369355281,
+      "learning_rate": 2.1429572648966956e-05,
+      "loss": 2.0597,
+      "step": 182
+    },
+    {
+      "epoch": 0.6182432432432432,
+      "grad_norm": 0.4870232191555961,
+      "learning_rate": 2.1183360308088636e-05,
+      "loss": 2.0595,
+      "step": 183
+    },
+    {
+      "epoch": 0.6216216216216216,
+      "grad_norm": 0.43367499716068364,
+      "learning_rate": 2.0937966362266443e-05,
+      "loss": 1.7438,
+      "step": 184
+    },
+    {
+      "epoch": 0.625,
+      "grad_norm": 0.3828393006442412,
+      "learning_rate": 2.06934226051758e-05,
+      "loss": 2.2068,
+      "step": 185
+    },
+    {
+      "epoch": 0.6283783783783784,
+      "grad_norm": 0.4948622876303367,
+      "learning_rate": 2.0449760720340155e-05,
+      "loss": 2.1126,
+      "step": 186
+    },
+    {
+      "epoch": 0.6317567567567568,
+      "grad_norm": 0.4073583504258278,
+      "learning_rate": 2.0207012277026016e-05,
+      "loss": 2.0846,
+      "step": 187
+    },
+    {
+      "epoch": 0.6351351351351351,
+      "grad_norm": 0.5109456209828284,
+      "learning_rate": 1.9965208726152813e-05,
+      "loss": 2.2111,
+      "step": 188
+    },
+    {
+      "epoch": 0.6385135135135135,
+      "grad_norm": 0.6271240563609504,
+      "learning_rate": 1.9724381396217996e-05,
+      "loss": 2.0691,
+      "step": 189
+    },
+    {
+      "epoch": 0.6418918918918919,
+      "grad_norm": 0.7995422703795263,
+      "learning_rate": 1.948456148923813e-05,
+      "loss": 2.0512,
+      "step": 190
+    },
+    {
+      "epoch": 0.6452702702702703,
+      "grad_norm": 0.5355787419178449,
+      "learning_rate": 1.9245780076706233e-05,
+      "loss": 1.9599,
+      "step": 191
+    },
+    {
+      "epoch": 0.6486486486486487,
+      "grad_norm": 0.37446732667446725,
+      "learning_rate": 1.9008068095566178e-05,
+      "loss": 1.9695,
+      "step": 192
+    },
+    {
+      "epoch": 0.652027027027027,
+      "grad_norm": 0.4147478978916466,
+      "learning_rate": 1.8771456344204385e-05,
+      "loss": 2.1151,
+      "step": 193
+    },
+    {
+      "epoch": 0.6554054054054054,
+      "grad_norm": 0.922383794900146,
+      "learning_rate": 1.8535975478459566e-05,
+      "loss": 2.0877,
+      "step": 194
+    },
+    {
+      "epoch": 0.6587837837837838,
+      "grad_norm": 0.5037479123782382,
+      "learning_rate": 1.830165600765087e-05,
+      "loss": 2.0584,
+      "step": 195
+    },
+    {
+      "epoch": 0.6621621621621622,
+      "grad_norm": 0.4337707246605357,
+      "learning_rate": 1.806852829062507e-05,
+      "loss": 2.1358,
+      "step": 196
+    },
+    {
+      "epoch": 0.6655405405405406,
+      "grad_norm": 1.0443488522366138,
+      "learning_rate": 1.783662253182316e-05,
+      "loss": 2.0686,
+      "step": 197
+    },
+    {
+      "epoch": 0.668918918918919,
+      "grad_norm": 0.3652644321832928,
+      "learning_rate": 1.7605968777367116e-05,
+      "loss": 2.1038,
+      "step": 198
+    },
+    {
+      "epoch": 0.6722972972972973,
+      "grad_norm": 0.44849286894487655,
+      "learning_rate": 1.7376596911166987e-05,
+      "loss": 2.2002,
+      "step": 199
+    },
+    {
+      "epoch": 0.6756756756756757,
+      "grad_norm": 0.6883412198217931,
+      "learning_rate": 1.7148536651049078e-05,
+      "loss": 2.1097,
+      "step": 200
+    },
+    {
+      "epoch": 0.6790540540540541,
+      "grad_norm": 0.38108344540028705,
+      "learning_rate": 1.692181754490571e-05,
+      "loss": 2.1487,
+      "step": 201
+    },
+    {
+      "epoch": 0.6824324324324325,
+      "grad_norm": 0.46596636892914783,
+      "learning_rate": 1.66964689668669e-05,
+      "loss": 2.1226,
+      "step": 202
+    },
+    {
+      "epoch": 0.6858108108108109,
+      "grad_norm": 0.3884889567792045,
+      "learning_rate": 1.6472520113494622e-05,
+      "loss": 2.0384,
+      "step": 203
+    },
+    {
+      "epoch": 0.6891891891891891,
+      "grad_norm": 0.5294696992679853,
+      "learning_rate": 1.6250000000000005e-05,
+      "loss": 2.0854,
+      "step": 204
+    },
+    {
+      "epoch": 0.6925675675675675,
+      "grad_norm": 0.6679912280514742,
+      "learning_rate": 1.6028937456484137e-05,
+      "loss": 2.1521,
+      "step": 205
+    },
+    {
+      "epoch": 0.6959459459459459,
+      "grad_norm": 1.1501328678855647,
+      "learning_rate": 1.580936112420275e-05,
+      "loss": 1.9754,
+      "step": 206
+    },
+    {
+      "epoch": 0.6993243243243243,
+      "grad_norm": 0.4599912704355663,
+      "learning_rate": 1.559129945185538e-05,
+      "loss": 2.0103,
+      "step": 207
+    },
+    {
+      "epoch": 0.7027027027027027,
+      "grad_norm": 0.4723451173074553,
+      "learning_rate": 1.5374780691899553e-05,
+      "loss": 2.0294,
+      "step": 208
+    },
+    {
+      "epoch": 0.706081081081081,
+      "grad_norm": 0.39614835888065925,
+      "learning_rate": 1.5159832896890324e-05,
+      "loss": 2.0027,
+      "step": 209
+    },
+    {
+      "epoch": 0.7094594594594594,
+      "grad_norm": 0.5674089763099159,
+      "learning_rate": 1.4946483915845752e-05,
+      "loss": 2.0972,
+      "step": 210
+    },
+    {
+      "epoch": 0.7094594594594594,
+      "eval_loss": 2.005359411239624,
+      "eval_runtime": 196.3936,
+      "eval_samples_per_second": 0.255,
+      "eval_steps_per_second": 0.066,
+      "step": 210
+    },
+    {
+      "epoch": 0.7128378378378378,
+      "grad_norm": 0.405311292588835,
+      "learning_rate": 1.473476139063869e-05,
+      "loss": 2.1081,
+      "step": 211
+    },
+    {
+      "epoch": 0.7162162162162162,
+      "grad_norm": 0.49941841218777355,
+      "learning_rate": 1.4524692752415493e-05,
+      "loss": 1.992,
+      "step": 212
+    },
+    {
+      "epoch": 0.7195945945945946,
+      "grad_norm": 0.3870145887647159,
+      "learning_rate": 1.4316305218041997e-05,
+      "loss": 2.0732,
+      "step": 213
+    },
+    {
+      "epoch": 0.722972972972973,
+      "grad_norm": 0.8010230321806846,
+      "learning_rate": 1.4109625786577236e-05,
+      "loss": 2.031,
+      "step": 214
+    },
+    {
+      "epoch": 0.7263513513513513,
+      "grad_norm": 0.41451504848593906,
+      "learning_rate": 1.3904681235775374e-05,
+      "loss": 1.9604,
+      "step": 215
+    },
+    {
+      "epoch": 0.7297297297297297,
+      "grad_norm": 0.42012094253734433,
+      "learning_rate": 1.370149811861638e-05,
+      "loss": 2.0378,
+      "step": 216
+    },
+    {
+      "epoch": 0.7331081081081081,
+      "grad_norm": 0.5698583105554919,
+      "learning_rate": 1.3500102759865758e-05,
+      "loss": 2.1111,
+      "step": 217
+    },
+    {
+      "epoch": 0.7364864864864865,
+      "grad_norm": 0.3926750948896524,
+      "learning_rate": 1.330052125266382e-05,
+      "loss": 2.1808,
+      "step": 218
+    },
+    {
+      "epoch": 0.7398648648648649,
+      "grad_norm": 0.436664706138193,
+      "learning_rate": 1.310277945514508e-05,
+      "loss": 2.0413,
+      "step": 219
+    },
+    {
+      "epoch": 0.7432432432432432,
+      "grad_norm": 0.5181403672040665,
+      "learning_rate": 1.2906902987087994e-05,
+      "loss": 2.0807,
+      "step": 220
+    },
+    {
+      "epoch": 0.7466216216216216,
+      "grad_norm": 0.400974423390128,
+      "learning_rate": 1.2712917226595616e-05,
+      "loss": 2.0943,
+      "step": 221
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.4460134300706243,
+      "learning_rate": 1.2520847306807589e-05,
+      "loss": 2.0802,
+      "step": 222
+    },
+    {
+      "epoch": 0.7533783783783784,
+      "grad_norm": 0.4131195235199069,
+      "learning_rate": 1.2330718112643792e-05,
+      "loss": 2.0257,
+      "step": 223
+    },
+    {
+      "epoch": 0.7567567567567568,
+      "grad_norm": 0.6039834322797323,
+      "learning_rate": 1.2142554277580288e-05,
+      "loss": 2.1739,
+      "step": 224
+    },
+    {
+      "epoch": 0.7601351351351351,
+      "grad_norm": 0.4403554972296534,
+      "learning_rate": 1.1956380180457688e-05,
+      "loss": 1.929,
+      "step": 225
+    },
+    {
+      "epoch": 0.7635135135135135,
+      "grad_norm": 0.6136709559845319,
+      "learning_rate": 1.1772219942322659e-05,
+      "loss": 2.0956,
+      "step": 226
+    },
+    {
+      "epoch": 0.7668918918918919,
+      "grad_norm": 0.3721043934343538,
+      "learning_rate": 1.1590097423302684e-05,
+      "loss": 2.0624,
+      "step": 227
+    },
+    {
+      "epoch": 0.7702702702702703,
+      "grad_norm": 0.3751617145534424,
+      "learning_rate": 1.1410036219514762e-05,
+      "loss": 2.1306,
+      "step": 228
+    },
+    {
+      "epoch": 0.7736486486486487,
+      "grad_norm": 0.4615904753215304,
+      "learning_rate": 1.1232059660008237e-05,
+      "loss": 2.1549,
+      "step": 229
+    },
+    {
+      "epoch": 0.777027027027027,
+      "grad_norm": 0.34886057454784225,
+      "learning_rate": 1.1056190803742208e-05,
+      "loss": 1.8962,
+      "step": 230
+    },
+    {
+      "epoch": 0.7804054054054054,
+      "grad_norm": 0.379971430732989,
+      "learning_rate": 1.088245243659803e-05,
+      "loss": 2.1084,
+      "step": 231
+    },
+    {
+      "epoch": 0.7837837837837838,
+      "grad_norm": 0.4076749411088033,
+      "learning_rate": 1.0710867068427078e-05,
+      "loss": 2.0582,
+      "step": 232
+    },
+    {
+      "epoch": 0.7871621621621622,
+      "grad_norm": 0.37680297146772307,
+      "learning_rate": 1.0541456930134383e-05,
+      "loss": 2.1541,
+      "step": 233
+    },
+    {
+      "epoch": 0.7905405405405406,
+      "grad_norm": 0.3986805252065202,
+      "learning_rate": 1.0374243970798297e-05,
+      "loss": 2.0239,
+      "step": 234
+    },
+    {
+      "epoch": 0.793918918918919,
+      "grad_norm": 0.4216381351000177,
+      "learning_rate": 1.0209249854826793e-05,
+      "loss": 1.9384,
+      "step": 235
+    },
+    {
+      "epoch": 0.7972972972972973,
+      "grad_norm": 3.4841982335621053,
+      "learning_rate": 1.0046495959150554e-05,
+      "loss": 2.1648,
+      "step": 236
+    },
+    {
+      "epoch": 0.8006756756756757,
+      "grad_norm": 0.550886438560183,
+      "learning_rate": 9.88600337045333e-06,
+      "loss": 2.0643,
+      "step": 237
+    },
+    {
+      "epoch": 0.8040540540540541,
+      "grad_norm": 0.5312105646194669,
+      "learning_rate": 9.727792882439938e-06,
+      "loss": 2.1149,
+      "step": 238
+    },
+    {
+      "epoch": 0.8074324324324325,
+      "grad_norm": 1.2275389743439304,
+      "learning_rate": 9.57188499314215e-06,
+      "loss": 2.2107,
+      "step": 239
+    },
+    {
+      "epoch": 0.8108108108108109,
+      "grad_norm": 0.755407956454159,
+      "learning_rate": 9.41829990226299e-06,
+      "loss": 2.1761,
+      "step": 240
+    },
+    {
+      "epoch": 0.8108108108108109,
+      "eval_loss": 2.0042636394500732,
+      "eval_runtime": 195.0896,
+      "eval_samples_per_second": 0.256,
+      "eval_steps_per_second": 0.067,
+      "step": 240
+    },
+    {
+      "epoch": 0.8141891891891891,
+      "grad_norm": 0.48379829157730286,
+      "learning_rate": 9.267057508559592e-06,
+      "loss": 2.101,
+      "step": 241
+    },
+    {
+      "epoch": 0.8175675675675675,
+      "grad_norm": 0.44013024339715,
+      "learning_rate": 9.118177407265056e-06,
+      "loss": 2.1196,
+      "step": 242
+    },
+    {
+      "epoch": 0.8209459459459459,
+      "grad_norm": 0.561381768462023,
+      "learning_rate": 8.971678887549712e-06,
+      "loss": 2.149,
+      "step": 243
+    },
+    {
+      "epoch": 0.8243243243243243,
+      "grad_norm": 0.37732815129251784,
+      "learning_rate": 8.827580930021936e-06,
+      "loss": 2.1098,
+      "step": 244
+    },
+    {
+      "epoch": 0.8277027027027027,
+      "grad_norm": 0.6499115709600214,
+      "learning_rate": 8.685902204269012e-06,
+      "loss": 1.9987,
+      "step": 245
+    },
+    {
+      "epoch": 0.831081081081081,
+      "grad_norm": 0.3777685916732647,
+      "learning_rate": 8.546661066438243e-06,
+      "loss": 2.1232,
+      "step": 246
+    },
+    {
+      "epoch": 0.8344594594594594,
+      "grad_norm": 0.997691102079538,
+      "learning_rate": 8.409875556858728e-06,
+      "loss": 2.2468,
+      "step": 247
+    },
+    {
+      "epoch": 0.8378378378378378,
+      "grad_norm": 1.0313423837079596,
+      "learning_rate": 8.275563397704008e-06,
+      "loss": 2.0612,
+      "step": 248
+    },
+    {
+      "epoch": 0.8412162162162162,
+      "grad_norm": 0.9545397247064048,
+      "learning_rate": 8.143741990695947e-06,
+      "loss": 2.0295,
+      "step": 249
+    },
+    {
+      "epoch": 0.8445945945945946,
+      "grad_norm": 0.48112662217454943,
+      "learning_rate": 8.01442841485013e-06,
+      "loss": 2.016,
+      "step": 250
+    },
+    {
+      "epoch": 0.847972972972973,
+      "grad_norm": 0.4525753292819526,
+      "learning_rate": 7.887639424263095e-06,
+      "loss": 2.1112,
+      "step": 251
+    },
+    {
+      "epoch": 0.8513513513513513,
+      "grad_norm": 0.373269871593188,
+      "learning_rate": 7.76339144594163e-06,
+      "loss": 2.1332,
+      "step": 252
+    },
+    {
+      "epoch": 0.8547297297297297,
+      "grad_norm": 0.4714615102710501,
+      "learning_rate": 7.64170057767445e-06,
+      "loss": 2.1127,
+      "step": 253
+    },
+    {
+      "epoch": 0.8581081081081081,
+      "grad_norm": 0.41117656642398737,
+      "learning_rate": 7.522582585946558e-06,
+      "loss": 2.0966,
+      "step": 254
+    },
+    {
+      "epoch": 0.8614864864864865,
+      "grad_norm": 0.39133852746779035,
+      "learning_rate": 7.406052903896504e-06,
+      "loss": 2.1377,
+      "step": 255
+    },
+    {
+      "epoch": 0.8648648648648649,
+      "grad_norm": 0.4976521052133734,
+      "learning_rate": 7.292126629316841e-06,
+      "loss": 2.1264,
+      "step": 256
+    },
+    {
+      "epoch": 0.8682432432432432,
+      "grad_norm": 0.5748609387660506,
+      "learning_rate": 7.1808185226980144e-06,
+      "loss": 2.078,
+      "step": 257
+    },
+    {
+      "epoch": 0.8716216216216216,
+      "grad_norm": 0.6237308439362074,
+      "learning_rate": 7.072143005315997e-06,
+      "loss": 2.1125,
+      "step": 258
+    },
+    {
+      "epoch": 0.875,
+      "grad_norm": 0.45517735653366964,
+      "learning_rate": 6.966114157363821e-06,
+      "loss": 2.1685,
+      "step": 259
+    },
+    {
+      "epoch": 0.8783783783783784,
+      "grad_norm": 0.37602443499611526,
+      "learning_rate": 6.862745716127312e-06,
+      "loss": 2.1349,
+      "step": 260
+    },
+    {
+      "epoch": 0.8817567567567568,
+      "grad_norm": 0.43834187328579305,
+      "learning_rate": 6.762051074205289e-06,
+      "loss": 2.0764,
+      "step": 261
+    },
+    {
+      "epoch": 0.8851351351351351,
+      "grad_norm": 0.4170412679000652,
+      "learning_rate": 6.6640432777743815e-06,
+      "loss": 2.0925,
+      "step": 262
+    },
+    {
+      "epoch": 0.8885135135135135,
+      "grad_norm": 0.410513816061216,
+      "learning_rate": 6.568735024898755e-06,
+      "loss": 2.103,
+      "step": 263
+    },
+    {
+      "epoch": 0.8918918918918919,
+      "grad_norm": 0.3898098438793997,
+      "learning_rate": 6.476138663884902e-06,
+      "loss": 2.1138,
+      "step": 264
+    },
+    {
+      "epoch": 0.8952702702702703,
+      "grad_norm": 0.3975585157411543,
+      "learning_rate": 6.386266191681797e-06,
+      "loss": 2.0935,
+      "step": 265
+    },
+    {
+      "epoch": 0.8986486486486487,
+      "grad_norm": 0.35048570379733146,
+      "learning_rate": 6.299129252326541e-06,
+      "loss": 2.0116,
+      "step": 266
+    },
+    {
+      "epoch": 0.902027027027027,
+      "grad_norm": 0.3919603572459853,
+      "learning_rate": 6.214739135435742e-06,
+      "loss": 2.0657,
+      "step": 267
+    },
+    {
+      "epoch": 0.9054054054054054,
+      "grad_norm": 0.557862525874976,
+      "learning_rate": 6.133106774742819e-06,
+      "loss": 2.0765,
+      "step": 268
+    },
+    {
+      "epoch": 0.9087837837837838,
+      "grad_norm": 0.39493324357545007,
+      "learning_rate": 6.054242746681382e-06,
+      "loss": 2.1249,
+      "step": 269
+    },
+    {
+      "epoch": 0.9121621621621622,
+      "grad_norm": 1.9963252009359573,
+      "learning_rate": 5.9781572690149684e-06,
+      "loss": 2.1288,
+      "step": 270
+    },
+    {
+      "epoch": 0.9121621621621622,
+      "eval_loss": 2.0036399364471436,
+      "eval_runtime": 195.7956,
+      "eval_samples_per_second": 0.255,
+      "eval_steps_per_second": 0.066,
+      "step": 270
+    },
+    {
+      "epoch": 0.9155405405405406,
+      "grad_norm": 0.5120749568149566,
+      "learning_rate": 5.904860199513184e-06,
+      "loss": 2.0928,
+      "step": 271
+    },
+    {
+      "epoch": 0.918918918918919,
+      "grad_norm": 0.43131777610423944,
+      "learning_rate": 5.834361034674521e-06,
+      "loss": 2.0637,
+      "step": 272
+    },
+    {
+      "epoch": 0.9222972972972973,
+      "grad_norm": 0.35373523391766476,
+      "learning_rate": 5.766668908495966e-06,
+      "loss": 2.0978,
+      "step": 273
+    },
+    {
+      "epoch": 0.9256756756756757,
+      "grad_norm": 0.5253182101700391,
+      "learning_rate": 5.701792591289609e-06,
+      "loss": 1.9415,
+      "step": 274
+    },
+    {
+      "epoch": 0.9290540540540541,
+      "grad_norm": 0.37847718648888473,
+      "learning_rate": 5.6397404885463175e-06,
+      "loss": 2.0668,
+      "step": 275
+    },
+    {
+      "epoch": 0.9324324324324325,
+      "grad_norm": 0.40314006462162916,
+      "learning_rate": 5.580520639846723e-06,
+      "loss": 2.1415,
+      "step": 276
+    },
+    {
+      "epoch": 0.9358108108108109,
+      "grad_norm": 0.7112783095355424,
+      "learning_rate": 5.524140717819588e-06,
+      "loss": 2.0203,
+      "step": 277
+    },
+    {
+      "epoch": 0.9391891891891891,
+      "grad_norm": 0.3708045804577933,
+      "learning_rate": 5.470608027147735e-06,
+      "loss": 2.0389,
+      "step": 278
+    },
+    {
+      "epoch": 0.9425675675675675,
+      "grad_norm": 0.41978840812377416,
+      "learning_rate": 5.4199295036216396e-06,
+      "loss": 2.058,
+      "step": 279
+    },
+    {
+      "epoch": 0.9459459459459459,
+      "grad_norm": 0.36678087910383106,
+      "learning_rate": 5.372111713240805e-06,
+      "loss": 2.0516,
+      "step": 280
+    },
+    {
+      "epoch": 0.9493243243243243,
+      "grad_norm": 0.3872031527878056,
+      "learning_rate": 5.3271608513630705e-06,
+      "loss": 2.0408,
+      "step": 281
+    },
+    {
+      "epoch": 0.9527027027027027,
+      "grad_norm": 0.34292484653055205,
+      "learning_rate": 5.285082741901934e-06,
+      "loss": 2.1431,
+      "step": 282
+    },
+    {
+      "epoch": 0.956081081081081,
+      "grad_norm": 0.3956805273525828,
+      "learning_rate": 5.245882836571982e-06,
+      "loss": 2.0295,
+      "step": 283
+    },
+    {
+      "epoch": 0.9594594594594594,
+      "grad_norm": 0.4535838267215395,
+      "learning_rate": 5.209566214182558e-06,
+      "loss": 2.0857,
+      "step": 284
+    },
+    {
+      "epoch": 0.9628378378378378,
+      "grad_norm": 0.3550925640606362,
+      "learning_rate": 5.176137579979761e-06,
+      "loss": 2.1073,
+      "step": 285
+    },
+    {
+      "epoch": 0.9662162162162162,
+      "grad_norm": 0.4137692684011193,
+      "learning_rate": 5.1456012650368e-06,
+      "loss": 2.0959,
+      "step": 286
+    },
+    {
+      "epoch": 0.9695945945945946,
+      "grad_norm": 0.428088826724393,
+      "learning_rate": 5.117961225692866e-06,
+      "loss": 2.0772,
+      "step": 287
+    },
+    {
+      "epoch": 0.972972972972973,
+      "grad_norm": 0.44380038516311177,
+      "learning_rate": 5.093221043040547e-06,
+      "loss": 2.1148,
+      "step": 288
+    },
+    {
+      "epoch": 0.9763513513513513,
+      "grad_norm": 0.4304760622149439,
+      "learning_rate": 5.071383922461841e-06,
+      "loss": 2.1174,
+      "step": 289
+    },
+    {
+      "epoch": 0.9797297297297297,
+      "grad_norm": 0.3887153913842021,
+      "learning_rate": 5.052452693212867e-06,
+      "loss": 2.1782,
+      "step": 290
+    },
+    {
+      "epoch": 0.9831081081081081,
+      "grad_norm": 0.4408096502071823,
+      "learning_rate": 5.036429808057314e-06,
+      "loss": 1.9739,
+      "step": 291
+    },
+    {
+      "epoch": 0.9864864864864865,
+      "grad_norm": 0.38739812414286,
+      "learning_rate": 5.023317342948631e-06,
+      "loss": 2.0595,
+      "step": 292
+    },
+    {
+      "epoch": 0.9898648648648649,
+      "grad_norm": 0.40388722876679783,
+      "learning_rate": 5.013116996761085e-06,
+      "loss": 1.9943,
+      "step": 293
+    },
+    {
+      "epoch": 0.9932432432432432,
+      "grad_norm": 1.380480700103222,
+      "learning_rate": 5.005830091069644e-06,
+      "loss": 2.1601,
+      "step": 294
+    },
+    {
+      "epoch": 0.9966216216216216,
+      "grad_norm": 0.5043028943315486,
+      "learning_rate": 5.001457569978752e-06,
+      "loss": 2.1212,
+      "step": 295
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.41095001521536584,
+      "learning_rate": 5e-06,
+      "loss": 2.1448,
+      "step": 296
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 296,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 60,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 507967929581568.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-296/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f751e98c8537f36dd5103f084d3b16fe579ab9d618ac216bfedc4e2f22ecfad6
+size 11640

checkpoint-296/zero_to_fp32.py

@@ -0,0 +1,760 @@
+#!/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 gc
+import json
+import numpy as np
+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, weights_only=False)
+
+        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 tqdm(files, desc='Loading checkpoint shards'):
+        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
+        # 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}")
+
+    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] 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")
+
+
+class GatheredTensor:
+    """
+    A pseudo tensor that collects partitioned weights.
+    It is more memory efficient when there are multiple groups.
+    """
+
+    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
+        self.flat_groups = flat_groups
+        self.flat_groups_offset = flat_groups_offset
+        self.offset = offset
+        self.partitioned_numel = partitioned_numel
+        self.shape = shape
+        self.dtype = self.flat_groups[0][0].dtype
+
+    def contiguous(self):
+        """
+        Merge partitioned weights from flat_groups into a single tensor.
+        """
+        end_idx = self.offset + self.partitioned_numel
+        world_size = len(self.flat_groups)
+        pad_flat_param_chunks = []
+
+        for rank_i in range(world_size):
+            # for each rank, we need to collect weights from related group/groups
+            flat_groups_at_rank_i = self.flat_groups[rank_i]
+            start_group_id = None
+            end_group_id = None
+            for group_id in range(len(self.flat_groups_offset)):
+                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
+                    start_group_id = group_id
+                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
+                    end_group_id = group_id
+                    break
+            # collect weights from related group/groups
+            for group_id in range(start_group_id, end_group_id + 1):
+                flat_tensor = flat_groups_at_rank_i[group_id]
+                start_offset = self.offset - self.flat_groups_offset[group_id]
+                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
+                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
+
+        # collect weights from all ranks
+        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
+        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
+        return param
+
+
+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 = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * 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
+    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[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}"
+            )
+
+        # memory efficient tensor
+        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
+        state_dict[name] = tensor
+        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 to_torch_tensor(state_dict, return_empty_tensor=False):
+    """
+    Convert state_dict of GatheredTensor to torch tensor
+    """
+    torch_state_dict = {}
+    converted_tensors = {}
+    for name, tensor in state_dict.items():
+        tensor_id = id(tensor)
+        if tensor_id in converted_tensors:  # shared tensors
+            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
+            torch_state_dict[name] = shared_tensor
+        else:
+            converted_tensors[tensor_id] = name
+            if return_empty_tensor:
+                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
+            else:
+                torch_state_dict[name] = tensor.contiguous()
+    return torch_state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                             tag=None,
+                                             exclude_frozen_parameters=False,
+                                             lazy_mode=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
+        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
+          Convert the pesduo tensor to torch tensor by ``.contiguous()``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    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.
+
+    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
+    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint. Or you can load state_dict in lazy mode ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
+        for name, lazy_tensor in state_dict.item():
+            tensor = lazy_tensor.contiguous()  # to cpu
+            print(name, tensor)
+            # del tensor to release memory if it no longer in use
+    """
+    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")
+
+    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+    if lazy_mode:
+        return state_dict
+    else:
+        return to_torch_tensor(state_dict)
+
+
+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,
+                                                          lazy_mode=True)
+
+    # 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")
+        # an memory-efficient approach for sharding
+        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
+        state_dict_split = split_torch_state_dict_into_shards(empty_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 by shard
+    os.makedirs(output_dir, exist_ok=True)
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
+        shard_state_dict = to_torch_tensor(shard_state_dict)
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard_state_dict, output_path)
+        # release the memory of current shard
+        for tensor_name in list(shard_state_dict.keys()):
+            del state_dict[tensor_name]
+            del shard_state_dict[tensor_name]
+        del shard_state_dict
+        gc.collect()
+
+    # 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)