Upload folder using huggingface_hub

mistral7b_kto_w2_delta0.005/README.md

@@ -0,0 +1,64 @@
+---
+library_name: peft
+license: other
+base_model: mistralai/Mistral-7B-v0.1
+tags:
+- base_model:adapter:mistralai/Mistral-7B-v0.1
+- llama-factory
+- lora
+- transformers
+pipeline_tag: text-generation
+model-index:
+- name: mistral7b_kto_w2_delta0.005
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+# mistral7b_kto_w2_delta0.005
+
+This model is a fine-tuned version of [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) on the shp_kto_train dataset.
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 1e-06
+- train_batch_size: 8
+- eval_batch_size: 8
+- seed: 42
+- distributed_type: multi-GPU
+- num_devices: 8
+- gradient_accumulation_steps: 8
+- total_train_batch_size: 512
+- total_eval_batch_size: 64
+- optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
+- lr_scheduler_type: cosine
+- lr_scheduler_warmup_steps: 100
+- num_epochs: 1.0
+
+### Training results
+
+
+
+### Framework versions
+
+- PEFT 0.17.1
+- Transformers 4.57.1
+- Pytorch 2.10.0+cu128
+- Datasets 4.0.0
+- Tokenizers 0.22.2

mistral7b_kto_w2_delta0.005/adapter_config.json

@@ -0,0 +1,42 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "mistralai/Mistral-7B-v0.1",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "k_proj",
+    "o_proj",
+    "up_proj",
+    "gate_proj",
+    "v_proj",
+    "down_proj",
+    "q_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

mistral7b_kto_w2_delta0.005/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35828ffbc8422a5b814447238f31afc10f35e9d7b503717dc4679511dba37b55
+size 42002584

mistral7b_kto_w2_delta0.005/all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.9280117068971906,
+    "train_runtime": 4303.5052,
+    "train_samples_per_second": 23.237,
+    "train_steps_per_second": 0.046
+}

mistral7b_kto_w2_delta0.005/chat_template.jinja

@@ -0,0 +1,3 @@
+{{ '<s>' }}{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% else %}{% set loop_messages = messages %}{% endif %}{% for message in loop_messages %}{% if loop.index0 == 0 and system_message is defined %}{% set content = system_message + '
+
+' + message['content'] %}{% else %}{% set content = message['content'] %}{% endif %}{% if message['role'] == 'user' %}{{ '[INST] ' + content + '[/INST]' }}{% elif message['role'] == 'assistant' %}{{ ' ' + content + '</s>' }}{% endif %}{% endfor %}

mistral7b_kto_w2_delta0.005/checkpoint-196/README.md

@@ -0,0 +1,208 @@
+---
+base_model: mistralai/Mistral-7B-v0.1
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:mistralai/Mistral-7B-v0.1
+- llama-factory
+- lora
+- transformers
+---
+
+# 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 0.17.1

mistral7b_kto_w2_delta0.005/checkpoint-196/adapter_config.json

@@ -0,0 +1,42 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "mistralai/Mistral-7B-v0.1",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "k_proj",
+    "o_proj",
+    "up_proj",
+    "gate_proj",
+    "v_proj",
+    "down_proj",
+    "q_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

mistral7b_kto_w2_delta0.005/checkpoint-196/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35828ffbc8422a5b814447238f31afc10f35e9d7b503717dc4679511dba37b55
+size 42002584

mistral7b_kto_w2_delta0.005/checkpoint-196/chat_template.jinja

@@ -0,0 +1,3 @@
+{{ '<s>' }}{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% else %}{% set loop_messages = messages %}{% endif %}{% for message in loop_messages %}{% if loop.index0 == 0 and system_message is defined %}{% set content = system_message + '
+
+' + message['content'] %}{% else %}{% set content = message['content'] %}{% endif %}{% if message['role'] == 'user' %}{{ '[INST] ' + content + '[/INST]' }}{% elif message['role'] == 'assistant' %}{{ ' ' + content + '</s>' }}{% endif %}{% endfor %}

mistral7b_kto_w2_delta0.005/checkpoint-196/global_step196/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1dd9bafbfae251ae3887812445b6d1daf17fb951a59ca01e7e43bfb9c65c86d7
+size 31467333

mistral7b_kto_w2_delta0.005/checkpoint-196/global_step196/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:988641547a6d6865903758927ba3de7f6963f555e5c6ffccc3c0924e1c6fabf7
+size 31467333

mistral7b_kto_w2_delta0.005/checkpoint-196/global_step196/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:d3e75cea27fdaeb8b2e2cc1e61bef2929cb0df28071cd59ddea19185597073cd
+size 31467333

mistral7b_kto_w2_delta0.005/checkpoint-196/global_step196/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:af070a2306c428d3dd2f4af03cc5ea52ab5ad8d2b6f076bd6485f011ca15875b
+size 31467397

mistral7b_kto_w2_delta0.005/checkpoint-196/global_step196/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c11dcbdc28de6d1e5f62f9544ccd48016ae5d6de4653588204d4bc4eb419be3f
+size 31467397

mistral7b_kto_w2_delta0.005/checkpoint-196/global_step196/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e00bb3e28e4a37b92f39c543edc9bf8e8972534bbd2351b821e1746eb053eacc
+size 31467397

mistral7b_kto_w2_delta0.005/checkpoint-196/global_step196/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9f7555cc3a3ccb6239096d6eb5af71b0a8707b2f382e15f3c3833cf72661d8f
+size 31467397

mistral7b_kto_w2_delta0.005/checkpoint-196/global_step196/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:283a047c63142ff64bc8d3b732cfb22dc08e4c76c170ab7fe4f6a560f726eea6
+size 31467397

mistral7b_kto_w2_delta0.005/checkpoint-196/global_step196/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b9280d068de7b25f4e4c3ce5c5b53d9879cb10a0f185c967578b11ddcaa0c05
+size 42288409

mistral7b_kto_w2_delta0.005/checkpoint-196/latest

@@ -0,0 +1 @@
+global_step196

mistral7b_kto_w2_delta0.005/checkpoint-196/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7efbc3b39e557ed047a255851058c3d62216b6bc5794c19289879cf099f52035
+size 16389

mistral7b_kto_w2_delta0.005/checkpoint-196/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce29a8767a7d907dd24987aa2c3e654d4317f3042fbc13b5b72cadb46d43311a
+size 16389

mistral7b_kto_w2_delta0.005/checkpoint-196/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61a48db011646b4e9a867bf12f4a233cad5dfbfe309686f8996c250196d3783a
+size 16389

mistral7b_kto_w2_delta0.005/checkpoint-196/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9562ee822472a4f01dcd6349ab3d1ef42a48915fe3b92e843a0c37db53c8421
+size 16389

mistral7b_kto_w2_delta0.005/checkpoint-196/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7d2767d83c3bf27f12db022b0632e2c4f8c164274ba75e380cf18f9d5f21819
+size 16389

mistral7b_kto_w2_delta0.005/checkpoint-196/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76816358d4e5db8149d60d55234db658d67a13c0c1ce05d7404cf7125a676a5c
+size 16389

mistral7b_kto_w2_delta0.005/checkpoint-196/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1562e7520c977d178183d641f70abcf3f57da2489938756cfbebf9b6e6c1a9fd
+size 16389

mistral7b_kto_w2_delta0.005/checkpoint-196/rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6b6cabaed045c5398cd1b732f7ec48bd363f3b43cd24e0e70e641a42bd00c28
+size 16389

mistral7b_kto_w2_delta0.005/checkpoint-196/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83820952e8eb59075696a2d6e1f7ef27d70c5576cb2781806606f38f86685d3e
+size 1401

mistral7b_kto_w2_delta0.005/checkpoint-196/special_tokens_map.json

@@ -0,0 +1,24 @@
+{
+  "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": "</s>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

mistral7b_kto_w2_delta0.005/checkpoint-196/tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dadfd56d766715c61d2ef780a525ab43b8e6da4de6865bda3d95fdef5e134055
+size 493443

mistral7b_kto_w2_delta0.005/checkpoint-196/tokenizer_config.json

@@ -0,0 +1,46 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [],
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "extra_special_tokens": {},
+  "legacy": false,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "</s>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "split_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

mistral7b_kto_w2_delta0.005/checkpoint-196/trainer_state.json

@@ -0,0 +1,965 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 196,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.025591810620601407,
+      "grad_norm": 0.577160120010376,
+      "kl": 0.008340898901224136,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59718413.473684214,
+      "logits/rejected": -50413519.23809524,
+      "logps/chosen": -439.03885690789474,
+      "logps/rejected": -198.42229352678572,
+      "loss": 0.9479,
+      "loss/base_kto": 0.4997499883174896,
+      "metrics/advantage_var": 0.052478689700365067,
+      "regularization/grad_norm_sq": 4334.98388671875,
+      "regularization/w2": 0.45370370149612427,
+      "rewards/chosen": 0.0006774790388973136,
+      "rewards/margins": 0.0019659657722203654,
+      "rewards/rejected": -0.0012884867333230517,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.611538290977478,
+      "kl": 0.016795814037322998,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59302692.571428575,
+      "logits/rejected": -50021761.54216868,
+      "logps/chosen": -429.92659293831167,
+      "logps/rejected": -192.85163309487953,
+      "loss": 0.9286,
+      "loss/base_kto": 0.5000812411308289,
+      "metrics/advantage_var": 0.0676804929971695,
+      "regularization/grad_norm_sq": 3756.94775390625,
+      "regularization/w2": 0.4279446601867676,
+      "rewards/chosen": 0.0011910808744368614,
+      "rewards/margins": -0.00043923452559481435,
+      "rewards/rejected": 0.0016303154000316758,
+      "stability/repetition_rate_mean": 0.3492690920829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.3662109375,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.6086092591285706,
+      "kl": 0.020404411479830742,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56236609.163636364,
+      "logits/rejected": -51916776.87741935,
+      "logps/chosen": -404.12135416666666,
+      "logps/rejected": -200.67870463709679,
+      "loss": 0.9272,
+      "loss/base_kto": 0.49999886751174927,
+      "metrics/advantage_var": 0.05835923179984093,
+      "regularization/grad_norm_sq": 4170.2373046875,
+      "regularization/w2": 0.44162631034851074,
+      "rewards/chosen": 0.0009465837117397424,
+      "rewards/margins": -0.0005884172047105119,
+      "rewards/rejected": 0.0015350009164502543,
+      "stability/repetition_rate_mean": 0.28515625,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.92341423034668,
+      "stability/response_length_var": 895.4107055664062,
+      "stability/token_entropy_mean": 3.7548828125,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.6190893650054932,
+      "kl": 0.021697884425520897,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51675481.6,
+      "logits/rejected": -48057888.0,
+      "logps/chosen": -343.6622802734375,
+      "logps/rejected": -225.616259765625,
+      "loss": 0.9324,
+      "loss/base_kto": 0.5001762509346008,
+      "metrics/advantage_var": 0.06153888627886772,
+      "regularization/grad_norm_sq": 3680.827392578125,
+      "regularization/w2": 0.42263123393058777,
+      "rewards/chosen": 0.0034450016915798186,
+      "rewards/margins": -0.001264280080795288,
+      "rewards/rejected": 0.004709281772375107,
+      "stability/repetition_rate_mean": 0.2096664160490036,
+      "stability/response_length_mean": 44.0,
+      "stability/response_length_std": 28.0611572265625,
+      "stability/response_length_var": 787.4285888671875,
+      "stability/token_entropy_mean": 4.42578125,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.5808376669883728,
+      "kl": 0.018133217468857765,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56471566.12413793,
+      "logits/rejected": -52024934.4,
+      "logps/chosen": -378.4438577586207,
+      "logps/rejected": -202.7503125,
+      "loss": 0.9391,
+      "loss/base_kto": 0.49999186396598816,
+      "metrics/advantage_var": 0.06101418659090996,
+      "regularization/grad_norm_sq": 4002.333984375,
+      "regularization/w2": 0.4309103488922119,
+      "rewards/chosen": 0.0047595426954072095,
+      "rewards/margins": 0.0005058412129068605,
+      "rewards/rejected": 0.004253701482500349,
+      "stability/repetition_rate_mean": 0.18583066761493683,
+      "stability/response_length_mean": 43.375,
+      "stability/response_length_std": 28.54538917541504,
+      "stability/response_length_var": 814.8392944335938,
+      "stability/token_entropy_mean": 4.75390625,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.6033640503883362,
+      "kl": 0.022612739354372025,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57729188.939597316,
+      "logits/rejected": -51520841.35672515,
+      "logps/chosen": -468.5989932885906,
+      "logps/rejected": -198.2083561769006,
+      "loss": 0.9337,
+      "loss/base_kto": 0.5006771683692932,
+      "metrics/advantage_var": 0.09293336421251297,
+      "regularization/grad_norm_sq": 4491.3251953125,
+      "regularization/w2": 0.45624828338623047,
+      "rewards/chosen": -0.0017742578615278207,
+      "rewards/margins": -0.005727118693742176,
+      "rewards/rejected": 0.0039528608322143555,
+      "stability/repetition_rate_mean": 0.1583891361951828,
+      "stability/response_length_mean": 43.0,
+      "stability/response_length_std": 29.602123260498047,
+      "stability/response_length_var": 876.2857055664062,
+      "stability/token_entropy_mean": 4.642578125,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6803989410400391,
+      "kl": 0.02071223594248295,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57644589.36708861,
+      "logits/rejected": -53911292.83950617,
+      "logps/chosen": -385.9240506329114,
+      "logps/rejected": -195.117862654321,
+      "loss": 0.938,
+      "loss/base_kto": 0.4999936521053314,
+      "metrics/advantage_var": 0.07669314742088318,
+      "regularization/grad_norm_sq": 3758.585205078125,
+      "regularization/w2": 0.4280441403388977,
+      "rewards/chosen": 0.001247800886631012,
+      "rewards/margins": 0.0003867014138786881,
+      "rewards/rejected": 0.0008610994727523239,
+      "stability/repetition_rate_mean": 0.3108258843421936,
+      "stability/response_length_mean": 50.5,
+      "stability/response_length_std": 25.048524856567383,
+      "stability/response_length_var": 627.4285888671875,
+      "stability/token_entropy_mean": 5.58203125,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.5847389698028564,
+      "kl": 0.02293821983039379,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56634060.8,
+      "logits/rejected": -50219801.6,
+      "logps/chosen": -421.899560546875,
+      "logps/rejected": -219.2647705078125,
+      "loss": 0.9341,
+      "loss/base_kto": 0.5005573630332947,
+      "metrics/advantage_var": 0.06820251047611237,
+      "regularization/grad_norm_sq": 5033.30859375,
+      "regularization/w2": 0.4727831482887268,
+      "rewards/chosen": 0.0003104138188064098,
+      "rewards/margins": -0.0041379273869097235,
+      "rewards/rejected": 0.004448341205716133,
+      "stability/repetition_rate_mean": 0.16694079339504242,
+      "stability/response_length_mean": 35.0,
+      "stability/response_length_std": 31.518701553344727,
+      "stability/response_length_var": 993.4285888671875,
+      "stability/token_entropy_mean": 4.89453125,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.6400073766708374,
+      "kl": 0.0444142147898674,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52707451.70469799,
+      "logits/rejected": -50772524.9122807,
+      "logps/chosen": -352.2749056208054,
+      "logps/rejected": -243.8424707602339,
+      "loss": 0.9389,
+      "loss/base_kto": 0.4996757507324219,
+      "metrics/advantage_var": 0.06386065483093262,
+      "regularization/grad_norm_sq": 3936.931640625,
+      "regularization/w2": 0.42452260851860046,
+      "rewards/chosen": 0.0053554469306997,
+      "rewards/margins": 0.0021862895790329766,
+      "rewards/rejected": 0.0031691573516667237,
+      "stability/repetition_rate_mean": 0.1052611693739891,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.87324333190918,
+      "stability/response_length_var": 892.4107055664062,
+      "stability/token_entropy_mean": 4.771484375,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.5788847208023071,
+      "kl": 0.04537983611226082,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54037710.17449664,
+      "logits/rejected": -48296085.70760234,
+      "logps/chosen": -375.1397336409396,
+      "logps/rejected": -180.18807108918128,
+      "loss": 0.9201,
+      "loss/base_kto": 0.49983420968055725,
+      "metrics/advantage_var": 0.052370816469192505,
+      "regularization/grad_norm_sq": 3871.12744140625,
+      "regularization/w2": 0.4266422390937805,
+      "rewards/chosen": 0.005941886629834271,
+      "rewards/margins": 0.0006887215409353172,
+      "rewards/rejected": 0.005253165088898954,
+      "stability/repetition_rate_mean": 0.3655294179916382,
+      "stability/response_length_mean": 53.375,
+      "stability/response_length_std": 21.480472564697266,
+      "stability/response_length_var": 461.41070556640625,
+      "stability/token_entropy_mean": 3.8212890625,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.6621019244194031,
+      "kl": 0.05794268846511841,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56953277.217391305,
+      "logits/rejected": -49571939.82389937,
+      "logps/chosen": -344.48738354037266,
+      "logps/rejected": -230.3876523191824,
+      "loss": 0.9291,
+      "loss/base_kto": 0.5004059076309204,
+      "metrics/advantage_var": 0.05569034814834595,
+      "regularization/grad_norm_sq": 3976.169921875,
+      "regularization/w2": 0.43878814578056335,
+      "rewards/chosen": 0.0024627199084121987,
+      "rewards/margins": -0.00357790456916621,
+      "rewards/rejected": 0.006040624477578409,
+      "stability/repetition_rate_mean": 0.10418051481246948,
+      "stability/response_length_mean": 22.375,
+      "stability/response_length_std": 30.236862182617188,
+      "stability/response_length_var": 914.2678833007812,
+      "stability/token_entropy_mean": 4.3173828125,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.5812336206436157,
+      "kl": 0.05679145082831383,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52678345.696969695,
+      "logits/rejected": -47821840.51612903,
+      "logps/chosen": -359.6124053030303,
+      "logps/rejected": -178.17867943548387,
+      "loss": 0.933,
+      "loss/base_kto": 0.4992280900478363,
+      "metrics/advantage_var": 0.057238996028900146,
+      "regularization/grad_norm_sq": 3165.790283203125,
+      "regularization/w2": 0.3937515616416931,
+      "rewards/chosen": 0.010837621399850556,
+      "rewards/margins": 0.00601367213975416,
+      "rewards/rejected": 0.004823949260096396,
+      "stability/repetition_rate_mean": 0.22545291483402252,
+      "stability/response_length_mean": 44.625,
+      "stability/response_length_std": 27.8820858001709,
+      "stability/response_length_var": 777.4107055664062,
+      "stability/token_entropy_mean": 4.646484375,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.5738043785095215,
+      "kl": 0.06266746670007706,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57563182.24516129,
+      "logits/rejected": -47268345.7939394,
+      "logps/chosen": -381.86370967741937,
+      "logps/rejected": -183.5180516098485,
+      "loss": 0.939,
+      "loss/base_kto": 0.4997069537639618,
+      "metrics/advantage_var": 0.07446863502264023,
+      "regularization/grad_norm_sq": 3868.786865234375,
+      "regularization/w2": 0.4315285384654999,
+      "rewards/chosen": 0.008308312200730847,
+      "rewards/margins": 0.001878772895939772,
+      "rewards/rejected": 0.006429539304791075,
+      "stability/repetition_rate_mean": 0.19313038885593414,
+      "stability/response_length_mean": 20.625,
+      "stability/response_length_std": 27.176342010498047,
+      "stability/response_length_var": 738.5535888671875,
+      "stability/token_entropy_mean": 5.55859375,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5531291365623474,
+      "kl": 0.08709845691919327,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52753698.68387097,
+      "logits/rejected": -49023162.18181818,
+      "logps/chosen": -378.9974798387097,
+      "logps/rejected": -220.35189393939393,
+      "loss": 0.9497,
+      "loss/base_kto": 0.49925193190574646,
+      "metrics/advantage_var": 0.08100283890962601,
+      "regularization/grad_norm_sq": 4483.40087890625,
+      "regularization/w2": 0.45682960748672485,
+      "rewards/chosen": 0.01309710318042386,
+      "rewards/margins": 0.006069044376981108,
+      "rewards/rejected": 0.007028058803442753,
+      "stability/repetition_rate_mean": 0.111328125,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.3388671875,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.611277163028717,
+      "kl": 0.09017860144376755,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49498509.83783784,
+      "logits/rejected": -47885377.488372095,
+      "logps/chosen": -314.13988597972974,
+      "logps/rejected": -228.78901707848837,
+      "loss": 0.9402,
+      "loss/base_kto": 0.49950870871543884,
+      "metrics/advantage_var": 0.06859137862920761,
+      "regularization/grad_norm_sq": 3736.681396484375,
+      "regularization/w2": 0.42404261231422424,
+      "rewards/chosen": 0.013092591955855087,
+      "rewards/margins": 0.0035108895627003218,
+      "rewards/rejected": 0.009581702393154765,
+      "stability/repetition_rate_mean": 0.10066105425357819,
+      "stability/response_length_mean": 22.75,
+      "stability/response_length_std": 31.07479476928711,
+      "stability/response_length_var": 965.6428833007812,
+      "stability/token_entropy_mean": 2.130859375,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.6037386655807495,
+      "kl": 0.13238738477230072,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58626637.98675497,
+      "logits/rejected": -53447752.71005917,
+      "logps/chosen": -403.87065397350995,
+      "logps/rejected": -200.96398853550295,
+      "loss": 0.9316,
+      "loss/base_kto": 0.4992172420024872,
+      "metrics/advantage_var": 0.07778961956501007,
+      "regularization/grad_norm_sq": 3624.650390625,
+      "regularization/w2": 0.42037779092788696,
+      "rewards/chosen": 0.0159459240389186,
+      "rewards/margins": 0.006766194357169991,
+      "rewards/rejected": 0.00917972968174861,
+      "stability/repetition_rate_mean": 0.19088542461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.23828125,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.6116781234741211,
+      "kl": 0.12365061044692993,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51478686.55483871,
+      "logits/rejected": -47141816.63030303,
+      "logps/chosen": -358.13480342741934,
+      "logps/rejected": -230.34330018939394,
+      "loss": 0.9311,
+      "loss/base_kto": 0.4985641539096832,
+      "metrics/advantage_var": 0.10404322296380997,
+      "regularization/grad_norm_sq": 3923.891357421875,
+      "regularization/w2": 0.43580102920532227,
+      "rewards/chosen": 0.01977017310357863,
+      "rewards/margins": 0.011619663471583632,
+      "rewards/rejected": 0.008150509631994998,
+      "stability/repetition_rate_mean": 0.103515625,
+      "stability/response_length_mean": 32.5,
+      "stability/response_length_std": 33.67915725708008,
+      "stability/response_length_var": 1134.2857666015625,
+      "stability/token_entropy_mean": 4.9375,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.6381218433380127,
+      "kl": 0.16665403544902802,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58412438.974358976,
+      "logits/rejected": -50588893.65853658,
+      "logps/chosen": -380.5612980769231,
+      "logps/rejected": -187.1324433117378,
+      "loss": 0.9286,
+      "loss/base_kto": 0.4982132017612457,
+      "metrics/advantage_var": 0.08798178285360336,
+      "regularization/grad_norm_sq": 3925.050048828125,
+      "regularization/w2": 0.43326812982559204,
+      "rewards/chosen": 0.022810822878128443,
+      "rewards/margins": 0.013375084872839225,
+      "rewards/rejected": 0.009435738005289218,
+      "stability/repetition_rate_mean": 0.3356415629386902,
+      "stability/response_length_mean": 30.375,
+      "stability/response_length_std": 29.061203002929688,
+      "stability/response_length_var": 844.5535888671875,
+      "stability/token_entropy_mean": 3.72265625,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.6163039803504944,
+      "kl": 0.15967555344104767,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50379520.0,
+      "logits/rejected": -48967028.938271604,
+      "logps/chosen": -314.77383306962025,
+      "logps/rejected": -207.04853877314815,
+      "loss": 0.9271,
+      "loss/base_kto": 0.4984891414642334,
+      "metrics/advantage_var": 0.1260511875152588,
+      "regularization/grad_norm_sq": 3692.677490234375,
+      "regularization/w2": 0.42039623856544495,
+      "rewards/chosen": 0.021952728681926485,
+      "rewards/margins": 0.012433580661829717,
+      "rewards/rejected": 0.009519148020096768,
+      "stability/repetition_rate_mean": 0.19374233484268188,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 31.1264705657959,
+      "stability/response_length_var": 968.8571166992188,
+      "stability/token_entropy_mean": 3.75,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.6088441014289856,
+      "kl": 0.1950296312570572,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54698091.24324324,
+      "logits/rejected": -50908255.255813956,
+      "logps/chosen": -359.5659311655405,
+      "logps/rejected": -216.3930323401163,
+      "loss": 0.9328,
+      "loss/base_kto": 0.49854350090026855,
+      "metrics/advantage_var": 0.13550297915935516,
+      "regularization/grad_norm_sq": 3865.565673828125,
+      "regularization/w2": 0.42957553267478943,
+      "rewards/chosen": 0.03042653444651011,
+      "rewards/margins": 0.011843225627781683,
+      "rewards/rejected": 0.018583308818728425,
+      "stability/repetition_rate_mean": 0.228515625,
+      "stability/response_length_mean": 41.25,
+      "stability/response_length_std": 31.476749420166016,
+      "stability/response_length_var": 990.7857055664062,
+      "stability/token_entropy_mean": 4.078125,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6745646595954895,
+      "kl": 0.21558904647827148,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59378317.87951807,
+      "logits/rejected": -53788419.32467532,
+      "logps/chosen": -399.4925169427711,
+      "logps/rejected": -209.84361049107142,
+      "loss": 0.9365,
+      "loss/base_kto": 0.49687233567237854,
+      "metrics/advantage_var": 0.2065424919128418,
+      "regularization/grad_norm_sq": 3994.716064453125,
+      "regularization/w2": 0.4382788836956024,
+      "rewards/chosen": 0.04099334004413651,
+      "rewards/margins": 0.024971716804247736,
+      "rewards/rejected": 0.016021623239888774,
+      "stability/repetition_rate_mean": 0.1877790093421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.353515625,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.6502033472061157,
+      "kl": 0.26441773772239685,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57391930.60240964,
+      "logits/rejected": -53613953.66233766,
+      "logps/chosen": -378.218938253012,
+      "logps/rejected": -205.88195008116884,
+      "loss": 0.9259,
+      "loss/base_kto": 0.49810346961021423,
+      "metrics/advantage_var": 0.21781854331493378,
+      "regularization/grad_norm_sq": 3637.52197265625,
+      "regularization/w2": 0.42161688208580017,
+      "rewards/chosen": 0.0344322733132236,
+      "rewards/margins": 0.013813094566021703,
+      "rewards/rejected": 0.020619178747201896,
+      "stability/repetition_rate_mean": 0.30853456258773804,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.911476135253906,
+      "stability/response_length_var": 894.6964111328125,
+      "stability/token_entropy_mean": 4.404296875,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.6496742367744446,
+      "kl": 0.3085666298866272,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60598233.11392405,
+      "logits/rejected": -53416372.14814815,
+      "logps/chosen": -364.10173556170884,
+      "logps/rejected": -200.97486255787038,
+      "loss": 0.9207,
+      "loss/base_kto": 0.4965952932834625,
+      "metrics/advantage_var": 0.25335976481437683,
+      "regularization/grad_norm_sq": 3352.978271484375,
+      "regularization/w2": 0.4012783169746399,
+      "rewards/chosen": 0.04908096337620216,
+      "rewards/margins": 0.02691114084816366,
+      "rewards/rejected": 0.022169822528038497,
+      "stability/repetition_rate_mean": 0.2194010466337204,
+      "stability/response_length_mean": 49.25,
+      "stability/response_length_std": 27.316923141479492,
+      "stability/response_length_var": 746.2142944335938,
+      "stability/token_entropy_mean": 4.75,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.6193932294845581,
+      "kl": 0.3443666994571686,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57739732.49673203,
+      "logits/rejected": -54352828.5508982,
+      "logps/chosen": -383.7316687091503,
+      "logps/rejected": -211.93726609281438,
+      "loss": 0.936,
+      "loss/base_kto": 0.49703770875930786,
+      "metrics/advantage_var": 0.3692956864833832,
+      "regularization/grad_norm_sq": 4454.72216796875,
+      "regularization/w2": 0.4518660008907318,
+      "rewards/chosen": 0.05132290896247415,
+      "rewards/margins": 0.023414049823756954,
+      "rewards/rejected": 0.027908859138717195,
+      "stability/repetition_rate_mean": 0.15568576753139496,
+      "stability/response_length_mean": 27.625,
+      "stability/response_length_std": 30.486238479614258,
+      "stability/response_length_var": 929.4107055664062,
+      "stability/token_entropy_mean": 4.693359375,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.6110739707946777,
+      "kl": 0.3985693156719208,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60667195.61643836,
+      "logits/rejected": -55641405.79310345,
+      "logps/chosen": -416.79403895547944,
+      "logps/rejected": -216.0777658045977,
+      "loss": 0.9231,
+      "loss/base_kto": 0.49641838669776917,
+      "metrics/advantage_var": 0.559248149394989,
+      "regularization/grad_norm_sq": 3774.068115234375,
+      "regularization/w2": 0.4254666864871979,
+      "rewards/chosen": 0.05853001058918156,
+      "rewards/margins": 0.029587951762866566,
+      "rewards/rejected": 0.028942058826314992,
+      "stability/repetition_rate_mean": 0.19144965708255768,
+      "stability/response_length_mean": 23.125,
+      "stability/response_length_std": 29.347366333007812,
+      "stability/response_length_var": 861.2678833007812,
+      "stability/token_entropy_mean": 3.353515625,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.5931130051612854,
+      "kl": 0.45674076676368713,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57375400.464516126,
+      "logits/rejected": -53823680.38787879,
+      "logps/chosen": -391.9906754032258,
+      "logps/rejected": -183.6497159090909,
+      "loss": 0.9394,
+      "loss/base_kto": 0.49498748779296875,
+      "metrics/advantage_var": 0.7409353256225586,
+      "regularization/grad_norm_sq": 3780.29150390625,
+      "regularization/w2": 0.42766115069389343,
+      "rewards/chosen": 0.06672527559341923,
+      "rewards/margins": 0.039806725622272214,
+      "rewards/rejected": 0.026918549971147017,
+      "stability/repetition_rate_mean": 0.24460937082767487,
+      "stability/response_length_mean": 43.25,
+      "stability/response_length_std": 29.620214462280273,
+      "stability/response_length_var": 877.3571166992188,
+      "stability/token_entropy_mean": 3.640625,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5910015106201172,
+      "kl": 0.5136879086494446,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59843408.09815951,
+      "logits/rejected": -55000846.67515924,
+      "logps/chosen": -400.98552530674846,
+      "logps/rejected": -247.51219148089172,
+      "loss": 0.9147,
+      "loss/base_kto": 0.49399566650390625,
+      "metrics/advantage_var": 0.9681418538093567,
+      "regularization/grad_norm_sq": 4315.08056640625,
+      "regularization/w2": 0.4519772231578827,
+      "rewards/chosen": 0.0830619554578161,
+      "rewards/margins": 0.04720748222231525,
+      "rewards/rejected": 0.035854473235500846,
+      "stability/repetition_rate_mean": 0.2230902761220932,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.936838150024414,
+      "stability/response_length_var": 896.2142944335938,
+      "stability/token_entropy_mean": 4.05078125,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.7235668301582336,
+      "kl": 0.4908495843410492,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57076500.16969697,
+      "logits/rejected": -50077570.477419354,
+      "logps/chosen": -366.3874053030303,
+      "logps/rejected": -189.92144657258063,
+      "loss": 0.93,
+      "loss/base_kto": 0.49192824959754944,
+      "metrics/advantage_var": 1.9476429224014282,
+      "regularization/grad_norm_sq": 5980.30029296875,
+      "regularization/w2": 0.48489466309547424,
+      "rewards/chosen": 0.08023344097715436,
+      "rewards/margins": 0.06893024071686896,
+      "rewards/rejected": 0.011303200260285409,
+      "stability/repetition_rate_mean": 0.302734375,
+      "stability/response_length_mean": 56.25,
+      "stability/response_length_std": 21.920310974121094,
+      "stability/response_length_var": 480.5,
+      "stability/token_entropy_mean": 5.001953125,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.6801520586013794,
+      "kl": 0.577669084072113,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51146073.44578313,
+      "logits/rejected": -51662515.53246753,
+      "logps/chosen": -329.4719267695783,
+      "logps/rejected": -199.15390371347402,
+      "loss": 0.9176,
+      "loss/base_kto": 0.49409326910972595,
+      "metrics/advantage_var": 0.8393446207046509,
+      "regularization/grad_norm_sq": 3416.243408203125,
+      "regularization/w2": 0.40856075286865234,
+      "rewards/chosen": 0.0799884566341538,
+      "rewards/margins": 0.046587763137053366,
+      "rewards/rejected": 0.03340069349710043,
+      "stability/repetition_rate_mean": 0.21016408503055573,
+      "stability/response_length_mean": 22.625,
+      "stability/response_length_std": 26.359804153442383,
+      "stability/response_length_var": 694.8392944335938,
+      "stability/token_entropy_mean": 5.66015625,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.6449370384216309,
+      "kl": 0.6454542875289917,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55421090.63529412,
+      "logits/rejected": -50187294.72,
+      "logps/chosen": -396.8174632352941,
+      "logps/rejected": -207.35263020833332,
+      "loss": 0.923,
+      "loss/base_kto": 0.4903319776058197,
+      "metrics/advantage_var": 1.2783325910568237,
+      "regularization/grad_norm_sq": 4074.032470703125,
+      "regularization/w2": 0.4475216865539551,
+      "rewards/chosen": 0.11169673695283777,
+      "rewards/margins": 0.0743670695435767,
+      "rewards/rejected": 0.037329667409261066,
+      "stability/repetition_rate_mean": 0.23066024482250214,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.193626403808594,
+      "stability/response_length_var": 852.2678833007812,
+      "stability/token_entropy_mean": 4.6015625,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.6441305875778198,
+      "kl": 0.6456610560417175,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54003495.152941175,
+      "logits/rejected": -49346880.85333333,
+      "logps/chosen": -360.09657628676473,
+      "logps/rejected": -191.00634114583335,
+      "loss": 0.9163,
+      "loss/base_kto": 0.4928789734840393,
+      "metrics/advantage_var": 2.1478357315063477,
+      "regularization/grad_norm_sq": 4929.31494140625,
+      "regularization/w2": 0.46441006660461426,
+      "rewards/chosen": 0.08927134345559513,
+      "rewards/margins": 0.059344200994454185,
+      "rewards/rejected": 0.029927142461140952,
+      "stability/repetition_rate_mean": 0.2978527545928955,
+      "stability/response_length_mean": 58.125,
+      "stability/response_length_std": 10.960155487060547,
+      "stability/response_length_var": 120.125,
+      "stability/token_entropy_mean": 4.56640625,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.6161492466926575,
+      "kl": 0.7202507257461548,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55241313.52380952,
+      "logits/rejected": -48956978.526315786,
+      "logps/chosen": -415.5804501488095,
+      "logps/rejected": -186.47481496710526,
+      "loss": 0.9248,
+      "loss/base_kto": 0.48859721422195435,
+      "metrics/advantage_var": 1.9541778564453125,
+      "regularization/grad_norm_sq": 4000.94580078125,
+      "regularization/w2": 0.43814727663993835,
+      "rewards/chosen": 0.1314907301039923,
+      "rewards/margins": 0.08923867576403129,
+      "rewards/rejected": 0.042252054339961,
+      "stability/repetition_rate_mean": 0.31046873331069946,
+      "stability/response_length_mean": 32.5,
+      "stability/response_length_std": 30.528675079345703,
+      "stability/response_length_var": 932.0,
+      "stability/token_entropy_mean": 4.3720703125,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.7250444293022156,
+      "kl": 0.7532386779785156,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54405402.48275862,
+      "logits/rejected": -49781749.47945205,
+      "logps/chosen": -369.750628591954,
+      "logps/rejected": -212.1415881849315,
+      "loss": 0.9192,
+      "loss/base_kto": 0.49109888076782227,
+      "metrics/advantage_var": 1.7532415390014648,
+      "regularization/grad_norm_sq": 3549.75244140625,
+      "regularization/w2": 0.4151502549648285,
+      "rewards/chosen": 0.11661315786427465,
+      "rewards/margins": 0.07156126962048184,
+      "rewards/rejected": 0.04505188824379281,
+      "stability/repetition_rate_mean": 0.22421875596046448,
+      "stability/response_length_mean": 19.875,
+      "stability/response_length_std": 27.771709442138672,
+      "stability/response_length_var": 771.2678833007812,
+      "stability/token_entropy_mean": 4.890625,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.6213644742965698,
+      "kl": 0.9238977432250977,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59125432.80473373,
+      "logits/rejected": -52567887.682119206,
+      "logps/chosen": -439.6600406804734,
+      "logps/rejected": -190.58578228476821,
+      "loss": 0.905,
+      "loss/base_kto": 0.4874550998210907,
+      "metrics/advantage_var": 2.416893720626831,
+      "regularization/grad_norm_sq": 3930.178955078125,
+      "regularization/w2": 0.4287300109863281,
+      "rewards/chosen": 0.16818395309899686,
+      "rewards/margins": 0.10098015347862745,
+      "rewards/rejected": 0.06720379962036942,
+      "stability/repetition_rate_mean": 0.23209978640079498,
+      "stability/response_length_mean": 36.25,
+      "stability/response_length_std": 30.76988410949707,
+      "stability/response_length_var": 946.7857055664062,
+      "stability/token_entropy_mean": 3.67822265625,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.7216498851776123,
+      "kl": 0.9429834485054016,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56652960.38554217,
+      "logits/rejected": -48022956.883116886,
+      "logps/chosen": -354.89241340361446,
+      "logps/rejected": -194.30808644480518,
+      "loss": 0.911,
+      "loss/base_kto": 0.48693132400512695,
+      "metrics/advantage_var": 4.248536586761475,
+      "regularization/grad_norm_sq": 4139.484375,
+      "regularization/w2": 0.4321208894252777,
+      "rewards/chosen": 0.15448007238916603,
+      "rewards/margins": 0.1045362072695292,
+      "rewards/rejected": 0.04994386511963683,
+      "stability/repetition_rate_mean": 0.1976061761379242,
+      "stability/response_length_mean": 30.5,
+      "stability/response_length_std": 26.42509651184082,
+      "stability/response_length_var": 698.2857055664062,
+      "stability/token_entropy_mean": 4.8369140625,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.6428859233856201,
+      "kl": 1.0304526090621948,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52658986.40490798,
+      "logits/rejected": -47980801.63057325,
+      "logps/chosen": -374.418615797546,
+      "logps/rejected": -210.53871417197453,
+      "loss": 0.9028,
+      "loss/base_kto": 0.4877626597881317,
+      "metrics/advantage_var": 4.026914119720459,
+      "regularization/grad_norm_sq": 3664.698974609375,
+      "regularization/w2": 0.41559115052223206,
+      "rewards/chosen": 0.1727198852351838,
+      "rewards/margins": 0.09677210547646317,
+      "rewards/rejected": 0.07594777975872063,
+      "stability/repetition_rate_mean": 0.25664061307907104,
+      "stability/response_length_mean": 34.75,
+      "stability/response_length_std": 31.54928207397461,
+      "stability/response_length_var": 995.3571166992188,
+      "stability/token_entropy_mean": 3.3203125,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.6485155820846558,
+      "kl": 1.06327223777771,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54608421.925925925,
+      "logits/rejected": -49521022.37974683,
+      "logps/chosen": -310.32325424382714,
+      "logps/rejected": -196.57655508306962,
+      "loss": 0.9317,
+      "loss/base_kto": 0.48661017417907715,
+      "metrics/advantage_var": 7.680975437164307,
+      "regularization/grad_norm_sq": 7077.32763671875,
+      "regularization/w2": 0.47685596346855164,
+      "rewards/chosen": 0.16781084037121433,
+      "rewards/margins": 0.12155084260498517,
+      "rewards/rejected": 0.04625999776622917,
+      "stability/repetition_rate_mean": 0.19140625,
+      "stability/response_length_mean": 32.875,
+      "stability/response_length_std": 33.30996322631836,
+      "stability/response_length_var": 1109.5535888671875,
+      "stability/token_entropy_mean": 3.701171875,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.6963216662406921,
+      "kl": 1.1201375722885132,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53089470.08383234,
+      "logits/rejected": -49691681.46405229,
+      "logps/chosen": -390.30389221556885,
+      "logps/rejected": -210.46195874183007,
+      "loss": 0.9237,
+      "loss/base_kto": 0.4832589626312256,
+      "metrics/advantage_var": 7.02877950668335,
+      "regularization/grad_norm_sq": 5314.27880859375,
+      "regularization/w2": 0.4580075442790985,
+      "rewards/chosen": 0.2101439972837528,
+      "rewards/margins": 0.1364806098605055,
+      "rewards/rejected": 0.0736633874232473,
+      "stability/repetition_rate_mean": 0.1987847238779068,
+      "stability/response_length_mean": 19.125,
+      "stability/response_length_std": 28.296579360961914,
+      "stability/response_length_var": 800.6964111328125,
+      "stability/token_entropy_mean": 4.759765625,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.6389023065567017,
+      "kl": 1.1507009267807007,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60032350.14193548,
+      "logits/rejected": -53812130.90909091,
+      "logps/chosen": -390.8807459677419,
+      "logps/rejected": -197.07857481060606,
+      "loss": 0.9173,
+      "loss/base_kto": 0.4865249693393707,
+      "metrics/advantage_var": 4.4115118980407715,
+      "regularization/grad_norm_sq": 3978.581298828125,
+      "regularization/w2": 0.43020516633987427,
+      "rewards/chosen": 0.1904176281344506,
+      "rewards/margins": 0.10634150966521232,
+      "rewards/rejected": 0.08407611846923828,
+      "stability/repetition_rate_mean": 0.3856026828289032,
+      "stability/response_length_mean": 49.0,
+      "stability/response_length_std": 27.82085609436035,
+      "stability/response_length_var": 774.0,
+      "stability/token_entropy_mean": 5.1796875,
+      "step": 195
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 196,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 200,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.1861866257511875e+18,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}

mistral7b_kto_w2_delta0.005/checkpoint-196/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b00a81f45dd8c085139037c534b1aff270699bd4417624f905a02a36fe627e05
+size 8081

mistral7b_kto_w2_delta0.005/checkpoint-196/zero_to_fp32.py

@@ -0,0 +1,604 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None, exclude_frozen_parameters=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, 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_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("--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_file,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)

mistral7b_kto_w2_delta0.005/special_tokens_map.json

@@ -0,0 +1,24 @@
+{
+  "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": "</s>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

mistral7b_kto_w2_delta0.005/stability_metrics.jsonl

@@ -0,0 +1,38 @@
+{"stability/response_length_mean": 46.125, "stability/response_length_std": 27.231481552124023, "stability/response_length_var": 741.5535888671875, "stability/token_entropy_mean": 3.3662109375, "stability/repetition_rate_mean": 0.3492690920829773, "step": 10}
+{"stability/response_length_mean": 29.375, "stability/response_length_std": 29.92341423034668, "stability/response_length_var": 895.4107055664062, "stability/token_entropy_mean": 3.7548828125, "stability/repetition_rate_mean": 0.28515625, "step": 15}
+{"stability/response_length_mean": 44.0, "stability/response_length_std": 28.0611572265625, "stability/response_length_var": 787.4285888671875, "stability/token_entropy_mean": 4.42578125, "stability/repetition_rate_mean": 0.2096664160490036, "step": 20}
+{"stability/response_length_mean": 43.375, "stability/response_length_std": 28.54538917541504, "stability/response_length_var": 814.8392944335938, "stability/token_entropy_mean": 4.75390625, "stability/repetition_rate_mean": 0.18583066761493683, "step": 25}
+{"stability/response_length_mean": 43.0, "stability/response_length_std": 29.602123260498047, "stability/response_length_var": 876.2857055664062, "stability/token_entropy_mean": 4.642578125, "stability/repetition_rate_mean": 0.1583891361951828, "step": 30}
+{"stability/response_length_mean": 50.5, "stability/response_length_std": 25.048524856567383, "stability/response_length_var": 627.4285888671875, "stability/token_entropy_mean": 5.58203125, "stability/repetition_rate_mean": 0.3108258843421936, "step": 35}
+{"stability/response_length_mean": 35.0, "stability/response_length_std": 31.518701553344727, "stability/response_length_var": 993.4285888671875, "stability/token_entropy_mean": 4.89453125, "stability/repetition_rate_mean": 0.16694079339504242, "step": 40}
+{"stability/response_length_mean": 28.875, "stability/response_length_std": 29.87324333190918, "stability/response_length_var": 892.4107055664062, "stability/token_entropy_mean": 4.771484375, "stability/repetition_rate_mean": 0.1052611693739891, "step": 45}
+{"stability/response_length_mean": 53.375, "stability/response_length_std": 21.480472564697266, "stability/response_length_var": 461.41070556640625, "stability/token_entropy_mean": 3.8212890625, "stability/repetition_rate_mean": 0.3655294179916382, "step": 50}
+{"stability/response_length_mean": 22.375, "stability/response_length_std": 30.236862182617188, "stability/response_length_var": 914.2678833007812, "stability/token_entropy_mean": 4.3173828125, "stability/repetition_rate_mean": 0.10418051481246948, "step": 55}
+{"stability/response_length_mean": 44.625, "stability/response_length_std": 27.8820858001709, "stability/response_length_var": 777.4107055664062, "stability/token_entropy_mean": 4.646484375, "stability/repetition_rate_mean": 0.22545291483402252, "step": 60}
+{"stability/response_length_mean": 20.625, "stability/response_length_std": 27.176342010498047, "stability/response_length_var": 738.5535888671875, "stability/token_entropy_mean": 5.55859375, "stability/repetition_rate_mean": 0.19313038885593414, "step": 65}
+{"stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.3388671875, "stability/repetition_rate_mean": 0.111328125, "step": 70}
+{"stability/response_length_mean": 22.75, "stability/response_length_std": 31.07479476928711, "stability/response_length_var": 965.6428833007812, "stability/token_entropy_mean": 2.130859375, "stability/repetition_rate_mean": 0.10066105425357819, "step": 75}
+{"stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.23828125, "stability/repetition_rate_mean": 0.19088542461395264, "step": 80}
+{"stability/response_length_mean": 32.5, "stability/response_length_std": 33.67915725708008, "stability/response_length_var": 1134.2857666015625, "stability/token_entropy_mean": 4.9375, "stability/repetition_rate_mean": 0.103515625, "step": 85}
+{"stability/response_length_mean": 30.375, "stability/response_length_std": 29.061203002929688, "stability/response_length_var": 844.5535888671875, "stability/token_entropy_mean": 3.72265625, "stability/repetition_rate_mean": 0.3356415629386902, "step": 90}
+{"stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.75, "stability/repetition_rate_mean": 0.19374233484268188, "step": 95}
+{"stability/response_length_mean": 41.25, "stability/response_length_std": 31.476749420166016, "stability/response_length_var": 990.7857055664062, "stability/token_entropy_mean": 4.078125, "stability/repetition_rate_mean": 0.228515625, "step": 100}
+{"stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.353515625, "stability/repetition_rate_mean": 0.1877790093421936, "step": 105}
+{"stability/response_length_mean": 28.875, "stability/response_length_std": 29.911476135253906, "stability/response_length_var": 894.6964111328125, "stability/token_entropy_mean": 4.404296875, "stability/repetition_rate_mean": 0.30853456258773804, "step": 110}
+{"stability/response_length_mean": 49.25, "stability/response_length_std": 27.316923141479492, "stability/response_length_var": 746.2142944335938, "stability/token_entropy_mean": 4.75, "stability/repetition_rate_mean": 0.2194010466337204, "step": 115}
+{"stability/response_length_mean": 27.625, "stability/response_length_std": 30.486238479614258, "stability/response_length_var": 929.4107055664062, "stability/token_entropy_mean": 4.693359375, "stability/repetition_rate_mean": 0.15568576753139496, "step": 120}
+{"stability/response_length_mean": 23.125, "stability/response_length_std": 29.347366333007812, "stability/response_length_var": 861.2678833007812, "stability/token_entropy_mean": 3.353515625, "stability/repetition_rate_mean": 0.19144965708255768, "step": 125}
+{"stability/response_length_mean": 43.25, "stability/response_length_std": 29.620214462280273, "stability/response_length_var": 877.3571166992188, "stability/token_entropy_mean": 3.640625, "stability/repetition_rate_mean": 0.24460937082767487, "step": 130}
+{"stability/response_length_mean": 29.25, "stability/response_length_std": 29.936838150024414, "stability/response_length_var": 896.2142944335938, "stability/token_entropy_mean": 4.05078125, "stability/repetition_rate_mean": 0.2230902761220932, "step": 135}
+{"stability/response_length_mean": 56.25, "stability/response_length_std": 21.920310974121094, "stability/response_length_var": 480.5, "stability/token_entropy_mean": 5.001953125, "stability/repetition_rate_mean": 0.302734375, "step": 140}
+{"stability/response_length_mean": 22.625, "stability/response_length_std": 26.359804153442383, "stability/response_length_var": 694.8392944335938, "stability/token_entropy_mean": 5.66015625, "stability/repetition_rate_mean": 0.21016408503055573, "step": 145}
+{"stability/response_length_mean": 29.375, "stability/response_length_std": 29.193626403808594, "stability/response_length_var": 852.2678833007812, "stability/token_entropy_mean": 4.6015625, "stability/repetition_rate_mean": 0.23066024482250214, "step": 150}
+{"stability/response_length_mean": 58.125, "stability/response_length_std": 10.960155487060547, "stability/response_length_var": 120.125, "stability/token_entropy_mean": 4.56640625, "stability/repetition_rate_mean": 0.2978527545928955, "step": 155}
+{"stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.3720703125, "stability/repetition_rate_mean": 0.31046873331069946, "step": 160}
+{"stability/response_length_mean": 19.875, "stability/response_length_std": 27.771709442138672, "stability/response_length_var": 771.2678833007812, "stability/token_entropy_mean": 4.890625, "stability/repetition_rate_mean": 0.22421875596046448, "step": 165}
+{"stability/response_length_mean": 36.25, "stability/response_length_std": 30.76988410949707, "stability/response_length_var": 946.7857055664062, "stability/token_entropy_mean": 3.67822265625, "stability/repetition_rate_mean": 0.23209978640079498, "step": 170}
+{"stability/response_length_mean": 30.5, "stability/response_length_std": 26.42509651184082, "stability/response_length_var": 698.2857055664062, "stability/token_entropy_mean": 4.8369140625, "stability/repetition_rate_mean": 0.1976061761379242, "step": 175}
+{"stability/response_length_mean": 34.75, "stability/response_length_std": 31.54928207397461, "stability/response_length_var": 995.3571166992188, "stability/token_entropy_mean": 3.3203125, "stability/repetition_rate_mean": 0.25664061307907104, "step": 180}
+{"stability/response_length_mean": 32.875, "stability/response_length_std": 33.30996322631836, "stability/response_length_var": 1109.5535888671875, "stability/token_entropy_mean": 3.701171875, "stability/repetition_rate_mean": 0.19140625, "step": 185}
+{"stability/response_length_mean": 19.125, "stability/response_length_std": 28.296579360961914, "stability/response_length_var": 800.6964111328125, "stability/token_entropy_mean": 4.759765625, "stability/repetition_rate_mean": 0.1987847238779068, "step": 190}
+{"stability/response_length_mean": 49.0, "stability/response_length_std": 27.82085609436035, "stability/response_length_var": 774.0, "stability/token_entropy_mean": 5.1796875, "stability/repetition_rate_mean": 0.3856026828289032, "step": 195}

mistral7b_kto_w2_delta0.005/tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dadfd56d766715c61d2ef780a525ab43b8e6da4de6865bda3d95fdef5e134055
+size 493443

mistral7b_kto_w2_delta0.005/tokenizer_config.json

@@ -0,0 +1,46 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [],
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "extra_special_tokens": {},
+  "legacy": false,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "</s>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "split_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

mistral7b_kto_w2_delta0.005/train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.9280117068971906,
+    "train_runtime": 4303.5052,
+    "train_samples_per_second": 23.237,
+    "train_steps_per_second": 0.046
+}

mistral7b_kto_w2_delta0.005/trainer_log.jsonl

@@ -0,0 +1,40 @@
+{"current_steps": 5, "total_steps": 196, "loss": 0.9479, "lr": 4e-08, "epoch": 0.025591810620601407, "percentage": 2.55, "elapsed_time": "0:01:49", "remaining_time": "1:09:25", "rewards/chosen": 0.0006774790388973136, "logps/chosen": -439.03885690789474, "logits/chosen": -59718413.473684214, "rewards/rejected": -0.0012884867333230517, "logps/rejected": -198.42229352678572, "logits/rejected": -50413519.23809524, "rewards/margins": 0.0019659657722203654, "regularization/w2": 0.45370370149612427, "regularization/grad_norm_sq": 4334.98388671875, "kl": 0.008340898901224136}
+{"current_steps": 10, "total_steps": 196, "loss": 0.9286, "lr": 9e-08, "epoch": 0.05118362124120281, "percentage": 5.1, "elapsed_time": "0:03:39", "remaining_time": "1:07:56", "rewards/chosen": 0.0011910808744368614, "logps/chosen": -429.92659293831167, "logits/chosen": -59302692.571428575, "rewards/rejected": 0.0016303154000316758, "logps/rejected": -192.85163309487953, "logits/rejected": -50021761.54216868, "rewards/margins": -0.00043923452559481435, "regularization/w2": 0.4279446601867676, "regularization/grad_norm_sq": 3756.94775390625, "kl": 0.016795814037322998, "stability/response_length_mean": 46.125, "stability/response_length_std": 27.231481552124023, "stability/response_length_var": 741.5535888671875, "stability/token_entropy_mean": 3.3662109375, "stability/repetition_rate_mean": 0.3492690920829773}
+{"current_steps": 15, "total_steps": 196, "loss": 0.9272, "lr": 1.4e-07, "epoch": 0.07677543186180422, "percentage": 7.65, "elapsed_time": "0:05:27", "remaining_time": "1:05:54", "rewards/chosen": 0.0009465837117397424, "logps/chosen": -404.12135416666666, "logits/chosen": -56236609.163636364, "rewards/rejected": 0.0015350009164502543, "logps/rejected": -200.67870463709679, "logits/rejected": -51916776.87741935, "rewards/margins": -0.0005884172047105119, "regularization/w2": 0.44162631034851074, "regularization/grad_norm_sq": 4170.2373046875, "kl": 0.020404411479830742, "stability/response_length_mean": 29.375, "stability/response_length_std": 29.92341423034668, "stability/response_length_var": 895.4107055664062, "stability/token_entropy_mean": 3.7548828125, "stability/repetition_rate_mean": 0.28515625}
+{"current_steps": 20, "total_steps": 196, "loss": 0.9324, "lr": 1.8999999999999998e-07, "epoch": 0.10236724248240563, "percentage": 10.2, "elapsed_time": "0:07:19", "remaining_time": "1:04:23", "rewards/chosen": 0.0034450016915798186, "logps/chosen": -343.6622802734375, "logits/chosen": -51675481.6, "rewards/rejected": 0.004709281772375107, "logps/rejected": -225.616259765625, "logits/rejected": -48057888.0, "rewards/margins": -0.001264280080795288, "regularization/w2": 0.42263123393058777, "regularization/grad_norm_sq": 3680.827392578125, "kl": 0.021697884425520897, "stability/response_length_mean": 44.0, "stability/response_length_std": 28.0611572265625, "stability/response_length_var": 787.4285888671875, "stability/token_entropy_mean": 4.42578125, "stability/repetition_rate_mean": 0.2096664160490036}
+{"current_steps": 25, "total_steps": 196, "loss": 0.9391, "lr": 2.4e-07, "epoch": 0.12795905310300704, "percentage": 12.76, "elapsed_time": "0:09:07", "remaining_time": "1:02:26", "rewards/chosen": 0.0047595426954072095, "logps/chosen": -378.4438577586207, "logits/chosen": -56471566.12413793, "rewards/rejected": 0.004253701482500349, "logps/rejected": -202.7503125, "logits/rejected": -52024934.4, "rewards/margins": 0.0005058412129068605, "regularization/w2": 0.4309103488922119, "regularization/grad_norm_sq": 4002.333984375, "kl": 0.018133217468857765, "stability/response_length_mean": 43.375, "stability/response_length_std": 28.54538917541504, "stability/response_length_var": 814.8392944335938, "stability/token_entropy_mean": 4.75390625, "stability/repetition_rate_mean": 0.18583066761493683}
+{"current_steps": 30, "total_steps": 196, "loss": 0.9337, "lr": 2.9e-07, "epoch": 0.15355086372360843, "percentage": 15.31, "elapsed_time": "0:10:57", "remaining_time": "1:00:35", "rewards/chosen": -0.0017742578615278207, "logps/chosen": -468.5989932885906, "logits/chosen": -57729188.939597316, "rewards/rejected": 0.0039528608322143555, "logps/rejected": -198.2083561769006, "logits/rejected": -51520841.35672515, "rewards/margins": -0.005727118693742176, "regularization/w2": 0.45624828338623047, "regularization/grad_norm_sq": 4491.3251953125, "kl": 0.022612739354372025, "stability/response_length_mean": 43.0, "stability/response_length_std": 29.602123260498047, "stability/response_length_var": 876.2857055664062, "stability/token_entropy_mean": 4.642578125, "stability/repetition_rate_mean": 0.1583891361951828}
+{"current_steps": 35, "total_steps": 196, "loss": 0.938, "lr": 3.4000000000000003e-07, "epoch": 0.17914267434420986, "percentage": 17.86, "elapsed_time": "0:12:46", "remaining_time": "0:58:47", "rewards/chosen": 0.001247800886631012, "logps/chosen": -385.9240506329114, "logits/chosen": -57644589.36708861, "rewards/rejected": 0.0008610994727523239, "logps/rejected": -195.117862654321, "logits/rejected": -53911292.83950617, "rewards/margins": 0.0003867014138786881, "regularization/w2": 0.4280441403388977, "regularization/grad_norm_sq": 3758.585205078125, "kl": 0.02071223594248295, "stability/response_length_mean": 50.5, "stability/response_length_std": 25.048524856567383, "stability/response_length_var": 627.4285888671875, "stability/token_entropy_mean": 5.58203125, "stability/repetition_rate_mean": 0.3108258843421936}
+{"current_steps": 40, "total_steps": 196, "loss": 0.9341, "lr": 3.8999999999999997e-07, "epoch": 0.20473448496481125, "percentage": 20.41, "elapsed_time": "0:14:37", "remaining_time": "0:57:01", "rewards/chosen": 0.0003104138188064098, "logps/chosen": -421.899560546875, "logits/chosen": -56634060.8, "rewards/rejected": 0.004448341205716133, "logps/rejected": -219.2647705078125, "logits/rejected": -50219801.6, "rewards/margins": -0.0041379273869097235, "regularization/w2": 0.4727831482887268, "regularization/grad_norm_sq": 5033.30859375, "kl": 0.02293821983039379, "stability/response_length_mean": 35.0, "stability/response_length_std": 31.518701553344727, "stability/response_length_var": 993.4285888671875, "stability/token_entropy_mean": 4.89453125, "stability/repetition_rate_mean": 0.16694079339504242}
+{"current_steps": 45, "total_steps": 196, "loss": 0.9389, "lr": 4.3999999999999997e-07, "epoch": 0.23032629558541268, "percentage": 22.96, "elapsed_time": "0:16:27", "remaining_time": "0:55:13", "rewards/chosen": 0.0053554469306997, "logps/chosen": -352.2749056208054, "logits/chosen": -52707451.70469799, "rewards/rejected": 0.0031691573516667237, "logps/rejected": -243.8424707602339, "logits/rejected": -50772524.9122807, "rewards/margins": 0.0021862895790329766, "regularization/w2": 0.42452260851860046, "regularization/grad_norm_sq": 3936.931640625, "kl": 0.0444142147898674, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.87324333190918, "stability/response_length_var": 892.4107055664062, "stability/token_entropy_mean": 4.771484375, "stability/repetition_rate_mean": 0.1052611693739891}
+{"current_steps": 50, "total_steps": 196, "loss": 0.9201, "lr": 4.9e-07, "epoch": 0.2559181062060141, "percentage": 25.51, "elapsed_time": "0:18:16", "remaining_time": "0:53:22", "rewards/chosen": 0.005941886629834271, "logps/chosen": -375.1397336409396, "logits/chosen": -54037710.17449664, "rewards/rejected": 0.005253165088898954, "logps/rejected": -180.18807108918128, "logits/rejected": -48296085.70760234, "rewards/margins": 0.0006887215409353172, "regularization/w2": 0.4266422390937805, "regularization/grad_norm_sq": 3871.12744140625, "kl": 0.04537983611226082, "stability/response_length_mean": 53.375, "stability/response_length_std": 21.480472564697266, "stability/response_length_var": 461.41070556640625, "stability/token_entropy_mean": 3.8212890625, "stability/repetition_rate_mean": 0.3655294179916382}
+{"current_steps": 55, "total_steps": 196, "loss": 0.9291, "lr": 5.4e-07, "epoch": 0.28150991682661547, "percentage": 28.06, "elapsed_time": "0:20:05", "remaining_time": "0:51:31", "rewards/chosen": 0.0024627199084121987, "logps/chosen": -344.48738354037266, "logits/chosen": -56953277.217391305, "rewards/rejected": 0.006040624477578409, "logps/rejected": -230.3876523191824, "logits/rejected": -49571939.82389937, "rewards/margins": -0.00357790456916621, "regularization/w2": 0.43878814578056335, "regularization/grad_norm_sq": 3976.169921875, "kl": 0.05794268846511841, "stability/response_length_mean": 22.375, "stability/response_length_std": 30.236862182617188, "stability/response_length_var": 914.2678833007812, "stability/token_entropy_mean": 4.3173828125, "stability/repetition_rate_mean": 0.10418051481246948}
+{"current_steps": 60, "total_steps": 196, "loss": 0.933, "lr": 5.9e-07, "epoch": 0.30710172744721687, "percentage": 30.61, "elapsed_time": "0:21:56", "remaining_time": "0:49:44", "rewards/chosen": 0.010837621399850556, "logps/chosen": -359.6124053030303, "logits/chosen": -52678345.696969695, "rewards/rejected": 0.004823949260096396, "logps/rejected": -178.17867943548387, "logits/rejected": -47821840.51612903, "rewards/margins": 0.00601367213975416, "regularization/w2": 0.3937515616416931, "regularization/grad_norm_sq": 3165.790283203125, "kl": 0.05679145082831383, "stability/response_length_mean": 44.625, "stability/response_length_std": 27.8820858001709, "stability/response_length_var": 777.4107055664062, "stability/token_entropy_mean": 4.646484375, "stability/repetition_rate_mean": 0.22545291483402252}
+{"current_steps": 65, "total_steps": 196, "loss": 0.939, "lr": 6.4e-07, "epoch": 0.3326935380678183, "percentage": 33.16, "elapsed_time": "0:23:47", "remaining_time": "0:47:56", "rewards/chosen": 0.008308312200730847, "logps/chosen": -381.86370967741937, "logits/chosen": -57563182.24516129, "rewards/rejected": 0.006429539304791075, "logps/rejected": -183.5180516098485, "logits/rejected": -47268345.7939394, "rewards/margins": 0.001878772895939772, "regularization/w2": 0.4315285384654999, "regularization/grad_norm_sq": 3868.786865234375, "kl": 0.06266746670007706, "stability/response_length_mean": 20.625, "stability/response_length_std": 27.176342010498047, "stability/response_length_var": 738.5535888671875, "stability/token_entropy_mean": 5.55859375, "stability/repetition_rate_mean": 0.19313038885593414}
+{"current_steps": 70, "total_steps": 196, "loss": 0.9497, "lr": 6.9e-07, "epoch": 0.3582853486884197, "percentage": 35.71, "elapsed_time": "0:25:36", "remaining_time": "0:46:04", "rewards/chosen": 0.01309710318042386, "logps/chosen": -378.9974798387097, "logits/chosen": -52753698.68387097, "rewards/rejected": 0.007028058803442753, "logps/rejected": -220.35189393939393, "logits/rejected": -49023162.18181818, "rewards/margins": 0.006069044376981108, "regularization/w2": 0.45682960748672485, "regularization/grad_norm_sq": 4483.40087890625, "kl": 0.08709845691919327, "stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.3388671875, "stability/repetition_rate_mean": 0.111328125}
+{"current_steps": 75, "total_steps": 196, "loss": 0.9402, "lr": 7.4e-07, "epoch": 0.3838771593090211, "percentage": 38.27, "elapsed_time": "0:27:26", "remaining_time": "0:44:16", "rewards/chosen": 0.013092591955855087, "logps/chosen": -314.13988597972974, "logits/chosen": -49498509.83783784, "rewards/rejected": 0.009581702393154765, "logps/rejected": -228.78901707848837, "logits/rejected": -47885377.488372095, "rewards/margins": 0.0035108895627003218, "regularization/w2": 0.42404261231422424, "regularization/grad_norm_sq": 3736.681396484375, "kl": 0.09017860144376755, "stability/response_length_mean": 22.75, "stability/response_length_std": 31.07479476928711, "stability/response_length_var": 965.6428833007812, "stability/token_entropy_mean": 2.130859375, "stability/repetition_rate_mean": 0.10066105425357819}
+{"current_steps": 80, "total_steps": 196, "loss": 0.9316, "lr": 7.9e-07, "epoch": 0.4094689699296225, "percentage": 40.82, "elapsed_time": "0:29:17", "remaining_time": "0:42:28", "rewards/chosen": 0.0159459240389186, "logps/chosen": -403.87065397350995, "logits/chosen": -58626637.98675497, "rewards/rejected": 0.00917972968174861, "logps/rejected": -200.96398853550295, "logits/rejected": -53447752.71005917, "rewards/margins": 0.006766194357169991, "regularization/w2": 0.42037779092788696, "regularization/grad_norm_sq": 3624.650390625, "kl": 0.13238738477230072, "stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.23828125, "stability/repetition_rate_mean": 0.19088542461395264}
+{"current_steps": 85, "total_steps": 196, "loss": 0.9311, "lr": 8.399999999999999e-07, "epoch": 0.4350607805502239, "percentage": 43.37, "elapsed_time": "0:31:07", "remaining_time": "0:40:38", "rewards/chosen": 0.01977017310357863, "logps/chosen": -358.13480342741934, "logits/chosen": -51478686.55483871, "rewards/rejected": 0.008150509631994998, "logps/rejected": -230.34330018939394, "logits/rejected": -47141816.63030303, "rewards/margins": 0.011619663471583632, "regularization/w2": 0.43580102920532227, "regularization/grad_norm_sq": 3923.891357421875, "kl": 0.12365061044692993, "stability/response_length_mean": 32.5, "stability/response_length_std": 33.67915725708008, "stability/response_length_var": 1134.2857666015625, "stability/token_entropy_mean": 4.9375, "stability/repetition_rate_mean": 0.103515625}
+{"current_steps": 90, "total_steps": 196, "loss": 0.9286, "lr": 8.9e-07, "epoch": 0.46065259117082535, "percentage": 45.92, "elapsed_time": "0:32:58", "remaining_time": "0:38:50", "rewards/chosen": 0.022810822878128443, "logps/chosen": -380.5612980769231, "logits/chosen": -58412438.974358976, "rewards/rejected": 0.009435738005289218, "logps/rejected": -187.1324433117378, "logits/rejected": -50588893.65853658, "rewards/margins": 0.013375084872839225, "regularization/w2": 0.43326812982559204, "regularization/grad_norm_sq": 3925.050048828125, "kl": 0.16665403544902802, "stability/response_length_mean": 30.375, "stability/response_length_std": 29.061203002929688, "stability/response_length_var": 844.5535888671875, "stability/token_entropy_mean": 3.72265625, "stability/repetition_rate_mean": 0.3356415629386902}
+{"current_steps": 95, "total_steps": 196, "loss": 0.9271, "lr": 9.399999999999999e-07, "epoch": 0.48624440179142675, "percentage": 48.47, "elapsed_time": "0:34:48", "remaining_time": "0:37:00", "rewards/chosen": 0.021952728681926485, "logps/chosen": -314.77383306962025, "logits/chosen": -50379520.0, "rewards/rejected": 0.009519148020096768, "logps/rejected": -207.04853877314815, "logits/rejected": -48967028.938271604, "rewards/margins": 0.012433580661829717, "regularization/w2": 0.42039623856544495, "regularization/grad_norm_sq": 3692.677490234375, "kl": 0.15967555344104767, "stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.75, "stability/repetition_rate_mean": 0.19374233484268188}
+{"current_steps": 100, "total_steps": 196, "loss": 0.9328, "lr": 9.9e-07, "epoch": 0.5118362124120281, "percentage": 51.02, "elapsed_time": "0:36:38", "remaining_time": "0:35:10", "rewards/chosen": 0.03042653444651011, "logps/chosen": -359.5659311655405, "logits/chosen": -54698091.24324324, "rewards/rejected": 0.018583308818728425, "logps/rejected": -216.3930323401163, "logits/rejected": -50908255.255813956, "rewards/margins": 0.011843225627781683, "regularization/w2": 0.42957553267478943, "regularization/grad_norm_sq": 3865.565673828125, "kl": 0.1950296312570572, "stability/response_length_mean": 41.25, "stability/response_length_std": 31.476749420166016, "stability/response_length_var": 990.7857055664062, "stability/token_entropy_mean": 4.078125, "stability/repetition_rate_mean": 0.228515625}
+{"current_steps": 105, "total_steps": 196, "loss": 0.9365, "lr": 9.999815554351224e-07, "epoch": 0.5374280230326296, "percentage": 53.57, "elapsed_time": "0:38:28", "remaining_time": "0:33:20", "rewards/chosen": 0.04099334004413651, "logps/chosen": -399.4925169427711, "logits/chosen": -59378317.87951807, "rewards/rejected": 0.016021623239888774, "logps/rejected": -209.84361049107142, "logits/rejected": -53788419.32467532, "rewards/margins": 0.024971716804247736, "regularization/w2": 0.4382788836956024, "regularization/grad_norm_sq": 3994.716064453125, "kl": 0.21558904647827148, "stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.353515625, "stability/repetition_rate_mean": 0.1877790093421936}
+{"current_steps": 110, "total_steps": 196, "loss": 0.9259, "lr": 9.999066267225444e-07, "epoch": 0.5630198336532309, "percentage": 56.12, "elapsed_time": "0:40:19", "remaining_time": "0:31:31", "rewards/chosen": 0.0344322733132236, "logps/chosen": -378.218938253012, "logits/chosen": -57391930.60240964, "rewards/rejected": 0.020619178747201896, "logps/rejected": -205.88195008116884, "logits/rejected": -53613953.66233766, "rewards/margins": 0.013813094566021703, "regularization/w2": 0.42161688208580017, "regularization/grad_norm_sq": 3637.52197265625, "kl": 0.26441773772239685, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.911476135253906, "stability/response_length_var": 894.6964111328125, "stability/token_entropy_mean": 4.404296875, "stability/repetition_rate_mean": 0.30853456258773804}
+{"current_steps": 115, "total_steps": 196, "loss": 0.9207, "lr": 9.997740697079592e-07, "epoch": 0.5886116442738324, "percentage": 58.67, "elapsed_time": "0:42:11", "remaining_time": "0:29:42", "rewards/chosen": 0.04908096337620216, "logps/chosen": -364.10173556170884, "logits/chosen": -60598233.11392405, "rewards/rejected": 0.022169822528038497, "logps/rejected": -200.97486255787038, "logits/rejected": -53416372.14814815, "rewards/margins": 0.02691114084816366, "regularization/w2": 0.4012783169746399, "regularization/grad_norm_sq": 3352.978271484375, "kl": 0.3085666298866272, "stability/response_length_mean": 49.25, "stability/response_length_std": 27.316923141479492, "stability/response_length_var": 746.2142944335938, "stability/token_entropy_mean": 4.75, "stability/repetition_rate_mean": 0.2194010466337204}
+{"current_steps": 120, "total_steps": 196, "loss": 0.936, "lr": 9.995838996722914e-07, "epoch": 0.6142034548944337, "percentage": 61.22, "elapsed_time": "0:43:59", "remaining_time": "0:27:51", "rewards/chosen": 0.05132290896247415, "logps/chosen": -383.7316687091503, "logits/chosen": -57739732.49673203, "rewards/rejected": 0.027908859138717195, "logps/rejected": -211.93726609281438, "logits/rejected": -54352828.5508982, "rewards/margins": 0.023414049823756954, "regularization/w2": 0.4518660008907318, "regularization/grad_norm_sq": 4454.72216796875, "kl": 0.3443666994571686, "stability/response_length_mean": 27.625, "stability/response_length_std": 30.486238479614258, "stability/response_length_var": 929.4107055664062, "stability/token_entropy_mean": 4.693359375, "stability/repetition_rate_mean": 0.15568576753139496}
+{"current_steps": 125, "total_steps": 196, "loss": 0.9231, "lr": 9.993361385379874e-07, "epoch": 0.6397952655150352, "percentage": 63.78, "elapsed_time": "0:45:50", "remaining_time": "0:26:02", "rewards/chosen": 0.05853001058918156, "logps/chosen": -416.79403895547944, "logits/chosen": -60667195.61643836, "rewards/rejected": 0.028942058826314992, "logps/rejected": -216.0777658045977, "logits/rejected": -55641405.79310345, "rewards/margins": 0.029587951762866566, "regularization/w2": 0.4254666864871979, "regularization/grad_norm_sq": 3774.068115234375, "kl": 0.3985693156719208, "stability/response_length_mean": 23.125, "stability/response_length_std": 29.347366333007812, "stability/response_length_var": 861.2678833007812, "stability/token_entropy_mean": 3.353515625, "stability/repetition_rate_mean": 0.19144965708255768}
+{"current_steps": 130, "total_steps": 196, "loss": 0.9394, "lr": 9.99030814866488e-07, "epoch": 0.6653870761356366, "percentage": 66.33, "elapsed_time": "0:47:40", "remaining_time": "0:24:12", "rewards/chosen": 0.06672527559341923, "logps/chosen": -391.9906754032258, "logits/chosen": -57375400.464516126, "rewards/rejected": 0.026918549971147017, "logps/rejected": -183.6497159090909, "logits/rejected": -53823680.38787879, "rewards/margins": 0.039806725622272214, "regularization/w2": 0.42766115069389343, "regularization/grad_norm_sq": 3780.29150390625, "kl": 0.45674076676368713, "stability/response_length_mean": 43.25, "stability/response_length_std": 29.620214462280273, "stability/response_length_var": 877.3571166992188, "stability/token_entropy_mean": 3.640625, "stability/repetition_rate_mean": 0.24460937082767487}
+{"current_steps": 135, "total_steps": 196, "loss": 0.9147, "lr": 9.986679638549358e-07, "epoch": 0.690978886756238, "percentage": 68.88, "elapsed_time": "0:49:29", "remaining_time": "0:22:21", "rewards/chosen": 0.0830619554578161, "logps/chosen": -400.98552530674846, "logits/chosen": -59843408.09815951, "rewards/rejected": 0.035854473235500846, "logps/rejected": -247.51219148089172, "logits/rejected": -55000846.67515924, "rewards/margins": 0.04720748222231525, "regularization/w2": 0.4519772231578827, "regularization/grad_norm_sq": 4315.08056640625, "kl": 0.5136879086494446, "stability/response_length_mean": 29.25, "stability/response_length_std": 29.936838150024414, "stability/response_length_var": 896.2142944335938, "stability/token_entropy_mean": 4.05078125, "stability/repetition_rate_mean": 0.2230902761220932}
+{"current_steps": 140, "total_steps": 196, "loss": 0.93, "lr": 9.982476273321175e-07, "epoch": 0.7165706973768394, "percentage": 71.43, "elapsed_time": "0:51:19", "remaining_time": "0:20:31", "rewards/chosen": 0.08023344097715436, "logps/chosen": -366.3874053030303, "logits/chosen": -57076500.16969697, "rewards/rejected": 0.011303200260285409, "logps/rejected": -189.92144657258063, "logits/rejected": -50077570.477419354, "rewards/margins": 0.06893024071686896, "regularization/w2": 0.48489466309547424, "regularization/grad_norm_sq": 5980.30029296875, "kl": 0.4908495843410492, "stability/response_length_mean": 56.25, "stability/response_length_std": 21.920310974121094, "stability/response_length_var": 480.5, "stability/token_entropy_mean": 5.001953125, "stability/repetition_rate_mean": 0.302734375}
+{"current_steps": 145, "total_steps": 196, "loss": 0.9176, "lr": 9.977698537536417e-07, "epoch": 0.7421625079974408, "percentage": 73.98, "elapsed_time": "0:53:08", "remaining_time": "0:18:41", "rewards/chosen": 0.0799884566341538, "logps/chosen": -329.4719267695783, "logits/chosen": -51146073.44578313, "rewards/rejected": 0.03340069349710043, "logps/rejected": -199.15390371347402, "logits/rejected": -51662515.53246753, "rewards/margins": 0.046587763137053366, "regularization/w2": 0.40856075286865234, "regularization/grad_norm_sq": 3416.243408203125, "kl": 0.577669084072113, "stability/response_length_mean": 22.625, "stability/response_length_std": 26.359804153442383, "stability/response_length_var": 694.8392944335938, "stability/token_entropy_mean": 5.66015625, "stability/repetition_rate_mean": 0.21016408503055573}
+{"current_steps": 150, "total_steps": 196, "loss": 0.923, "lr": 9.972346981963546e-07, "epoch": 0.7677543186180422, "percentage": 76.53, "elapsed_time": "0:54:58", "remaining_time": "0:16:51", "rewards/chosen": 0.11169673695283777, "logps/chosen": -396.8174632352941, "logits/chosen": -55421090.63529412, "rewards/rejected": 0.037329667409261066, "logps/rejected": -207.35263020833332, "logits/rejected": -50187294.72, "rewards/margins": 0.0743670695435767, "regularization/w2": 0.4475216865539551, "regularization/grad_norm_sq": 4074.032470703125, "kl": 0.6454542875289917, "stability/response_length_mean": 29.375, "stability/response_length_std": 29.193626403808594, "stability/response_length_var": 852.2678833007812, "stability/token_entropy_mean": 4.6015625, "stability/repetition_rate_mean": 0.23066024482250214}
+{"current_steps": 155, "total_steps": 196, "loss": 0.9163, "lr": 9.966422223519885e-07, "epoch": 0.7933461292386437, "percentage": 79.08, "elapsed_time": "0:56:49", "remaining_time": "0:15:01", "rewards/chosen": 0.08927134345559513, "logps/chosen": -360.09657628676473, "logits/chosen": -54003495.152941175, "rewards/rejected": 0.029927142461140952, "logps/rejected": -191.00634114583335, "logits/rejected": -49346880.85333333, "rewards/margins": 0.059344200994454185, "regularization/w2": 0.46441006660461426, "regularization/grad_norm_sq": 4929.31494140625, "kl": 0.6456610560417175, "stability/response_length_mean": 58.125, "stability/response_length_std": 10.960155487060547, "stability/response_length_var": 120.125, "stability/token_entropy_mean": 4.56640625, "stability/repetition_rate_mean": 0.2978527545928955}
+{"current_steps": 160, "total_steps": 196, "loss": 0.9248, "lr": 9.959924945200523e-07, "epoch": 0.818937939859245, "percentage": 81.63, "elapsed_time": "0:58:39", "remaining_time": "0:13:11", "rewards/chosen": 0.1314907301039923, "logps/chosen": -415.5804501488095, "logits/chosen": -55241313.52380952, "rewards/rejected": 0.042252054339961, "logps/rejected": -186.47481496710526, "logits/rejected": -48956978.526315786, "rewards/margins": 0.08923867576403129, "regularization/w2": 0.43814727663993835, "regularization/grad_norm_sq": 4000.94580078125, "kl": 0.7202507257461548, "stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.3720703125, "stability/repetition_rate_mean": 0.31046873331069946}
+{"current_steps": 165, "total_steps": 196, "loss": 0.9192, "lr": 9.952855895999568e-07, "epoch": 0.8445297504798465, "percentage": 84.18, "elapsed_time": "1:00:29", "remaining_time": "0:11:21", "rewards/chosen": 0.11661315786427465, "logps/chosen": -369.750628591954, "logits/chosen": -54405402.48275862, "rewards/rejected": 0.04505188824379281, "logps/rejected": -212.1415881849315, "logits/rejected": -49781749.47945205, "rewards/margins": 0.07156126962048184, "regularization/w2": 0.4151502549648285, "regularization/grad_norm_sq": 3549.75244140625, "kl": 0.7532386779785156, "stability/response_length_mean": 19.875, "stability/response_length_std": 27.771709442138672, "stability/response_length_var": 771.2678833007812, "stability/token_entropy_mean": 4.890625, "stability/repetition_rate_mean": 0.22421875596046448}
+{"current_steps": 170, "total_steps": 196, "loss": 0.905, "lr": 9.9452158908238e-07, "epoch": 0.8701215611004478, "percentage": 86.73, "elapsed_time": "1:02:20", "remaining_time": "0:09:32", "rewards/chosen": 0.16818395309899686, "logps/chosen": -439.6600406804734, "logits/chosen": -59125432.80473373, "rewards/rejected": 0.06720379962036942, "logps/rejected": -190.58578228476821, "logits/rejected": -52567887.682119206, "rewards/margins": 0.10098015347862745, "regularization/w2": 0.4287300109863281, "regularization/grad_norm_sq": 3930.178955078125, "kl": 0.9238977432250977, "stability/response_length_mean": 36.25, "stability/response_length_std": 30.76988410949707, "stability/response_length_var": 946.7857055664062, "stability/token_entropy_mean": 3.67822265625, "stability/repetition_rate_mean": 0.23209978640079498}
+{"current_steps": 175, "total_steps": 196, "loss": 0.911, "lr": 9.937005810398745e-07, "epoch": 0.8957133717210493, "percentage": 89.29, "elapsed_time": "1:04:10", "remaining_time": "0:07:42", "rewards/chosen": 0.15448007238916603, "logps/chosen": -354.89241340361446, "logits/chosen": -56652960.38554217, "rewards/rejected": 0.04994386511963683, "logps/rejected": -194.30808644480518, "logits/rejected": -48022956.883116886, "rewards/margins": 0.1045362072695292, "regularization/w2": 0.4321208894252777, "regularization/grad_norm_sq": 4139.484375, "kl": 0.9429834485054016, "stability/response_length_mean": 30.5, "stability/response_length_std": 26.42509651184082, "stability/response_length_var": 698.2857055664062, "stability/token_entropy_mean": 4.8369140625, "stability/repetition_rate_mean": 0.1976061761379242}
+{"current_steps": 180, "total_steps": 196, "loss": 0.9028, "lr": 9.928226601167138e-07, "epoch": 0.9213051823416507, "percentage": 91.84, "elapsed_time": "1:06:00", "remaining_time": "0:05:52", "rewards/chosen": 0.1727198852351838, "logps/chosen": -374.418615797546, "logits/chosen": -52658986.40490798, "rewards/rejected": 0.07594777975872063, "logps/rejected": -210.53871417197453, "logits/rejected": -47980801.63057325, "rewards/margins": 0.09677210547646317, "regularization/w2": 0.41559115052223206, "regularization/grad_norm_sq": 3664.698974609375, "kl": 1.0304526090621948, "stability/response_length_mean": 34.75, "stability/response_length_std": 31.54928207397461, "stability/response_length_var": 995.3571166992188, "stability/token_entropy_mean": 3.3203125, "stability/repetition_rate_mean": 0.25664061307907104}
+{"current_steps": 185, "total_steps": 196, "loss": 0.9317, "lr": 9.918879275179817e-07, "epoch": 0.946896992962252, "percentage": 94.39, "elapsed_time": "1:07:49", "remaining_time": "0:04:01", "rewards/chosen": 0.16781084037121433, "logps/chosen": -310.32325424382714, "logits/chosen": -54608421.925925925, "rewards/rejected": 0.04625999776622917, "logps/rejected": -196.57655508306962, "logits/rejected": -49521022.37974683, "rewards/margins": 0.12155084260498517, "regularization/w2": 0.47685596346855164, "regularization/grad_norm_sq": 7077.32763671875, "kl": 1.06327223777771, "stability/response_length_mean": 32.875, "stability/response_length_std": 33.30996322631836, "stability/response_length_var": 1109.5535888671875, "stability/token_entropy_mean": 3.701171875, "stability/repetition_rate_mean": 0.19140625}
+{"current_steps": 190, "total_steps": 196, "loss": 0.9237, "lr": 9.90896490997906e-07, "epoch": 0.9724888035828535, "percentage": 96.94, "elapsed_time": "1:09:38", "remaining_time": "0:02:11", "rewards/chosen": 0.2101439972837528, "logps/chosen": -390.30389221556885, "logits/chosen": -53089470.08383234, "rewards/rejected": 0.0736633874232473, "logps/rejected": -210.46195874183007, "logits/rejected": -49691681.46405229, "rewards/margins": 0.1364806098605055, "regularization/w2": 0.4580075442790985, "regularization/grad_norm_sq": 5314.27880859375, "kl": 1.1201375722885132, "stability/response_length_mean": 19.125, "stability/response_length_std": 28.296579360961914, "stability/response_length_var": 800.6964111328125, "stability/token_entropy_mean": 4.759765625, "stability/repetition_rate_mean": 0.1987847238779068}
+{"current_steps": 195, "total_steps": 196, "loss": 0.9173, "lr": 9.898484648474362e-07, "epoch": 0.9980806142034548, "percentage": 99.49, "elapsed_time": "1:11:27", "remaining_time": "0:00:21", "rewards/chosen": 0.1904176281344506, "logps/chosen": -390.8807459677419, "logits/chosen": -60032350.14193548, "rewards/rejected": 0.08407611846923828, "logps/rejected": -197.07857481060606, "logits/rejected": -53812130.90909091, "rewards/margins": 0.10634150966521232, "regularization/w2": 0.43020516633987427, "regularization/grad_norm_sq": 3978.581298828125, "kl": 1.1507009267807007, "stability/response_length_mean": 49.0, "stability/response_length_std": 27.82085609436035, "stability/response_length_var": 774.0, "stability/token_entropy_mean": 5.1796875, "stability/repetition_rate_mean": 0.3856026828289032}
+{"current_steps": 196, "total_steps": 196, "epoch": 1.0, "percentage": 100.0, "elapsed_time": "1:11:43", "remaining_time": "0:00:00"}

mistral7b_kto_w2_delta0.005/trainer_state.json

@@ -0,0 +1,974 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 196,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.025591810620601407,
+      "grad_norm": 0.577160120010376,
+      "kl": 0.008340898901224136,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59718413.473684214,
+      "logits/rejected": -50413519.23809524,
+      "logps/chosen": -439.03885690789474,
+      "logps/rejected": -198.42229352678572,
+      "loss": 0.9479,
+      "loss/base_kto": 0.4997499883174896,
+      "metrics/advantage_var": 0.052478689700365067,
+      "regularization/grad_norm_sq": 4334.98388671875,
+      "regularization/w2": 0.45370370149612427,
+      "rewards/chosen": 0.0006774790388973136,
+      "rewards/margins": 0.0019659657722203654,
+      "rewards/rejected": -0.0012884867333230517,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.611538290977478,
+      "kl": 0.016795814037322998,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59302692.571428575,
+      "logits/rejected": -50021761.54216868,
+      "logps/chosen": -429.92659293831167,
+      "logps/rejected": -192.85163309487953,
+      "loss": 0.9286,
+      "loss/base_kto": 0.5000812411308289,
+      "metrics/advantage_var": 0.0676804929971695,
+      "regularization/grad_norm_sq": 3756.94775390625,
+      "regularization/w2": 0.4279446601867676,
+      "rewards/chosen": 0.0011910808744368614,
+      "rewards/margins": -0.00043923452559481435,
+      "rewards/rejected": 0.0016303154000316758,
+      "stability/repetition_rate_mean": 0.3492690920829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.3662109375,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.6086092591285706,
+      "kl": 0.020404411479830742,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56236609.163636364,
+      "logits/rejected": -51916776.87741935,
+      "logps/chosen": -404.12135416666666,
+      "logps/rejected": -200.67870463709679,
+      "loss": 0.9272,
+      "loss/base_kto": 0.49999886751174927,
+      "metrics/advantage_var": 0.05835923179984093,
+      "regularization/grad_norm_sq": 4170.2373046875,
+      "regularization/w2": 0.44162631034851074,
+      "rewards/chosen": 0.0009465837117397424,
+      "rewards/margins": -0.0005884172047105119,
+      "rewards/rejected": 0.0015350009164502543,
+      "stability/repetition_rate_mean": 0.28515625,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.92341423034668,
+      "stability/response_length_var": 895.4107055664062,
+      "stability/token_entropy_mean": 3.7548828125,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.6190893650054932,
+      "kl": 0.021697884425520897,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51675481.6,
+      "logits/rejected": -48057888.0,
+      "logps/chosen": -343.6622802734375,
+      "logps/rejected": -225.616259765625,
+      "loss": 0.9324,
+      "loss/base_kto": 0.5001762509346008,
+      "metrics/advantage_var": 0.06153888627886772,
+      "regularization/grad_norm_sq": 3680.827392578125,
+      "regularization/w2": 0.42263123393058777,
+      "rewards/chosen": 0.0034450016915798186,
+      "rewards/margins": -0.001264280080795288,
+      "rewards/rejected": 0.004709281772375107,
+      "stability/repetition_rate_mean": 0.2096664160490036,
+      "stability/response_length_mean": 44.0,
+      "stability/response_length_std": 28.0611572265625,
+      "stability/response_length_var": 787.4285888671875,
+      "stability/token_entropy_mean": 4.42578125,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.5808376669883728,
+      "kl": 0.018133217468857765,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56471566.12413793,
+      "logits/rejected": -52024934.4,
+      "logps/chosen": -378.4438577586207,
+      "logps/rejected": -202.7503125,
+      "loss": 0.9391,
+      "loss/base_kto": 0.49999186396598816,
+      "metrics/advantage_var": 0.06101418659090996,
+      "regularization/grad_norm_sq": 4002.333984375,
+      "regularization/w2": 0.4309103488922119,
+      "rewards/chosen": 0.0047595426954072095,
+      "rewards/margins": 0.0005058412129068605,
+      "rewards/rejected": 0.004253701482500349,
+      "stability/repetition_rate_mean": 0.18583066761493683,
+      "stability/response_length_mean": 43.375,
+      "stability/response_length_std": 28.54538917541504,
+      "stability/response_length_var": 814.8392944335938,
+      "stability/token_entropy_mean": 4.75390625,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.6033640503883362,
+      "kl": 0.022612739354372025,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57729188.939597316,
+      "logits/rejected": -51520841.35672515,
+      "logps/chosen": -468.5989932885906,
+      "logps/rejected": -198.2083561769006,
+      "loss": 0.9337,
+      "loss/base_kto": 0.5006771683692932,
+      "metrics/advantage_var": 0.09293336421251297,
+      "regularization/grad_norm_sq": 4491.3251953125,
+      "regularization/w2": 0.45624828338623047,
+      "rewards/chosen": -0.0017742578615278207,
+      "rewards/margins": -0.005727118693742176,
+      "rewards/rejected": 0.0039528608322143555,
+      "stability/repetition_rate_mean": 0.1583891361951828,
+      "stability/response_length_mean": 43.0,
+      "stability/response_length_std": 29.602123260498047,
+      "stability/response_length_var": 876.2857055664062,
+      "stability/token_entropy_mean": 4.642578125,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6803989410400391,
+      "kl": 0.02071223594248295,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57644589.36708861,
+      "logits/rejected": -53911292.83950617,
+      "logps/chosen": -385.9240506329114,
+      "logps/rejected": -195.117862654321,
+      "loss": 0.938,
+      "loss/base_kto": 0.4999936521053314,
+      "metrics/advantage_var": 0.07669314742088318,
+      "regularization/grad_norm_sq": 3758.585205078125,
+      "regularization/w2": 0.4280441403388977,
+      "rewards/chosen": 0.001247800886631012,
+      "rewards/margins": 0.0003867014138786881,
+      "rewards/rejected": 0.0008610994727523239,
+      "stability/repetition_rate_mean": 0.3108258843421936,
+      "stability/response_length_mean": 50.5,
+      "stability/response_length_std": 25.048524856567383,
+      "stability/response_length_var": 627.4285888671875,
+      "stability/token_entropy_mean": 5.58203125,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.5847389698028564,
+      "kl": 0.02293821983039379,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56634060.8,
+      "logits/rejected": -50219801.6,
+      "logps/chosen": -421.899560546875,
+      "logps/rejected": -219.2647705078125,
+      "loss": 0.9341,
+      "loss/base_kto": 0.5005573630332947,
+      "metrics/advantage_var": 0.06820251047611237,
+      "regularization/grad_norm_sq": 5033.30859375,
+      "regularization/w2": 0.4727831482887268,
+      "rewards/chosen": 0.0003104138188064098,
+      "rewards/margins": -0.0041379273869097235,
+      "rewards/rejected": 0.004448341205716133,
+      "stability/repetition_rate_mean": 0.16694079339504242,
+      "stability/response_length_mean": 35.0,
+      "stability/response_length_std": 31.518701553344727,
+      "stability/response_length_var": 993.4285888671875,
+      "stability/token_entropy_mean": 4.89453125,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.6400073766708374,
+      "kl": 0.0444142147898674,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52707451.70469799,
+      "logits/rejected": -50772524.9122807,
+      "logps/chosen": -352.2749056208054,
+      "logps/rejected": -243.8424707602339,
+      "loss": 0.9389,
+      "loss/base_kto": 0.4996757507324219,
+      "metrics/advantage_var": 0.06386065483093262,
+      "regularization/grad_norm_sq": 3936.931640625,
+      "regularization/w2": 0.42452260851860046,
+      "rewards/chosen": 0.0053554469306997,
+      "rewards/margins": 0.0021862895790329766,
+      "rewards/rejected": 0.0031691573516667237,
+      "stability/repetition_rate_mean": 0.1052611693739891,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.87324333190918,
+      "stability/response_length_var": 892.4107055664062,
+      "stability/token_entropy_mean": 4.771484375,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.5788847208023071,
+      "kl": 0.04537983611226082,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54037710.17449664,
+      "logits/rejected": -48296085.70760234,
+      "logps/chosen": -375.1397336409396,
+      "logps/rejected": -180.18807108918128,
+      "loss": 0.9201,
+      "loss/base_kto": 0.49983420968055725,
+      "metrics/advantage_var": 0.052370816469192505,
+      "regularization/grad_norm_sq": 3871.12744140625,
+      "regularization/w2": 0.4266422390937805,
+      "rewards/chosen": 0.005941886629834271,
+      "rewards/margins": 0.0006887215409353172,
+      "rewards/rejected": 0.005253165088898954,
+      "stability/repetition_rate_mean": 0.3655294179916382,
+      "stability/response_length_mean": 53.375,
+      "stability/response_length_std": 21.480472564697266,
+      "stability/response_length_var": 461.41070556640625,
+      "stability/token_entropy_mean": 3.8212890625,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.6621019244194031,
+      "kl": 0.05794268846511841,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56953277.217391305,
+      "logits/rejected": -49571939.82389937,
+      "logps/chosen": -344.48738354037266,
+      "logps/rejected": -230.3876523191824,
+      "loss": 0.9291,
+      "loss/base_kto": 0.5004059076309204,
+      "metrics/advantage_var": 0.05569034814834595,
+      "regularization/grad_norm_sq": 3976.169921875,
+      "regularization/w2": 0.43878814578056335,
+      "rewards/chosen": 0.0024627199084121987,
+      "rewards/margins": -0.00357790456916621,
+      "rewards/rejected": 0.006040624477578409,
+      "stability/repetition_rate_mean": 0.10418051481246948,
+      "stability/response_length_mean": 22.375,
+      "stability/response_length_std": 30.236862182617188,
+      "stability/response_length_var": 914.2678833007812,
+      "stability/token_entropy_mean": 4.3173828125,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.5812336206436157,
+      "kl": 0.05679145082831383,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52678345.696969695,
+      "logits/rejected": -47821840.51612903,
+      "logps/chosen": -359.6124053030303,
+      "logps/rejected": -178.17867943548387,
+      "loss": 0.933,
+      "loss/base_kto": 0.4992280900478363,
+      "metrics/advantage_var": 0.057238996028900146,
+      "regularization/grad_norm_sq": 3165.790283203125,
+      "regularization/w2": 0.3937515616416931,
+      "rewards/chosen": 0.010837621399850556,
+      "rewards/margins": 0.00601367213975416,
+      "rewards/rejected": 0.004823949260096396,
+      "stability/repetition_rate_mean": 0.22545291483402252,
+      "stability/response_length_mean": 44.625,
+      "stability/response_length_std": 27.8820858001709,
+      "stability/response_length_var": 777.4107055664062,
+      "stability/token_entropy_mean": 4.646484375,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.5738043785095215,
+      "kl": 0.06266746670007706,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57563182.24516129,
+      "logits/rejected": -47268345.7939394,
+      "logps/chosen": -381.86370967741937,
+      "logps/rejected": -183.5180516098485,
+      "loss": 0.939,
+      "loss/base_kto": 0.4997069537639618,
+      "metrics/advantage_var": 0.07446863502264023,
+      "regularization/grad_norm_sq": 3868.786865234375,
+      "regularization/w2": 0.4315285384654999,
+      "rewards/chosen": 0.008308312200730847,
+      "rewards/margins": 0.001878772895939772,
+      "rewards/rejected": 0.006429539304791075,
+      "stability/repetition_rate_mean": 0.19313038885593414,
+      "stability/response_length_mean": 20.625,
+      "stability/response_length_std": 27.176342010498047,
+      "stability/response_length_var": 738.5535888671875,
+      "stability/token_entropy_mean": 5.55859375,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5531291365623474,
+      "kl": 0.08709845691919327,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52753698.68387097,
+      "logits/rejected": -49023162.18181818,
+      "logps/chosen": -378.9974798387097,
+      "logps/rejected": -220.35189393939393,
+      "loss": 0.9497,
+      "loss/base_kto": 0.49925193190574646,
+      "metrics/advantage_var": 0.08100283890962601,
+      "regularization/grad_norm_sq": 4483.40087890625,
+      "regularization/w2": 0.45682960748672485,
+      "rewards/chosen": 0.01309710318042386,
+      "rewards/margins": 0.006069044376981108,
+      "rewards/rejected": 0.007028058803442753,
+      "stability/repetition_rate_mean": 0.111328125,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.3388671875,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.611277163028717,
+      "kl": 0.09017860144376755,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49498509.83783784,
+      "logits/rejected": -47885377.488372095,
+      "logps/chosen": -314.13988597972974,
+      "logps/rejected": -228.78901707848837,
+      "loss": 0.9402,
+      "loss/base_kto": 0.49950870871543884,
+      "metrics/advantage_var": 0.06859137862920761,
+      "regularization/grad_norm_sq": 3736.681396484375,
+      "regularization/w2": 0.42404261231422424,
+      "rewards/chosen": 0.013092591955855087,
+      "rewards/margins": 0.0035108895627003218,
+      "rewards/rejected": 0.009581702393154765,
+      "stability/repetition_rate_mean": 0.10066105425357819,
+      "stability/response_length_mean": 22.75,
+      "stability/response_length_std": 31.07479476928711,
+      "stability/response_length_var": 965.6428833007812,
+      "stability/token_entropy_mean": 2.130859375,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.6037386655807495,
+      "kl": 0.13238738477230072,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58626637.98675497,
+      "logits/rejected": -53447752.71005917,
+      "logps/chosen": -403.87065397350995,
+      "logps/rejected": -200.96398853550295,
+      "loss": 0.9316,
+      "loss/base_kto": 0.4992172420024872,
+      "metrics/advantage_var": 0.07778961956501007,
+      "regularization/grad_norm_sq": 3624.650390625,
+      "regularization/w2": 0.42037779092788696,
+      "rewards/chosen": 0.0159459240389186,
+      "rewards/margins": 0.006766194357169991,
+      "rewards/rejected": 0.00917972968174861,
+      "stability/repetition_rate_mean": 0.19088542461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.23828125,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.6116781234741211,
+      "kl": 0.12365061044692993,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51478686.55483871,
+      "logits/rejected": -47141816.63030303,
+      "logps/chosen": -358.13480342741934,
+      "logps/rejected": -230.34330018939394,
+      "loss": 0.9311,
+      "loss/base_kto": 0.4985641539096832,
+      "metrics/advantage_var": 0.10404322296380997,
+      "regularization/grad_norm_sq": 3923.891357421875,
+      "regularization/w2": 0.43580102920532227,
+      "rewards/chosen": 0.01977017310357863,
+      "rewards/margins": 0.011619663471583632,
+      "rewards/rejected": 0.008150509631994998,
+      "stability/repetition_rate_mean": 0.103515625,
+      "stability/response_length_mean": 32.5,
+      "stability/response_length_std": 33.67915725708008,
+      "stability/response_length_var": 1134.2857666015625,
+      "stability/token_entropy_mean": 4.9375,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.6381218433380127,
+      "kl": 0.16665403544902802,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58412438.974358976,
+      "logits/rejected": -50588893.65853658,
+      "logps/chosen": -380.5612980769231,
+      "logps/rejected": -187.1324433117378,
+      "loss": 0.9286,
+      "loss/base_kto": 0.4982132017612457,
+      "metrics/advantage_var": 0.08798178285360336,
+      "regularization/grad_norm_sq": 3925.050048828125,
+      "regularization/w2": 0.43326812982559204,
+      "rewards/chosen": 0.022810822878128443,
+      "rewards/margins": 0.013375084872839225,
+      "rewards/rejected": 0.009435738005289218,
+      "stability/repetition_rate_mean": 0.3356415629386902,
+      "stability/response_length_mean": 30.375,
+      "stability/response_length_std": 29.061203002929688,
+      "stability/response_length_var": 844.5535888671875,
+      "stability/token_entropy_mean": 3.72265625,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.6163039803504944,
+      "kl": 0.15967555344104767,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50379520.0,
+      "logits/rejected": -48967028.938271604,
+      "logps/chosen": -314.77383306962025,
+      "logps/rejected": -207.04853877314815,
+      "loss": 0.9271,
+      "loss/base_kto": 0.4984891414642334,
+      "metrics/advantage_var": 0.1260511875152588,
+      "regularization/grad_norm_sq": 3692.677490234375,
+      "regularization/w2": 0.42039623856544495,
+      "rewards/chosen": 0.021952728681926485,
+      "rewards/margins": 0.012433580661829717,
+      "rewards/rejected": 0.009519148020096768,
+      "stability/repetition_rate_mean": 0.19374233484268188,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 31.1264705657959,
+      "stability/response_length_var": 968.8571166992188,
+      "stability/token_entropy_mean": 3.75,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.6088441014289856,
+      "kl": 0.1950296312570572,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54698091.24324324,
+      "logits/rejected": -50908255.255813956,
+      "logps/chosen": -359.5659311655405,
+      "logps/rejected": -216.3930323401163,
+      "loss": 0.9328,
+      "loss/base_kto": 0.49854350090026855,
+      "metrics/advantage_var": 0.13550297915935516,
+      "regularization/grad_norm_sq": 3865.565673828125,
+      "regularization/w2": 0.42957553267478943,
+      "rewards/chosen": 0.03042653444651011,
+      "rewards/margins": 0.011843225627781683,
+      "rewards/rejected": 0.018583308818728425,
+      "stability/repetition_rate_mean": 0.228515625,
+      "stability/response_length_mean": 41.25,
+      "stability/response_length_std": 31.476749420166016,
+      "stability/response_length_var": 990.7857055664062,
+      "stability/token_entropy_mean": 4.078125,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6745646595954895,
+      "kl": 0.21558904647827148,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59378317.87951807,
+      "logits/rejected": -53788419.32467532,
+      "logps/chosen": -399.4925169427711,
+      "logps/rejected": -209.84361049107142,
+      "loss": 0.9365,
+      "loss/base_kto": 0.49687233567237854,
+      "metrics/advantage_var": 0.2065424919128418,
+      "regularization/grad_norm_sq": 3994.716064453125,
+      "regularization/w2": 0.4382788836956024,
+      "rewards/chosen": 0.04099334004413651,
+      "rewards/margins": 0.024971716804247736,
+      "rewards/rejected": 0.016021623239888774,
+      "stability/repetition_rate_mean": 0.1877790093421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.353515625,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.6502033472061157,
+      "kl": 0.26441773772239685,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57391930.60240964,
+      "logits/rejected": -53613953.66233766,
+      "logps/chosen": -378.218938253012,
+      "logps/rejected": -205.88195008116884,
+      "loss": 0.9259,
+      "loss/base_kto": 0.49810346961021423,
+      "metrics/advantage_var": 0.21781854331493378,
+      "regularization/grad_norm_sq": 3637.52197265625,
+      "regularization/w2": 0.42161688208580017,
+      "rewards/chosen": 0.0344322733132236,
+      "rewards/margins": 0.013813094566021703,
+      "rewards/rejected": 0.020619178747201896,
+      "stability/repetition_rate_mean": 0.30853456258773804,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.911476135253906,
+      "stability/response_length_var": 894.6964111328125,
+      "stability/token_entropy_mean": 4.404296875,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.6496742367744446,
+      "kl": 0.3085666298866272,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60598233.11392405,
+      "logits/rejected": -53416372.14814815,
+      "logps/chosen": -364.10173556170884,
+      "logps/rejected": -200.97486255787038,
+      "loss": 0.9207,
+      "loss/base_kto": 0.4965952932834625,
+      "metrics/advantage_var": 0.25335976481437683,
+      "regularization/grad_norm_sq": 3352.978271484375,
+      "regularization/w2": 0.4012783169746399,
+      "rewards/chosen": 0.04908096337620216,
+      "rewards/margins": 0.02691114084816366,
+      "rewards/rejected": 0.022169822528038497,
+      "stability/repetition_rate_mean": 0.2194010466337204,
+      "stability/response_length_mean": 49.25,
+      "stability/response_length_std": 27.316923141479492,
+      "stability/response_length_var": 746.2142944335938,
+      "stability/token_entropy_mean": 4.75,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.6193932294845581,
+      "kl": 0.3443666994571686,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57739732.49673203,
+      "logits/rejected": -54352828.5508982,
+      "logps/chosen": -383.7316687091503,
+      "logps/rejected": -211.93726609281438,
+      "loss": 0.936,
+      "loss/base_kto": 0.49703770875930786,
+      "metrics/advantage_var": 0.3692956864833832,
+      "regularization/grad_norm_sq": 4454.72216796875,
+      "regularization/w2": 0.4518660008907318,
+      "rewards/chosen": 0.05132290896247415,
+      "rewards/margins": 0.023414049823756954,
+      "rewards/rejected": 0.027908859138717195,
+      "stability/repetition_rate_mean": 0.15568576753139496,
+      "stability/response_length_mean": 27.625,
+      "stability/response_length_std": 30.486238479614258,
+      "stability/response_length_var": 929.4107055664062,
+      "stability/token_entropy_mean": 4.693359375,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.6110739707946777,
+      "kl": 0.3985693156719208,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60667195.61643836,
+      "logits/rejected": -55641405.79310345,
+      "logps/chosen": -416.79403895547944,
+      "logps/rejected": -216.0777658045977,
+      "loss": 0.9231,
+      "loss/base_kto": 0.49641838669776917,
+      "metrics/advantage_var": 0.559248149394989,
+      "regularization/grad_norm_sq": 3774.068115234375,
+      "regularization/w2": 0.4254666864871979,
+      "rewards/chosen": 0.05853001058918156,
+      "rewards/margins": 0.029587951762866566,
+      "rewards/rejected": 0.028942058826314992,
+      "stability/repetition_rate_mean": 0.19144965708255768,
+      "stability/response_length_mean": 23.125,
+      "stability/response_length_std": 29.347366333007812,
+      "stability/response_length_var": 861.2678833007812,
+      "stability/token_entropy_mean": 3.353515625,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.5931130051612854,
+      "kl": 0.45674076676368713,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57375400.464516126,
+      "logits/rejected": -53823680.38787879,
+      "logps/chosen": -391.9906754032258,
+      "logps/rejected": -183.6497159090909,
+      "loss": 0.9394,
+      "loss/base_kto": 0.49498748779296875,
+      "metrics/advantage_var": 0.7409353256225586,
+      "regularization/grad_norm_sq": 3780.29150390625,
+      "regularization/w2": 0.42766115069389343,
+      "rewards/chosen": 0.06672527559341923,
+      "rewards/margins": 0.039806725622272214,
+      "rewards/rejected": 0.026918549971147017,
+      "stability/repetition_rate_mean": 0.24460937082767487,
+      "stability/response_length_mean": 43.25,
+      "stability/response_length_std": 29.620214462280273,
+      "stability/response_length_var": 877.3571166992188,
+      "stability/token_entropy_mean": 3.640625,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5910015106201172,
+      "kl": 0.5136879086494446,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59843408.09815951,
+      "logits/rejected": -55000846.67515924,
+      "logps/chosen": -400.98552530674846,
+      "logps/rejected": -247.51219148089172,
+      "loss": 0.9147,
+      "loss/base_kto": 0.49399566650390625,
+      "metrics/advantage_var": 0.9681418538093567,
+      "regularization/grad_norm_sq": 4315.08056640625,
+      "regularization/w2": 0.4519772231578827,
+      "rewards/chosen": 0.0830619554578161,
+      "rewards/margins": 0.04720748222231525,
+      "rewards/rejected": 0.035854473235500846,
+      "stability/repetition_rate_mean": 0.2230902761220932,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.936838150024414,
+      "stability/response_length_var": 896.2142944335938,
+      "stability/token_entropy_mean": 4.05078125,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.7235668301582336,
+      "kl": 0.4908495843410492,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57076500.16969697,
+      "logits/rejected": -50077570.477419354,
+      "logps/chosen": -366.3874053030303,
+      "logps/rejected": -189.92144657258063,
+      "loss": 0.93,
+      "loss/base_kto": 0.49192824959754944,
+      "metrics/advantage_var": 1.9476429224014282,
+      "regularization/grad_norm_sq": 5980.30029296875,
+      "regularization/w2": 0.48489466309547424,
+      "rewards/chosen": 0.08023344097715436,
+      "rewards/margins": 0.06893024071686896,
+      "rewards/rejected": 0.011303200260285409,
+      "stability/repetition_rate_mean": 0.302734375,
+      "stability/response_length_mean": 56.25,
+      "stability/response_length_std": 21.920310974121094,
+      "stability/response_length_var": 480.5,
+      "stability/token_entropy_mean": 5.001953125,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.6801520586013794,
+      "kl": 0.577669084072113,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51146073.44578313,
+      "logits/rejected": -51662515.53246753,
+      "logps/chosen": -329.4719267695783,
+      "logps/rejected": -199.15390371347402,
+      "loss": 0.9176,
+      "loss/base_kto": 0.49409326910972595,
+      "metrics/advantage_var": 0.8393446207046509,
+      "regularization/grad_norm_sq": 3416.243408203125,
+      "regularization/w2": 0.40856075286865234,
+      "rewards/chosen": 0.0799884566341538,
+      "rewards/margins": 0.046587763137053366,
+      "rewards/rejected": 0.03340069349710043,
+      "stability/repetition_rate_mean": 0.21016408503055573,
+      "stability/response_length_mean": 22.625,
+      "stability/response_length_std": 26.359804153442383,
+      "stability/response_length_var": 694.8392944335938,
+      "stability/token_entropy_mean": 5.66015625,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.6449370384216309,
+      "kl": 0.6454542875289917,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55421090.63529412,
+      "logits/rejected": -50187294.72,
+      "logps/chosen": -396.8174632352941,
+      "logps/rejected": -207.35263020833332,
+      "loss": 0.923,
+      "loss/base_kto": 0.4903319776058197,
+      "metrics/advantage_var": 1.2783325910568237,
+      "regularization/grad_norm_sq": 4074.032470703125,
+      "regularization/w2": 0.4475216865539551,
+      "rewards/chosen": 0.11169673695283777,
+      "rewards/margins": 0.0743670695435767,
+      "rewards/rejected": 0.037329667409261066,
+      "stability/repetition_rate_mean": 0.23066024482250214,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.193626403808594,
+      "stability/response_length_var": 852.2678833007812,
+      "stability/token_entropy_mean": 4.6015625,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.6441305875778198,
+      "kl": 0.6456610560417175,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54003495.152941175,
+      "logits/rejected": -49346880.85333333,
+      "logps/chosen": -360.09657628676473,
+      "logps/rejected": -191.00634114583335,
+      "loss": 0.9163,
+      "loss/base_kto": 0.4928789734840393,
+      "metrics/advantage_var": 2.1478357315063477,
+      "regularization/grad_norm_sq": 4929.31494140625,
+      "regularization/w2": 0.46441006660461426,
+      "rewards/chosen": 0.08927134345559513,
+      "rewards/margins": 0.059344200994454185,
+      "rewards/rejected": 0.029927142461140952,
+      "stability/repetition_rate_mean": 0.2978527545928955,
+      "stability/response_length_mean": 58.125,
+      "stability/response_length_std": 10.960155487060547,
+      "stability/response_length_var": 120.125,
+      "stability/token_entropy_mean": 4.56640625,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.6161492466926575,
+      "kl": 0.7202507257461548,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55241313.52380952,
+      "logits/rejected": -48956978.526315786,
+      "logps/chosen": -415.5804501488095,
+      "logps/rejected": -186.47481496710526,
+      "loss": 0.9248,
+      "loss/base_kto": 0.48859721422195435,
+      "metrics/advantage_var": 1.9541778564453125,
+      "regularization/grad_norm_sq": 4000.94580078125,
+      "regularization/w2": 0.43814727663993835,
+      "rewards/chosen": 0.1314907301039923,
+      "rewards/margins": 0.08923867576403129,
+      "rewards/rejected": 0.042252054339961,
+      "stability/repetition_rate_mean": 0.31046873331069946,
+      "stability/response_length_mean": 32.5,
+      "stability/response_length_std": 30.528675079345703,
+      "stability/response_length_var": 932.0,
+      "stability/token_entropy_mean": 4.3720703125,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.7250444293022156,
+      "kl": 0.7532386779785156,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54405402.48275862,
+      "logits/rejected": -49781749.47945205,
+      "logps/chosen": -369.750628591954,
+      "logps/rejected": -212.1415881849315,
+      "loss": 0.9192,
+      "loss/base_kto": 0.49109888076782227,
+      "metrics/advantage_var": 1.7532415390014648,
+      "regularization/grad_norm_sq": 3549.75244140625,
+      "regularization/w2": 0.4151502549648285,
+      "rewards/chosen": 0.11661315786427465,
+      "rewards/margins": 0.07156126962048184,
+      "rewards/rejected": 0.04505188824379281,
+      "stability/repetition_rate_mean": 0.22421875596046448,
+      "stability/response_length_mean": 19.875,
+      "stability/response_length_std": 27.771709442138672,
+      "stability/response_length_var": 771.2678833007812,
+      "stability/token_entropy_mean": 4.890625,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.6213644742965698,
+      "kl": 0.9238977432250977,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59125432.80473373,
+      "logits/rejected": -52567887.682119206,
+      "logps/chosen": -439.6600406804734,
+      "logps/rejected": -190.58578228476821,
+      "loss": 0.905,
+      "loss/base_kto": 0.4874550998210907,
+      "metrics/advantage_var": 2.416893720626831,
+      "regularization/grad_norm_sq": 3930.178955078125,
+      "regularization/w2": 0.4287300109863281,
+      "rewards/chosen": 0.16818395309899686,
+      "rewards/margins": 0.10098015347862745,
+      "rewards/rejected": 0.06720379962036942,
+      "stability/repetition_rate_mean": 0.23209978640079498,
+      "stability/response_length_mean": 36.25,
+      "stability/response_length_std": 30.76988410949707,
+      "stability/response_length_var": 946.7857055664062,
+      "stability/token_entropy_mean": 3.67822265625,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.7216498851776123,
+      "kl": 0.9429834485054016,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56652960.38554217,
+      "logits/rejected": -48022956.883116886,
+      "logps/chosen": -354.89241340361446,
+      "logps/rejected": -194.30808644480518,
+      "loss": 0.911,
+      "loss/base_kto": 0.48693132400512695,
+      "metrics/advantage_var": 4.248536586761475,
+      "regularization/grad_norm_sq": 4139.484375,
+      "regularization/w2": 0.4321208894252777,
+      "rewards/chosen": 0.15448007238916603,
+      "rewards/margins": 0.1045362072695292,
+      "rewards/rejected": 0.04994386511963683,
+      "stability/repetition_rate_mean": 0.1976061761379242,
+      "stability/response_length_mean": 30.5,
+      "stability/response_length_std": 26.42509651184082,
+      "stability/response_length_var": 698.2857055664062,
+      "stability/token_entropy_mean": 4.8369140625,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.6428859233856201,
+      "kl": 1.0304526090621948,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52658986.40490798,
+      "logits/rejected": -47980801.63057325,
+      "logps/chosen": -374.418615797546,
+      "logps/rejected": -210.53871417197453,
+      "loss": 0.9028,
+      "loss/base_kto": 0.4877626597881317,
+      "metrics/advantage_var": 4.026914119720459,
+      "regularization/grad_norm_sq": 3664.698974609375,
+      "regularization/w2": 0.41559115052223206,
+      "rewards/chosen": 0.1727198852351838,
+      "rewards/margins": 0.09677210547646317,
+      "rewards/rejected": 0.07594777975872063,
+      "stability/repetition_rate_mean": 0.25664061307907104,
+      "stability/response_length_mean": 34.75,
+      "stability/response_length_std": 31.54928207397461,
+      "stability/response_length_var": 995.3571166992188,
+      "stability/token_entropy_mean": 3.3203125,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.6485155820846558,
+      "kl": 1.06327223777771,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54608421.925925925,
+      "logits/rejected": -49521022.37974683,
+      "logps/chosen": -310.32325424382714,
+      "logps/rejected": -196.57655508306962,
+      "loss": 0.9317,
+      "loss/base_kto": 0.48661017417907715,
+      "metrics/advantage_var": 7.680975437164307,
+      "regularization/grad_norm_sq": 7077.32763671875,
+      "regularization/w2": 0.47685596346855164,
+      "rewards/chosen": 0.16781084037121433,
+      "rewards/margins": 0.12155084260498517,
+      "rewards/rejected": 0.04625999776622917,
+      "stability/repetition_rate_mean": 0.19140625,
+      "stability/response_length_mean": 32.875,
+      "stability/response_length_std": 33.30996322631836,
+      "stability/response_length_var": 1109.5535888671875,
+      "stability/token_entropy_mean": 3.701171875,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.6963216662406921,
+      "kl": 1.1201375722885132,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53089470.08383234,
+      "logits/rejected": -49691681.46405229,
+      "logps/chosen": -390.30389221556885,
+      "logps/rejected": -210.46195874183007,
+      "loss": 0.9237,
+      "loss/base_kto": 0.4832589626312256,
+      "metrics/advantage_var": 7.02877950668335,
+      "regularization/grad_norm_sq": 5314.27880859375,
+      "regularization/w2": 0.4580075442790985,
+      "rewards/chosen": 0.2101439972837528,
+      "rewards/margins": 0.1364806098605055,
+      "rewards/rejected": 0.0736633874232473,
+      "stability/repetition_rate_mean": 0.1987847238779068,
+      "stability/response_length_mean": 19.125,
+      "stability/response_length_std": 28.296579360961914,
+      "stability/response_length_var": 800.6964111328125,
+      "stability/token_entropy_mean": 4.759765625,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.6389023065567017,
+      "kl": 1.1507009267807007,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60032350.14193548,
+      "logits/rejected": -53812130.90909091,
+      "logps/chosen": -390.8807459677419,
+      "logps/rejected": -197.07857481060606,
+      "loss": 0.9173,
+      "loss/base_kto": 0.4865249693393707,
+      "metrics/advantage_var": 4.4115118980407715,
+      "regularization/grad_norm_sq": 3978.581298828125,
+      "regularization/w2": 0.43020516633987427,
+      "rewards/chosen": 0.1904176281344506,
+      "rewards/margins": 0.10634150966521232,
+      "rewards/rejected": 0.08407611846923828,
+      "stability/repetition_rate_mean": 0.3856026828289032,
+      "stability/response_length_mean": 49.0,
+      "stability/response_length_std": 27.82085609436035,
+      "stability/response_length_var": 774.0,
+      "stability/token_entropy_mean": 5.1796875,
+      "step": 195
+    },
+    {
+      "epoch": 1.0,
+      "step": 196,
+      "total_flos": 3.1861866257511875e+18,
+      "train_loss": 0.9280117068971906,
+      "train_runtime": 4303.5052,
+      "train_samples_per_second": 23.237,
+      "train_steps_per_second": 0.046
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 196,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 200,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.1861866257511875e+18,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}

mistral7b_kto_w2_delta0.005/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b00a81f45dd8c085139037c534b1aff270699bd4417624f905a02a36fe627e05
+size 8081