Upload folder using huggingface_hub

mistral7b_kto_w2_delta0.001/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_delta0.001
+  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_delta0.001
+
+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_delta0.001/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": [
+    "gate_proj",
+    "down_proj",
+    "up_proj",
+    "q_proj",
+    "k_proj",
+    "o_proj",
+    "v_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

mistral7b_kto_w2_delta0.001/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87fd19c040097e11b7423fbb56ee2a5a2bbd7ece255a0895affdcff3a7497b83
+size 42002584

mistral7b_kto_w2_delta0.001/all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.5821207132266493,
+    "train_runtime": 4296.589,
+    "train_samples_per_second": 23.274,
+    "train_steps_per_second": 0.046
+}

mistral7b_kto_w2_delta0.001/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_delta0.001/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_delta0.001/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": [
+    "gate_proj",
+    "down_proj",
+    "up_proj",
+    "q_proj",
+    "k_proj",
+    "o_proj",
+    "v_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

mistral7b_kto_w2_delta0.001/checkpoint-196/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87fd19c040097e11b7423fbb56ee2a5a2bbd7ece255a0895affdcff3a7497b83
+size 42002584

mistral7b_kto_w2_delta0.001/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_delta0.001/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 sha256:51932fbb0b2ec1f77ef6f5a87e75bf9ecfd2c45c215f2c583d376a23c002bd98
+size 31467333

mistral7b_kto_w2_delta0.001/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 sha256:7cd238b5f26b477fc2a0a1976fe7d53ae32e34f0b22e2c53902dd38c79e6cd68
+size 31467333

mistral7b_kto_w2_delta0.001/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 sha256:26ff1ecc669a44ba6e03a56f290b0149f269e368ef3acd1bc8a1a8ed9bac1dcd
+size 31467333

mistral7b_kto_w2_delta0.001/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 sha256:405fb0fa8f3f1949767034a4104d9da6a8f74a887f2000ef4029cf56d6e044a2
+size 31467397

mistral7b_kto_w2_delta0.001/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 sha256:48b53dce4f43b501176d56ec5c0f7d90547afaf21c6b7de1d43ea429493374f9
+size 31467397

mistral7b_kto_w2_delta0.001/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:dae9d8ddf051ccadabbb76f48096a1e3b3d7f61a39198505d323af41bc8661cc
+size 31467397

mistral7b_kto_w2_delta0.001/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 sha256:9a0dfc4d2769cafede5557cde1010f76e589433ac81b0f46e3b0d6a7521b5afb
+size 31467397

mistral7b_kto_w2_delta0.001/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:b1a5da139fb7f8918236b19270d940cb2a432e544baa4ca12dc4a1f57041ce32
+size 31467397

mistral7b_kto_w2_delta0.001/checkpoint-196/global_step196/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:367b736115265c9830260836e27d9579ed26f1eced903c6ff02f3ca072ac708b
+size 42288409

mistral7b_kto_w2_delta0.001/checkpoint-196/latest

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

mistral7b_kto_w2_delta0.001/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_delta0.001/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_delta0.001/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_delta0.001/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_delta0.001/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_delta0.001/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_delta0.001/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_delta0.001/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_delta0.001/checkpoint-196/scheduler.pt

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

mistral7b_kto_w2_delta0.001/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_delta0.001/checkpoint-196/tokenizer.model

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

mistral7b_kto_w2_delta0.001/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_delta0.001/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": 0.5704261660575867,
+      "kl": 0.009683904238045216,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59713832.421052635,
+      "logits/rejected": -50416451.047619045,
+      "logps/chosen": -439.0024157072368,
+      "logps/rejected": -198.41252790178572,
+      "loss": 0.5885,
+      "loss/base_kto": 0.4994261860847473,
+      "metrics/advantage_var": 0.04821223020553589,
+      "regularization/grad_norm_sq": 4308.1396484375,
+      "regularization/w2": 0.09046166390180588,
+      "rewards/chosen": 0.004319707029744198,
+      "rewards/margins": 0.004632244073321347,
+      "rewards/rejected": -0.0003125370435771488,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.6019414663314819,
+      "kl": 0.0205225832760334,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59306230.02597403,
+      "logits/rejected": -50024096.38554217,
+      "logps/chosen": -429.9021915584416,
+      "logps/rejected": -192.85753953313252,
+      "loss": 0.586,
+      "loss/base_kto": 0.4996527135372162,
+      "metrics/advantage_var": 0.06569447368383408,
+      "regularization/grad_norm_sq": 3819.715576171875,
+      "regularization/w2": 0.08616509288549423,
+      "rewards/chosen": 0.003629611297087236,
+      "rewards/margins": 0.002590525050740383,
+      "rewards/rejected": 0.001039086246346853,
+      "stability/repetition_rate_mean": 0.3551284670829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.4365234375,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.6072039604187012,
+      "kl": 0.011671505868434906,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56229478.4,
+      "logits/rejected": -51918392.15483871,
+      "logps/chosen": -404.1024621212121,
+      "logps/rejected": -200.67379032258066,
+      "loss": 0.5856,
+      "loss/base_kto": 0.49987754225730896,
+      "metrics/advantage_var": 0.05594036728143692,
+      "regularization/grad_norm_sq": 4186.52001953125,
+      "regularization/w2": 0.08851373195648193,
+      "rewards/chosen": 0.002835910970514471,
+      "rewards/margins": 0.0008086012605348293,
+      "rewards/rejected": 0.0020273097099796416,
+      "stability/repetition_rate_mean": 0.2763020694255829,
+      "stability/response_length_mean": 28.5,
+      "stability/response_length_std": 30.10457992553711,
+      "stability/response_length_var": 906.2857055664062,
+      "stability/token_entropy_mean": 3.75390625,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.6190371513366699,
+      "kl": 0.01878364384174347,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51667321.6,
+      "logits/rejected": -48058720.0,
+      "logps/chosen": -343.685205078125,
+      "logps/rejected": -225.618603515625,
+      "loss": 0.5864,
+      "loss/base_kto": 0.5004865527153015,
+      "metrics/advantage_var": 0.055938970297575,
+      "regularization/grad_norm_sq": 3616.344970703125,
+      "regularization/w2": 0.08394529670476913,
+      "rewards/chosen": 0.0011490656994283198,
+      "rewards/margins": -0.003325168509036303,
+      "rewards/rejected": 0.004474234208464623,
+      "stability/repetition_rate_mean": 0.1925223171710968,
+      "stability/response_length_mean": 49.75,
+      "stability/response_length_std": 26.649845123291016,
+      "stability/response_length_var": 710.2142944335938,
+      "stability/token_entropy_mean": 4.154296875,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.5828595161437988,
+      "kl": 0.018041405826807022,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56484976.993103445,
+      "logits/rejected": -52025847.22285714,
+      "logps/chosen": -378.47941810344827,
+      "logps/rejected": -202.76964285714286,
+      "loss": 0.5883,
+      "loss/base_kto": 0.5001233220100403,
+      "metrics/advantage_var": 0.07194186747074127,
+      "regularization/grad_norm_sq": 4000.53515625,
+      "regularization/w2": 0.08617646992206573,
+      "rewards/chosen": 0.0012034501495032475,
+      "rewards/margins": -0.001115835549796156,
+      "rewards/rejected": 0.0023192856992994035,
+      "stability/repetition_rate_mean": 0.18387754261493683,
+      "stability/response_length_mean": 43.375,
+      "stability/response_length_std": 28.54538917541504,
+      "stability/response_length_var": 814.8392944335938,
+      "stability/token_entropy_mean": 4.73828125,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.6027837991714478,
+      "kl": 0.01663057506084442,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57733023.7852349,
+      "logits/rejected": -51524224.74853801,
+      "logps/chosen": -468.56889681208054,
+      "logps/rejected": -198.22884685672514,
+      "loss": 0.5872,
+      "loss/base_kto": 0.5000787377357483,
+      "metrics/advantage_var": 0.08475451171398163,
+      "regularization/grad_norm_sq": 4545.20458984375,
+      "regularization/w2": 0.09177713841199875,
+      "rewards/chosen": 0.001230826113848078,
+      "rewards/margins": -0.0006731302826898365,
+      "rewards/rejected": 0.0019039563965379145,
+      "stability/repetition_rate_mean": 0.1875,
+      "stability/response_length_mean": 48.375,
+      "stability/response_length_std": 28.93311309814453,
+      "stability/response_length_var": 837.125,
+      "stability/token_entropy_mean": 4.52734375,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6801803708076477,
+      "kl": 0.016775427386164665,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57625172.25316456,
+      "logits/rejected": -53896609.18518519,
+      "logps/chosen": -385.9157189477848,
+      "logps/rejected": -195.12061149691357,
+      "loss": 0.5878,
+      "loss/base_kto": 0.4998288154602051,
+      "metrics/advantage_var": 0.062246084213256836,
+      "regularization/grad_norm_sq": 3760.884521484375,
+      "regularization/w2": 0.08562777936458588,
+      "rewards/chosen": 0.002080178713496727,
+      "rewards/margins": 0.0014936748957853799,
+      "rewards/rejected": 0.0005865038177113475,
+      "stability/repetition_rate_mean": 0.2854352593421936,
+      "stability/response_length_mean": 50.5,
+      "stability/response_length_std": 25.048524856567383,
+      "stability/response_length_var": 627.4285888671875,
+      "stability/token_entropy_mean": 5.5234375,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.5847268104553223,
+      "kl": 0.017838044092059135,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56634124.8,
+      "logits/rejected": -50225545.6,
+      "logps/chosen": -421.87763671875,
+      "logps/rejected": -219.2741943359375,
+      "loss": 0.5859,
+      "loss/base_kto": 0.5001553297042847,
+      "metrics/advantage_var": 0.06722141057252884,
+      "regularization/grad_norm_sq": 4253.48828125,
+      "regularization/w2": 0.0906253382563591,
+      "rewards/chosen": 0.002502601221203804,
+      "rewards/margins": -0.0010038755834102628,
+      "rewards/rejected": 0.003506476804614067,
+      "stability/repetition_rate_mean": 0.16303454339504242,
+      "stability/response_length_mean": 35.0,
+      "stability/response_length_std": 31.518701553344727,
+      "stability/response_length_var": 993.4285888671875,
+      "stability/token_entropy_mean": 4.8115234375,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.6392990946769714,
+      "kl": 0.044592004269361496,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52689789.422818795,
+      "logits/rejected": -50767135.438596494,
+      "logps/chosen": -352.29244442114094,
+      "logps/rejected": -243.81542397660817,
+      "loss": 0.5871,
+      "loss/base_kto": 0.5003380179405212,
+      "metrics/advantage_var": 0.0578509084880352,
+      "regularization/grad_norm_sq": 3660.001953125,
+      "regularization/w2": 0.08328904211521149,
+      "rewards/chosen": 0.0036000581395706074,
+      "rewards/margins": -0.0022749988076132804,
+      "rewards/rejected": 0.005875056947183888,
+      "stability/repetition_rate_mean": 0.0915892943739891,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.51875877380371,
+      "stability/response_length_var": 871.3571166992188,
+      "stability/token_entropy_mean": 4.705078125,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.5789488554000854,
+      "kl": 0.04660215601325035,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54035864.912751675,
+      "logits/rejected": -48294564.67836257,
+      "logps/chosen": -375.1398385067114,
+      "logps/rejected": -180.18667763157896,
+      "loss": 0.5843,
+      "loss/base_kto": 0.4998607635498047,
+      "metrics/advantage_var": 0.06304984539747238,
+      "regularization/grad_norm_sq": 3861.982421875,
+      "regularization/w2": 0.08533728122711182,
+      "rewards/chosen": 0.005930147715062903,
+      "rewards/margins": 0.0005367304570512354,
+      "rewards/rejected": 0.005393417258011668,
+      "stability/repetition_rate_mean": 0.4206075370311737,
+      "stability/response_length_mean": 55.375,
+      "stability/response_length_std": 16.361431121826172,
+      "stability/response_length_var": 267.6964416503906,
+      "stability/token_entropy_mean": 3.806640625,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.6636475324630737,
+      "kl": 0.04949667677283287,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56956520.94409938,
+      "logits/rejected": -49569669.635220125,
+      "logps/chosen": -344.4948078416149,
+      "logps/rejected": -230.40541470125785,
+      "loss": 0.5859,
+      "loss/base_kto": 0.5003251433372498,
+      "metrics/advantage_var": 0.05528384447097778,
+      "regularization/grad_norm_sq": 3821.798583984375,
+      "regularization/w2": 0.08618461340665817,
+      "rewards/chosen": 0.0017190375683470542,
+      "rewards/margins": -0.0025420666225273456,
+      "rewards/rejected": 0.0042611041908743996,
+      "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": 4.3310546875,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.5860218405723572,
+      "kl": 0.052175115793943405,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52664307.587878786,
+      "logits/rejected": -47816974.86451613,
+      "logps/chosen": -359.62769886363634,
+      "logps/rejected": -178.1703377016129,
+      "loss": 0.5865,
+      "loss/base_kto": 0.4995388090610504,
+      "metrics/advantage_var": 0.05257711559534073,
+      "regularization/grad_norm_sq": 3147.68408203125,
+      "regularization/w2": 0.07853565365076065,
+      "rewards/chosen": 0.009310866124702222,
+      "rewards/margins": 0.003652031144322071,
+      "rewards/rejected": 0.005658834980380151,
+      "stability/repetition_rate_mean": 0.19273173809051514,
+      "stability/response_length_mean": 51.125,
+      "stability/response_length_std": 21.90523338317871,
+      "stability/response_length_var": 479.83929443359375,
+      "stability/token_entropy_mean": 5.384765625,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.5740265846252441,
+      "kl": 0.08242836594581604,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57557335.535483874,
+      "logits/rejected": -47273723.345454544,
+      "logps/chosen": -381.8203629032258,
+      "logps/rejected": -183.4941287878788,
+      "loss": 0.5876,
+      "loss/base_kto": 0.4996063709259033,
+      "metrics/advantage_var": 0.060703542083501816,
+      "regularization/grad_norm_sq": 3963.318359375,
+      "regularization/w2": 0.08724621683359146,
+      "rewards/chosen": 0.012640624661599436,
+      "rewards/margins": 0.003818599709317831,
+      "rewards/rejected": 0.008822024952281605,
+      "stability/repetition_rate_mean": 0.18141163885593414,
+      "stability/response_length_mean": 20.625,
+      "stability/response_length_std": 27.176342010498047,
+      "stability/response_length_var": 738.5535888671875,
+      "stability/token_entropy_mean": 5.541015625,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5525412559509277,
+      "kl": 0.08330817520618439,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52759789.83225807,
+      "logits/rejected": -49030733.57575758,
+      "logps/chosen": -379.01844758064516,
+      "logps/rejected": -220.32575757575756,
+      "loss": 0.5896,
+      "loss/base_kto": 0.4998234808444977,
+      "metrics/advantage_var": 0.07535155117511749,
+      "regularization/grad_norm_sq": 4505.4248046875,
+      "regularization/w2": 0.09144462645053864,
+      "rewards/chosen": 0.01100011948616274,
+      "rewards/margins": 0.001358813158111722,
+      "rewards/rejected": 0.009641306328051017,
+      "stability/repetition_rate_mean": 0.1875,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.734375,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.6087732315063477,
+      "kl": 0.08941509574651718,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49510995.027027026,
+      "logits/rejected": -47899594.41860465,
+      "logps/chosen": -314.11993243243245,
+      "logps/rejected": -228.82019894622093,
+      "loss": 0.5871,
+      "loss/base_kto": 0.4989282786846161,
+      "metrics/advantage_var": 0.0716477707028389,
+      "regularization/grad_norm_sq": 3729.305419921875,
+      "regularization/w2": 0.08471358567476273,
+      "rewards/chosen": 0.01508830366907893,
+      "rewards/margins": 0.008622753590176797,
+      "rewards/rejected": 0.0064655500789021336,
+      "stability/repetition_rate_mean": 0.07421875,
+      "stability/response_length_mean": 24.25,
+      "stability/response_length_std": 32.92306900024414,
+      "stability/response_length_var": 1083.9285888671875,
+      "stability/token_entropy_mean": 1.901123046875,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.6029742360115051,
+      "kl": 0.12230757623910904,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58619103.78807947,
+      "logits/rejected": -53444892.78106509,
+      "logps/chosen": -403.8790873344371,
+      "logps/rejected": -200.96717825443787,
+      "loss": 0.586,
+      "loss/base_kto": 0.49925336241722107,
+      "metrics/advantage_var": 0.08379612118005753,
+      "regularization/grad_norm_sq": 3586.81494140625,
+      "regularization/w2": 0.08361315727233887,
+      "rewards/chosen": 0.015101601745908624,
+      "rewards/margins": 0.006237886015433373,
+      "rewards/rejected": 0.008863715730475251,
+      "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": 5.22265625,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.6115427613258362,
+      "kl": 0.12014111131429672,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51479532.18064516,
+      "logits/rejected": -47144202.86060606,
+      "logps/chosen": -358.09873991935484,
+      "logps/rejected": -230.32935606060607,
+      "loss": 0.5852,
+      "loss/base_kto": 0.4983179271221161,
+      "metrics/advantage_var": 0.0890185683965683,
+      "regularization/grad_norm_sq": 3845.35009765625,
+      "regularization/w2": 0.08649836480617523,
+      "rewards/chosen": 0.023378246061263545,
+      "rewards/margins": 0.013833501425190287,
+      "rewards/rejected": 0.009544744636073257,
+      "stability/repetition_rate_mean": 0.1291852593421936,
+      "stability/response_length_mean": 28.0,
+      "stability/response_length_std": 31.181495666503906,
+      "stability/response_length_var": 972.2857055664062,
+      "stability/token_entropy_mean": 4.955078125,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.6375041604042053,
+      "kl": 0.16934870183467865,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58404365.12820513,
+      "logits/rejected": -50561935.6097561,
+      "logps/chosen": -380.53598257211536,
+      "logps/rejected": -187.11123285060975,
+      "loss": 0.5845,
+      "loss/base_kto": 0.4982176721096039,
+      "metrics/advantage_var": 0.09395573288202286,
+      "regularization/grad_norm_sq": 3960.31103515625,
+      "regularization/w2": 0.0869247317314148,
+      "rewards/chosen": 0.025340945292741824,
+      "rewards/margins": 0.013783786392271555,
+      "rewards/rejected": 0.01155715890047027,
+      "stability/repetition_rate_mean": 0.3175307810306549,
+      "stability/response_length_mean": 35.625,
+      "stability/response_length_std": 31.057262420654297,
+      "stability/response_length_var": 964.5535888671875,
+      "stability/token_entropy_mean": 3.80078125,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.6220579147338867,
+      "kl": 0.16499601304531097,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50380346.32911392,
+      "logits/rejected": -48968169.87654321,
+      "logps/chosen": -314.7436956091772,
+      "logps/rejected": -207.06233121141975,
+      "loss": 0.5842,
+      "loss/base_kto": 0.49793797731399536,
+      "metrics/advantage_var": 0.13462769985198975,
+      "regularization/grad_norm_sq": 3773.901123046875,
+      "regularization/w2": 0.08464869111776352,
+      "rewards/chosen": 0.02496972416020647,
+      "rewards/margins": 0.016831111695584253,
+      "rewards/rejected": 0.008138612464622216,
+      "stability/repetition_rate_mean": 0.09133065491914749,
+      "stability/response_length_mean": 21.875,
+      "stability/response_length_std": 27.304826736450195,
+      "stability/response_length_var": 745.5535888671875,
+      "stability/token_entropy_mean": 4.09375,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.6105950474739075,
+      "kl": 0.20538730919361115,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54697513.51351351,
+      "logits/rejected": -50905552.37209302,
+      "logps/chosen": -359.57635663006755,
+      "logps/rejected": -216.37804324127907,
+      "loss": 0.5846,
+      "loss/base_kto": 0.4987795054912567,
+      "metrics/advantage_var": 0.12774451076984406,
+      "regularization/grad_norm_sq": 3816.7734375,
+      "regularization/w2": 0.08535917103290558,
+      "rewards/chosen": 0.029381449158127244,
+      "rewards/margins": 0.009296495572790869,
+      "rewards/rejected": 0.020084953585336376,
+      "stability/repetition_rate_mean": 0.2265625,
+      "stability/response_length_mean": 41.25,
+      "stability/response_length_std": 31.476749420166016,
+      "stability/response_length_var": 990.7857055664062,
+      "stability/token_entropy_mean": 4.166015625,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6735162138938904,
+      "kl": 0.22678242623806,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59375134.84337349,
+      "logits/rejected": -53807054.12987013,
+      "logps/chosen": -399.4981645331325,
+      "logps/rejected": -209.86383928571428,
+      "loss": 0.5847,
+      "loss/base_kto": 0.49670204520225525,
+      "metrics/advantage_var": 0.2160111963748932,
+      "regularization/grad_norm_sq": 3959.833740234375,
+      "regularization/w2": 0.08724039793014526,
+      "rewards/chosen": 0.04043001151946654,
+      "rewards/margins": 0.02643077232863418,
+      "rewards/rejected": 0.01399923919083236,
+      "stability/repetition_rate_mean": 0.1897321343421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.359375,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.6461487412452698,
+      "kl": 0.2732836902141571,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57406969.8313253,
+      "logits/rejected": -53620639.584415585,
+      "logps/chosen": -378.2289627259036,
+      "logps/rejected": -205.90453784496754,
+      "loss": 0.5829,
+      "loss/base_kto": 0.4979163706302643,
+      "metrics/advantage_var": 0.1939479112625122,
+      "regularization/grad_norm_sq": 3633.171875,
+      "regularization/w2": 0.08427540957927704,
+      "rewards/chosen": 0.033429114215345265,
+      "rewards/margins": 0.015068041179646833,
+      "rewards/rejected": 0.018361073035698432,
+      "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": 4.416015625,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.6481019854545593,
+      "kl": 0.31861141324043274,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60599995.94936709,
+      "logits/rejected": -53401410.37037037,
+      "logps/chosen": -364.0679390822785,
+      "logps/rejected": -200.95594618055554,
+      "loss": 0.5815,
+      "loss/base_kto": 0.4965081214904785,
+      "metrics/advantage_var": 0.2759661376476288,
+      "regularization/grad_norm_sq": 3353.943115234375,
+      "regularization/w2": 0.08032859116792679,
+      "rewards/chosen": 0.052460066879851906,
+      "rewards/margins": 0.0283992906756132,
+      "rewards/rejected": 0.024060776204238705,
+      "stability/repetition_rate_mean": 0.3053385615348816,
+      "stability/response_length_mean": 49.0,
+      "stability/response_length_std": 27.795169830322266,
+      "stability/response_length_var": 772.5714111328125,
+      "stability/token_entropy_mean": 4.779296875,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.6184019446372986,
+      "kl": 0.3490689694881439,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57745769.4117647,
+      "logits/rejected": -54368427.68862276,
+      "logps/chosen": -383.6613562091503,
+      "logps/rejected": -211.98376684131736,
+      "loss": 0.5837,
+      "loss/base_kto": 0.49555841088294983,
+      "metrics/advantage_var": 0.4176083505153656,
+      "regularization/grad_norm_sq": 4317.544921875,
+      "regularization/w2": 0.08935306966304779,
+      "rewards/chosen": 0.05835722630320032,
+      "rewards/margins": 0.03509791492412741,
+      "rewards/rejected": 0.02325931137907291,
+      "stability/repetition_rate_mean": 0.1742897778749466,
+      "stability/response_length_mean": 27.875,
+      "stability/response_length_std": 30.31943130493164,
+      "stability/response_length_var": 919.2678833007812,
+      "stability/token_entropy_mean": 4.634765625,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.6115716099739075,
+      "kl": 0.39925938844680786,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60688299.83561644,
+      "logits/rejected": -55668524.137931034,
+      "logps/chosen": -416.8091823630137,
+      "logps/rejected": -216.12354076867817,
+      "loss": 0.582,
+      "loss/base_kto": 0.49605321884155273,
+      "metrics/advantage_var": 0.6236904263496399,
+      "regularization/grad_norm_sq": 3812.632568359375,
+      "regularization/w2": 0.08550175279378891,
+      "rewards/chosen": 0.057012499195255646,
+      "rewards/margins": 0.032647929103745005,
+      "rewards/rejected": 0.02436457009151064,
+      "stability/repetition_rate_mean": 0.16215278208255768,
+      "stability/response_length_mean": 15.25,
+      "stability/response_length_std": 24.93276596069336,
+      "stability/response_length_var": 621.6428833007812,
+      "stability/token_entropy_mean": 3.64453125,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.593185305595398,
+      "kl": 0.4498286247253418,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57376213.05806451,
+      "logits/rejected": -53835838.06060606,
+      "logps/chosen": -391.9687247983871,
+      "logps/rejected": -183.66205018939394,
+      "loss": 0.5824,
+      "loss/base_kto": 0.4946207106113434,
+      "metrics/advantage_var": 0.780052661895752,
+      "regularization/grad_norm_sq": 3791.43017578125,
+      "regularization/w2": 0.08550234884023666,
+      "rewards/chosen": 0.06892158754410282,
+      "rewards/margins": 0.04323520790913005,
+      "rewards/rejected": 0.025686379634972775,
+      "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": 3.6328125,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5935289263725281,
+      "kl": 0.5017997026443481,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59854006.18404908,
+      "logits/rejected": -55003403.41401274,
+      "logps/chosen": -400.9800373849693,
+      "logps/rejected": -247.51960589171975,
+      "loss": 0.5789,
+      "loss/base_kto": 0.49386605620384216,
+      "metrics/advantage_var": 1.076611042022705,
+      "regularization/grad_norm_sq": 4160.88037109375,
+      "regularization/w2": 0.08902523666620255,
+      "rewards/chosen": 0.08361249467346565,
+      "rewards/margins": 0.048496560192178954,
+      "rewards/rejected": 0.0351159344812867,
+      "stability/repetition_rate_mean": 0.2270585298538208,
+      "stability/response_length_mean": 29.875,
+      "stability/response_length_std": 29.502723693847656,
+      "stability/response_length_var": 870.4107055664062,
+      "stability/token_entropy_mean": 3.8955078125,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.7262139916419983,
+      "kl": 0.48766636848449707,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57086696.72727273,
+      "logits/rejected": -50077061.78064516,
+      "logps/chosen": -366.3728456439394,
+      "logps/rejected": -189.92496219758064,
+      "loss": 0.5811,
+      "loss/base_kto": 0.4916563630104065,
+      "metrics/advantage_var": 1.9005340337753296,
+      "regularization/grad_norm_sq": 5850.53369140625,
+      "regularization/w2": 0.09640365839004517,
+      "rewards/chosen": 0.08169195001775569,
+      "rewards/margins": 0.07074021258312237,
+      "rewards/rejected": 0.010951737434633317,
+      "stability/repetition_rate_mean": 0.2662760317325592,
+      "stability/response_length_mean": 52.75,
+      "stability/response_length_std": 22.72663688659668,
+      "stability/response_length_var": 516.5,
+      "stability/token_entropy_mean": 5.19921875,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.6740468144416809,
+      "kl": 0.5717339515686035,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51144241.34939759,
+      "logits/rejected": -51662345.97402597,
+      "logps/chosen": -329.4772684487952,
+      "logps/rejected": -199.15458857548703,
+      "loss": 0.5778,
+      "loss/base_kto": 0.4942333400249481,
+      "metrics/advantage_var": 0.8122803568840027,
+      "regularization/grad_norm_sq": 3395.665771484375,
+      "regularization/w2": 0.08150183409452438,
+      "rewards/chosen": 0.07945383600441806,
+      "rewards/margins": 0.04612136428073769,
+      "rewards/rejected": 0.033332471723680375,
+      "stability/repetition_rate_mean": 0.19258596003055573,
+      "stability/response_length_mean": 22.625,
+      "stability/response_length_std": 26.359804153442383,
+      "stability/response_length_var": 694.8392944335938,
+      "stability/token_entropy_mean": 5.59765625,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.6410212516784668,
+      "kl": 0.6380648016929626,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55422674.823529415,
+      "logits/rejected": -50191220.053333335,
+      "logps/chosen": -396.8014705882353,
+      "logps/rejected": -207.36661458333333,
+      "loss": 0.5788,
+      "loss/base_kto": 0.48983898758888245,
+      "metrics/advantage_var": 1.331944465637207,
+      "regularization/grad_norm_sq": 4066.153076171875,
+      "regularization/w2": 0.08941753953695297,
+      "rewards/chosen": 0.11329341215245864,
+      "rewards/margins": 0.0773618948693369,
+      "rewards/rejected": 0.035931517283121744,
+      "stability/repetition_rate_mean": 0.2115885466337204,
+      "stability/response_length_mean": 35.25,
+      "stability/response_length_std": 31.01957893371582,
+      "stability/response_length_var": 962.2142944335938,
+      "stability/token_entropy_mean": 4.6640625,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.6439304947853088,
+      "kl": 0.6510013341903687,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54019288.847058825,
+      "logits/rejected": -49348276.906666666,
+      "logps/chosen": -360.07109375,
+      "logps/rejected": -190.99877604166667,
+      "loss": 0.5778,
+      "loss/base_kto": 0.4926762580871582,
+      "metrics/advantage_var": 2.249819755554199,
+      "regularization/grad_norm_sq": 5026.2802734375,
+      "regularization/w2": 0.09407851099967957,
+      "rewards/chosen": 0.09181964537676643,
+      "rewards/margins": 0.06113565108355354,
+      "rewards/rejected": 0.03068399429321289,
+      "stability/repetition_rate_mean": 0.3444148898124695,
+      "stability/response_length_mean": 61.875,
+      "stability/response_length_std": 6.010407447814941,
+      "stability/response_length_var": 36.125,
+      "stability/token_entropy_mean": 4.5625,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.615806519985199,
+      "kl": 0.7197233438491821,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55244385.52380952,
+      "logits/rejected": -48958032.84210526,
+      "logps/chosen": -415.5634765625,
+      "logps/rejected": -186.464111328125,
+      "loss": 0.5777,
+      "loss/base_kto": 0.4885575473308563,
+      "metrics/advantage_var": 1.9970428943634033,
+      "regularization/grad_norm_sq": 4003.491455078125,
+      "regularization/w2": 0.08765041083097458,
+      "rewards/chosen": 0.13318413779849098,
+      "rewards/margins": 0.08986123582175501,
+      "rewards/rejected": 0.04332290197673597,
+      "stability/repetition_rate_mean": 0.30851560831069946,
+      "stability/response_length_mean": 32.5,
+      "stability/response_length_std": 30.528675079345703,
+      "stability/response_length_var": 932.0,
+      "stability/token_entropy_mean": 4.44921875,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.7225887179374695,
+      "kl": 0.7622379660606384,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54423646.16091954,
+      "logits/rejected": -49801538.63013699,
+      "logps/chosen": -369.7487652658046,
+      "logps/rejected": -212.1081843964041,
+      "loss": 0.575,
+      "loss/base_kto": 0.4914153218269348,
+      "metrics/advantage_var": 1.7091789245605469,
+      "regularization/grad_norm_sq": 3531.582763671875,
+      "regularization/w2": 0.0829373374581337,
+      "rewards/chosen": 0.11679701969541352,
+      "rewards/margins": 0.06840403473033733,
+      "rewards/rejected": 0.048392984965076186,
+      "stability/repetition_rate_mean": 0.2411106526851654,
+      "stability/response_length_mean": 22.5,
+      "stability/response_length_std": 28.33977508544922,
+      "stability/response_length_var": 803.1428833007812,
+      "stability/token_entropy_mean": 4.8671875,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.6213456988334656,
+      "kl": 0.9274574518203735,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59135054.76923077,
+      "logits/rejected": -52582671.25827815,
+      "logps/chosen": -439.6777921597633,
+      "logps/rejected": -190.59510813327816,
+      "loss": 0.5717,
+      "loss/base_kto": 0.4875049293041229,
+      "metrics/advantage_var": 2.3025033473968506,
+      "regularization/grad_norm_sq": 4036.350830078125,
+      "regularization/w2": 0.08665412664413452,
+      "rewards/chosen": 0.1664069587662375,
+      "rewards/margins": 0.1001336926497226,
+      "rewards/rejected": 0.0662732661165149,
+      "stability/repetition_rate_mean": 0.22428728640079498,
+      "stability/response_length_mean": 36.25,
+      "stability/response_length_std": 30.76988410949707,
+      "stability/response_length_var": 946.7857055664062,
+      "stability/token_entropy_mean": 3.7744140625,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.7222792506217957,
+      "kl": 0.9340354204177856,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56650924.72289157,
+      "logits/rejected": -48021038.54545455,
+      "logps/chosen": -354.87575301204816,
+      "logps/rejected": -194.314453125,
+      "loss": 0.5718,
+      "loss/base_kto": 0.48661375045776367,
+      "metrics/advantage_var": 4.268935680389404,
+      "regularization/grad_norm_sq": 4183.99560546875,
+      "regularization/w2": 0.08717496693134308,
+      "rewards/chosen": 0.15614137304834572,
+      "rewards/margins": 0.10683375422597546,
+      "rewards/rejected": 0.04930761882237026,
+      "stability/repetition_rate_mean": 0.2011384218931198,
+      "stability/response_length_mean": 28.125,
+      "stability/response_length_std": 25.357091903686523,
+      "stability/response_length_var": 642.9821166992188,
+      "stability/token_entropy_mean": 4.91015625,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.6457726955413818,
+      "kl": 1.0286387205123901,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52652798.42944785,
+      "logits/rejected": -47983420.33121019,
+      "logps/chosen": -374.4070408358896,
+      "logps/rejected": -210.5137838375796,
+      "loss": 0.5699,
+      "loss/base_kto": 0.48794251680374146,
+      "metrics/advantage_var": 4.224320411682129,
+      "regularization/grad_norm_sq": 3658.589111328125,
+      "regularization/w2": 0.08313455432653427,
+      "rewards/chosen": 0.1738798925481691,
+      "rewards/margins": 0.0954359765675324,
+      "rewards/rejected": 0.0784439159806367,
+      "stability/repetition_rate_mean": 0.1989397406578064,
+      "stability/response_length_mean": 35.25,
+      "stability/response_length_std": 31.129911422729492,
+      "stability/response_length_var": 969.0714111328125,
+      "stability/token_entropy_mean": 3.41015625,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.6506476402282715,
+      "kl": 1.0774658918380737,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54617227.061728396,
+      "logits/rejected": -49519227.1392405,
+      "logps/chosen": -310.31630979938274,
+      "logps/rejected": -196.56126384493672,
+      "loss": 0.5768,
+      "loss/base_kto": 0.486668199300766,
+      "metrics/advantage_var": 7.448278903961182,
+      "regularization/grad_norm_sq": 6933.6611328125,
+      "regularization/w2": 0.09475024789571762,
+      "rewards/chosen": 0.16850583347273462,
+      "rewards/margins": 0.12071590972032412,
+      "rewards/rejected": 0.0477899237524105,
+      "stability/repetition_rate_mean": 0.15234375,
+      "stability/response_length_mean": 32.875,
+      "stability/response_length_std": 33.30996322631836,
+      "stability/response_length_var": 1109.5535888671875,
+      "stability/token_entropy_mean": 3.6953125,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.6974467635154724,
+      "kl": 1.1172609329223633,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53099575.18562874,
+      "logits/rejected": -49682532.39215686,
+      "logps/chosen": -390.274186002994,
+      "logps/rejected": -210.4579248366013,
+      "loss": 0.572,
+      "loss/base_kto": 0.4829126298427582,
+      "metrics/advantage_var": 6.995049953460693,
+      "regularization/grad_norm_sq": 5340.63427734375,
+      "regularization/w2": 0.09194418042898178,
+      "rewards/chosen": 0.21311228860638098,
+      "rewards/margins": 0.13904476766693424,
+      "rewards/rejected": 0.07406752093944674,
+      "stability/repetition_rate_mean": 0.13363486528396606,
+      "stability/response_length_mean": 16.75,
+      "stability/response_length_std": 27.12800407409668,
+      "stability/response_length_var": 735.9285888671875,
+      "stability/token_entropy_mean": 5.107421875,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.6366767287254333,
+      "kl": 1.1507939100265503,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60035851.561290324,
+      "logits/rejected": -53796069.624242425,
+      "logps/chosen": -390.8335685483871,
+      "logps/rejected": -197.07270359848485,
+      "loss": 0.5705,
+      "loss/base_kto": 0.48604080080986023,
+      "metrics/advantage_var": 4.371114253997803,
+      "regularization/grad_norm_sq": 3933.094482421875,
+      "regularization/w2": 0.08576133102178574,
+      "rewards/chosen": 0.19513650709582914,
+      "rewards/margins": 0.11047319690036868,
+      "rewards/rejected": 0.08466331019546046,
+      "stability/repetition_rate_mean": 0.3953683078289032,
+      "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_delta0.001/checkpoint-196/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc269180f5d75bd1ef95cbfdbfbd583fa9ef80416dbb4748e6a57fe1cde09ea8
+size 8081

mistral7b_kto_w2_delta0.001/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_delta0.001/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_delta0.001/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": 3.4365234375, "stability/repetition_rate_mean": 0.3551284670829773, "step": 10}
+{"stability/response_length_mean": 28.5, "stability/response_length_std": 30.10457992553711, "stability/response_length_var": 906.2857055664062, "stability/token_entropy_mean": 3.75390625, "stability/repetition_rate_mean": 0.2763020694255829, "step": 15}
+{"stability/response_length_mean": 49.75, "stability/response_length_std": 26.649845123291016, "stability/response_length_var": 710.2142944335938, "stability/token_entropy_mean": 4.154296875, "stability/repetition_rate_mean": 0.1925223171710968, "step": 20}
+{"stability/response_length_mean": 43.375, "stability/response_length_std": 28.54538917541504, "stability/response_length_var": 814.8392944335938, "stability/token_entropy_mean": 4.73828125, "stability/repetition_rate_mean": 0.18387754261493683, "step": 25}
+{"stability/response_length_mean": 48.375, "stability/response_length_std": 28.93311309814453, "stability/response_length_var": 837.125, "stability/token_entropy_mean": 4.52734375, "stability/repetition_rate_mean": 0.1875, "step": 30}
+{"stability/response_length_mean": 50.5, "stability/response_length_std": 25.048524856567383, "stability/response_length_var": 627.4285888671875, "stability/token_entropy_mean": 5.5234375, "stability/repetition_rate_mean": 0.2854352593421936, "step": 35}
+{"stability/response_length_mean": 35.0, "stability/response_length_std": 31.518701553344727, "stability/response_length_var": 993.4285888671875, "stability/token_entropy_mean": 4.8115234375, "stability/repetition_rate_mean": 0.16303454339504242, "step": 40}
+{"stability/response_length_mean": 29.25, "stability/response_length_std": 29.51875877380371, "stability/response_length_var": 871.3571166992188, "stability/token_entropy_mean": 4.705078125, "stability/repetition_rate_mean": 0.0915892943739891, "step": 45}
+{"stability/response_length_mean": 55.375, "stability/response_length_std": 16.361431121826172, "stability/response_length_var": 267.6964416503906, "stability/token_entropy_mean": 3.806640625, "stability/repetition_rate_mean": 0.4206075370311737, "step": 50}
+{"stability/response_length_mean": 22.375, "stability/response_length_std": 30.236862182617188, "stability/response_length_var": 914.2678833007812, "stability/token_entropy_mean": 4.3310546875, "stability/repetition_rate_mean": 0.10418051481246948, "step": 55}
+{"stability/response_length_mean": 51.125, "stability/response_length_std": 21.90523338317871, "stability/response_length_var": 479.83929443359375, "stability/token_entropy_mean": 5.384765625, "stability/repetition_rate_mean": 0.19273173809051514, "step": 60}
+{"stability/response_length_mean": 20.625, "stability/response_length_std": 27.176342010498047, "stability/response_length_var": 738.5535888671875, "stability/token_entropy_mean": 5.541015625, "stability/repetition_rate_mean": 0.18141163885593414, "step": 65}
+{"stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.734375, "stability/repetition_rate_mean": 0.1875, "step": 70}
+{"stability/response_length_mean": 24.25, "stability/response_length_std": 32.92306900024414, "stability/response_length_var": 1083.9285888671875, "stability/token_entropy_mean": 1.901123046875, "stability/repetition_rate_mean": 0.07421875, "step": 75}
+{"stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.22265625, "stability/repetition_rate_mean": 0.19088542461395264, "step": 80}
+{"stability/response_length_mean": 28.0, "stability/response_length_std": 31.181495666503906, "stability/response_length_var": 972.2857055664062, "stability/token_entropy_mean": 4.955078125, "stability/repetition_rate_mean": 0.1291852593421936, "step": 85}
+{"stability/response_length_mean": 35.625, "stability/response_length_std": 31.057262420654297, "stability/response_length_var": 964.5535888671875, "stability/token_entropy_mean": 3.80078125, "stability/repetition_rate_mean": 0.3175307810306549, "step": 90}
+{"stability/response_length_mean": 21.875, "stability/response_length_std": 27.304826736450195, "stability/response_length_var": 745.5535888671875, "stability/token_entropy_mean": 4.09375, "stability/repetition_rate_mean": 0.09133065491914749, "step": 95}
+{"stability/response_length_mean": 41.25, "stability/response_length_std": 31.476749420166016, "stability/response_length_var": 990.7857055664062, "stability/token_entropy_mean": 4.166015625, "stability/repetition_rate_mean": 0.2265625, "step": 100}
+{"stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.359375, "stability/repetition_rate_mean": 0.1897321343421936, "step": 105}
+{"stability/response_length_mean": 28.875, "stability/response_length_std": 29.911476135253906, "stability/response_length_var": 894.6964111328125, "stability/token_entropy_mean": 4.416015625, "stability/repetition_rate_mean": 0.30853456258773804, "step": 110}
+{"stability/response_length_mean": 49.0, "stability/response_length_std": 27.795169830322266, "stability/response_length_var": 772.5714111328125, "stability/token_entropy_mean": 4.779296875, "stability/repetition_rate_mean": 0.3053385615348816, "step": 115}
+{"stability/response_length_mean": 27.875, "stability/response_length_std": 30.31943130493164, "stability/response_length_var": 919.2678833007812, "stability/token_entropy_mean": 4.634765625, "stability/repetition_rate_mean": 0.1742897778749466, "step": 120}
+{"stability/response_length_mean": 15.25, "stability/response_length_std": 24.93276596069336, "stability/response_length_var": 621.6428833007812, "stability/token_entropy_mean": 3.64453125, "stability/repetition_rate_mean": 0.16215278208255768, "step": 125}
+{"stability/response_length_mean": 43.25, "stability/response_length_std": 29.620214462280273, "stability/response_length_var": 877.3571166992188, "stability/token_entropy_mean": 3.6328125, "stability/repetition_rate_mean": 0.24460937082767487, "step": 130}
+{"stability/response_length_mean": 29.875, "stability/response_length_std": 29.502723693847656, "stability/response_length_var": 870.4107055664062, "stability/token_entropy_mean": 3.8955078125, "stability/repetition_rate_mean": 0.2270585298538208, "step": 135}
+{"stability/response_length_mean": 52.75, "stability/response_length_std": 22.72663688659668, "stability/response_length_var": 516.5, "stability/token_entropy_mean": 5.19921875, "stability/repetition_rate_mean": 0.2662760317325592, "step": 140}
+{"stability/response_length_mean": 22.625, "stability/response_length_std": 26.359804153442383, "stability/response_length_var": 694.8392944335938, "stability/token_entropy_mean": 5.59765625, "stability/repetition_rate_mean": 0.19258596003055573, "step": 145}
+{"stability/response_length_mean": 35.25, "stability/response_length_std": 31.01957893371582, "stability/response_length_var": 962.2142944335938, "stability/token_entropy_mean": 4.6640625, "stability/repetition_rate_mean": 0.2115885466337204, "step": 150}
+{"stability/response_length_mean": 61.875, "stability/response_length_std": 6.010407447814941, "stability/response_length_var": 36.125, "stability/token_entropy_mean": 4.5625, "stability/repetition_rate_mean": 0.3444148898124695, "step": 155}
+{"stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.44921875, "stability/repetition_rate_mean": 0.30851560831069946, "step": 160}
+{"stability/response_length_mean": 22.5, "stability/response_length_std": 28.33977508544922, "stability/response_length_var": 803.1428833007812, "stability/token_entropy_mean": 4.8671875, "stability/repetition_rate_mean": 0.2411106526851654, "step": 165}
+{"stability/response_length_mean": 36.25, "stability/response_length_std": 30.76988410949707, "stability/response_length_var": 946.7857055664062, "stability/token_entropy_mean": 3.7744140625, "stability/repetition_rate_mean": 0.22428728640079498, "step": 170}
+{"stability/response_length_mean": 28.125, "stability/response_length_std": 25.357091903686523, "stability/response_length_var": 642.9821166992188, "stability/token_entropy_mean": 4.91015625, "stability/repetition_rate_mean": 0.2011384218931198, "step": 175}
+{"stability/response_length_mean": 35.25, "stability/response_length_std": 31.129911422729492, "stability/response_length_var": 969.0714111328125, "stability/token_entropy_mean": 3.41015625, "stability/repetition_rate_mean": 0.1989397406578064, "step": 180}
+{"stability/response_length_mean": 32.875, "stability/response_length_std": 33.30996322631836, "stability/response_length_var": 1109.5535888671875, "stability/token_entropy_mean": 3.6953125, "stability/repetition_rate_mean": 0.15234375, "step": 185}
+{"stability/response_length_mean": 16.75, "stability/response_length_std": 27.12800407409668, "stability/response_length_var": 735.9285888671875, "stability/token_entropy_mean": 5.107421875, "stability/repetition_rate_mean": 0.13363486528396606, "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": 0.3953683078289032, "step": 195}

mistral7b_kto_w2_delta0.001/tokenizer.model

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

mistral7b_kto_w2_delta0.001/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_delta0.001/train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.5821207132266493,
+    "train_runtime": 4296.589,
+    "train_samples_per_second": 23.274,
+    "train_steps_per_second": 0.046
+}

mistral7b_kto_w2_delta0.001/trainer_log.jsonl

@@ -0,0 +1,40 @@
+{"current_steps": 5, "total_steps": 196, "loss": 0.5885, "lr": 4e-08, "epoch": 0.025591810620601407, "percentage": 2.55, "elapsed_time": "0:01:47", "remaining_time": "1:08:33", "rewards/chosen": 0.004319707029744198, "logps/chosen": -439.0024157072368, "logits/chosen": -59713832.421052635, "rewards/rejected": -0.0003125370435771488, "logps/rejected": -198.41252790178572, "logits/rejected": -50416451.047619045, "rewards/margins": 0.004632244073321347, "regularization/w2": 0.09046166390180588, "regularization/grad_norm_sq": 4308.1396484375, "kl": 0.009683904238045216}
+{"current_steps": 10, "total_steps": 196, "loss": 0.586, "lr": 9e-08, "epoch": 0.05118362124120281, "percentage": 5.1, "elapsed_time": "0:03:37", "remaining_time": "1:07:31", "rewards/chosen": 0.003629611297087236, "logps/chosen": -429.9021915584416, "logits/chosen": -59306230.02597403, "rewards/rejected": 0.001039086246346853, "logps/rejected": -192.85753953313252, "logits/rejected": -50024096.38554217, "rewards/margins": 0.002590525050740383, "regularization/w2": 0.08616509288549423, "regularization/grad_norm_sq": 3819.715576171875, "kl": 0.0205225832760334, "stability/response_length_mean": 46.125, "stability/response_length_std": 27.231481552124023, "stability/response_length_var": 741.5535888671875, "stability/token_entropy_mean": 3.4365234375, "stability/repetition_rate_mean": 0.3551284670829773}
+{"current_steps": 15, "total_steps": 196, "loss": 0.5856, "lr": 1.4e-07, "epoch": 0.07677543186180422, "percentage": 7.65, "elapsed_time": "0:05:26", "remaining_time": "1:05:38", "rewards/chosen": 0.002835910970514471, "logps/chosen": -404.1024621212121, "logits/chosen": -56229478.4, "rewards/rejected": 0.0020273097099796416, "logps/rejected": -200.67379032258066, "logits/rejected": -51918392.15483871, "rewards/margins": 0.0008086012605348293, "regularization/w2": 0.08851373195648193, "regularization/grad_norm_sq": 4186.52001953125, "kl": 0.011671505868434906, "stability/response_length_mean": 28.5, "stability/response_length_std": 30.10457992553711, "stability/response_length_var": 906.2857055664062, "stability/token_entropy_mean": 3.75390625, "stability/repetition_rate_mean": 0.2763020694255829}
+{"current_steps": 20, "total_steps": 196, "loss": 0.5864, "lr": 1.8999999999999998e-07, "epoch": 0.10236724248240563, "percentage": 10.2, "elapsed_time": "0:07:17", "remaining_time": "1:04:09", "rewards/chosen": 0.0011490656994283198, "logps/chosen": -343.685205078125, "logits/chosen": -51667321.6, "rewards/rejected": 0.004474234208464623, "logps/rejected": -225.618603515625, "logits/rejected": -48058720.0, "rewards/margins": -0.003325168509036303, "regularization/w2": 0.08394529670476913, "regularization/grad_norm_sq": 3616.344970703125, "kl": 0.01878364384174347, "stability/response_length_mean": 49.75, "stability/response_length_std": 26.649845123291016, "stability/response_length_var": 710.2142944335938, "stability/token_entropy_mean": 4.154296875, "stability/repetition_rate_mean": 0.1925223171710968}
+{"current_steps": 25, "total_steps": 196, "loss": 0.5883, "lr": 2.4e-07, "epoch": 0.12795905310300704, "percentage": 12.76, "elapsed_time": "0:09:06", "remaining_time": "1:02:15", "rewards/chosen": 0.0012034501495032475, "logps/chosen": -378.47941810344827, "logits/chosen": -56484976.993103445, "rewards/rejected": 0.0023192856992994035, "logps/rejected": -202.76964285714286, "logits/rejected": -52025847.22285714, "rewards/margins": -0.001115835549796156, "regularization/w2": 0.08617646992206573, "regularization/grad_norm_sq": 4000.53515625, "kl": 0.018041405826807022, "stability/response_length_mean": 43.375, "stability/response_length_std": 28.54538917541504, "stability/response_length_var": 814.8392944335938, "stability/token_entropy_mean": 4.73828125, "stability/repetition_rate_mean": 0.18387754261493683}
+{"current_steps": 30, "total_steps": 196, "loss": 0.5872, "lr": 2.9e-07, "epoch": 0.15355086372360843, "percentage": 15.31, "elapsed_time": "0:10:55", "remaining_time": "1:00:26", "rewards/chosen": 0.001230826113848078, "logps/chosen": -468.56889681208054, "logits/chosen": -57733023.7852349, "rewards/rejected": 0.0019039563965379145, "logps/rejected": -198.22884685672514, "logits/rejected": -51524224.74853801, "rewards/margins": -0.0006731302826898365, "regularization/w2": 0.09177713841199875, "regularization/grad_norm_sq": 4545.20458984375, "kl": 0.01663057506084442, "stability/response_length_mean": 48.375, "stability/response_length_std": 28.93311309814453, "stability/response_length_var": 837.125, "stability/token_entropy_mean": 4.52734375, "stability/repetition_rate_mean": 0.1875}
+{"current_steps": 35, "total_steps": 196, "loss": 0.5878, "lr": 3.4000000000000003e-07, "epoch": 0.17914267434420986, "percentage": 17.86, "elapsed_time": "0:12:45", "remaining_time": "0:58:40", "rewards/chosen": 0.002080178713496727, "logps/chosen": -385.9157189477848, "logits/chosen": -57625172.25316456, "rewards/rejected": 0.0005865038177113475, "logps/rejected": -195.12061149691357, "logits/rejected": -53896609.18518519, "rewards/margins": 0.0014936748957853799, "regularization/w2": 0.08562777936458588, "regularization/grad_norm_sq": 3760.884521484375, "kl": 0.016775427386164665, "stability/response_length_mean": 50.5, "stability/response_length_std": 25.048524856567383, "stability/response_length_var": 627.4285888671875, "stability/token_entropy_mean": 5.5234375, "stability/repetition_rate_mean": 0.2854352593421936}
+{"current_steps": 40, "total_steps": 196, "loss": 0.5859, "lr": 3.8999999999999997e-07, "epoch": 0.20473448496481125, "percentage": 20.41, "elapsed_time": "0:14:35", "remaining_time": "0:56:54", "rewards/chosen": 0.002502601221203804, "logps/chosen": -421.87763671875, "logits/chosen": -56634124.8, "rewards/rejected": 0.003506476804614067, "logps/rejected": -219.2741943359375, "logits/rejected": -50225545.6, "rewards/margins": -0.0010038755834102628, "regularization/w2": 0.0906253382563591, "regularization/grad_norm_sq": 4253.48828125, "kl": 0.017838044092059135, "stability/response_length_mean": 35.0, "stability/response_length_std": 31.518701553344727, "stability/response_length_var": 993.4285888671875, "stability/token_entropy_mean": 4.8115234375, "stability/repetition_rate_mean": 0.16303454339504242}
+{"current_steps": 45, "total_steps": 196, "loss": 0.5871, "lr": 4.3999999999999997e-07, "epoch": 0.23032629558541268, "percentage": 22.96, "elapsed_time": "0:16:25", "remaining_time": "0:55:07", "rewards/chosen": 0.0036000581395706074, "logps/chosen": -352.29244442114094, "logits/chosen": -52689789.422818795, "rewards/rejected": 0.005875056947183888, "logps/rejected": -243.81542397660817, "logits/rejected": -50767135.438596494, "rewards/margins": -0.0022749988076132804, "regularization/w2": 0.08328904211521149, "regularization/grad_norm_sq": 3660.001953125, "kl": 0.044592004269361496, "stability/response_length_mean": 29.25, "stability/response_length_std": 29.51875877380371, "stability/response_length_var": 871.3571166992188, "stability/token_entropy_mean": 4.705078125, "stability/repetition_rate_mean": 0.0915892943739891}
+{"current_steps": 50, "total_steps": 196, "loss": 0.5843, "lr": 4.9e-07, "epoch": 0.2559181062060141, "percentage": 25.51, "elapsed_time": "0:18:14", "remaining_time": "0:53:17", "rewards/chosen": 0.005930147715062903, "logps/chosen": -375.1398385067114, "logits/chosen": -54035864.912751675, "rewards/rejected": 0.005393417258011668, "logps/rejected": -180.18667763157896, "logits/rejected": -48294564.67836257, "rewards/margins": 0.0005367304570512354, "regularization/w2": 0.08533728122711182, "regularization/grad_norm_sq": 3861.982421875, "kl": 0.04660215601325035, "stability/response_length_mean": 55.375, "stability/response_length_std": 16.361431121826172, "stability/response_length_var": 267.6964416503906, "stability/token_entropy_mean": 3.806640625, "stability/repetition_rate_mean": 0.4206075370311737}
+{"current_steps": 55, "total_steps": 196, "loss": 0.5859, "lr": 5.4e-07, "epoch": 0.28150991682661547, "percentage": 28.06, "elapsed_time": "0:20:03", "remaining_time": "0:51:26", "rewards/chosen": 0.0017190375683470542, "logps/chosen": -344.4948078416149, "logits/chosen": -56956520.94409938, "rewards/rejected": 0.0042611041908743996, "logps/rejected": -230.40541470125785, "logits/rejected": -49569669.635220125, "rewards/margins": -0.0025420666225273456, "regularization/w2": 0.08618461340665817, "regularization/grad_norm_sq": 3821.798583984375, "kl": 0.04949667677283287, "stability/response_length_mean": 22.375, "stability/response_length_std": 30.236862182617188, "stability/response_length_var": 914.2678833007812, "stability/token_entropy_mean": 4.3310546875, "stability/repetition_rate_mean": 0.10418051481246948}
+{"current_steps": 60, "total_steps": 196, "loss": 0.5865, "lr": 5.9e-07, "epoch": 0.30710172744721687, "percentage": 30.61, "elapsed_time": "0:21:54", "remaining_time": "0:49:40", "rewards/chosen": 0.009310866124702222, "logps/chosen": -359.62769886363634, "logits/chosen": -52664307.587878786, "rewards/rejected": 0.005658834980380151, "logps/rejected": -178.1703377016129, "logits/rejected": -47816974.86451613, "rewards/margins": 0.003652031144322071, "regularization/w2": 0.07853565365076065, "regularization/grad_norm_sq": 3147.68408203125, "kl": 0.052175115793943405, "stability/response_length_mean": 51.125, "stability/response_length_std": 21.90523338317871, "stability/response_length_var": 479.83929443359375, "stability/token_entropy_mean": 5.384765625, "stability/repetition_rate_mean": 0.19273173809051514}
+{"current_steps": 65, "total_steps": 196, "loss": 0.5876, "lr": 6.4e-07, "epoch": 0.3326935380678183, "percentage": 33.16, "elapsed_time": "0:23:45", "remaining_time": "0:47:52", "rewards/chosen": 0.012640624661599436, "logps/chosen": -381.8203629032258, "logits/chosen": -57557335.535483874, "rewards/rejected": 0.008822024952281605, "logps/rejected": -183.4941287878788, "logits/rejected": -47273723.345454544, "rewards/margins": 0.003818599709317831, "regularization/w2": 0.08724621683359146, "regularization/grad_norm_sq": 3963.318359375, "kl": 0.08242836594581604, "stability/response_length_mean": 20.625, "stability/response_length_std": 27.176342010498047, "stability/response_length_var": 738.5535888671875, "stability/token_entropy_mean": 5.541015625, "stability/repetition_rate_mean": 0.18141163885593414}
+{"current_steps": 70, "total_steps": 196, "loss": 0.5896, "lr": 6.9e-07, "epoch": 0.3582853486884197, "percentage": 35.71, "elapsed_time": "0:25:34", "remaining_time": "0:46:01", "rewards/chosen": 0.01100011948616274, "logps/chosen": -379.01844758064516, "logits/chosen": -52759789.83225807, "rewards/rejected": 0.009641306328051017, "logps/rejected": -220.32575757575756, "logits/rejected": -49030733.57575758, "rewards/margins": 0.001358813158111722, "regularization/w2": 0.09144462645053864, "regularization/grad_norm_sq": 4505.4248046875, "kl": 0.08330817520618439, "stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.734375, "stability/repetition_rate_mean": 0.1875}
+{"current_steps": 75, "total_steps": 196, "loss": 0.5871, "lr": 7.4e-07, "epoch": 0.3838771593090211, "percentage": 38.27, "elapsed_time": "0:27:24", "remaining_time": "0:44:13", "rewards/chosen": 0.01508830366907893, "logps/chosen": -314.11993243243245, "logits/chosen": -49510995.027027026, "rewards/rejected": 0.0064655500789021336, "logps/rejected": -228.82019894622093, "logits/rejected": -47899594.41860465, "rewards/margins": 0.008622753590176797, "regularization/w2": 0.08471358567476273, "regularization/grad_norm_sq": 3729.305419921875, "kl": 0.08941509574651718, "stability/response_length_mean": 24.25, "stability/response_length_std": 32.92306900024414, "stability/response_length_var": 1083.9285888671875, "stability/token_entropy_mean": 1.901123046875, "stability/repetition_rate_mean": 0.07421875}
+{"current_steps": 80, "total_steps": 196, "loss": 0.586, "lr": 7.9e-07, "epoch": 0.4094689699296225, "percentage": 40.82, "elapsed_time": "0:29:15", "remaining_time": "0:42:25", "rewards/chosen": 0.015101601745908624, "logps/chosen": -403.8790873344371, "logits/chosen": -58619103.78807947, "rewards/rejected": 0.008863715730475251, "logps/rejected": -200.96717825443787, "logits/rejected": -53444892.78106509, "rewards/margins": 0.006237886015433373, "regularization/w2": 0.08361315727233887, "regularization/grad_norm_sq": 3586.81494140625, "kl": 0.12230757623910904, "stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.22265625, "stability/repetition_rate_mean": 0.19088542461395264}
+{"current_steps": 85, "total_steps": 196, "loss": 0.5852, "lr": 8.399999999999999e-07, "epoch": 0.4350607805502239, "percentage": 43.37, "elapsed_time": "0:31:05", "remaining_time": "0:40:35", "rewards/chosen": 0.023378246061263545, "logps/chosen": -358.09873991935484, "logits/chosen": -51479532.18064516, "rewards/rejected": 0.009544744636073257, "logps/rejected": -230.32935606060607, "logits/rejected": -47144202.86060606, "rewards/margins": 0.013833501425190287, "regularization/w2": 0.08649836480617523, "regularization/grad_norm_sq": 3845.35009765625, "kl": 0.12014111131429672, "stability/response_length_mean": 28.0, "stability/response_length_std": 31.181495666503906, "stability/response_length_var": 972.2857055664062, "stability/token_entropy_mean": 4.955078125, "stability/repetition_rate_mean": 0.1291852593421936}
+{"current_steps": 90, "total_steps": 196, "loss": 0.5845, "lr": 8.9e-07, "epoch": 0.46065259117082535, "percentage": 45.92, "elapsed_time": "0:32:56", "remaining_time": "0:38:47", "rewards/chosen": 0.025340945292741824, "logps/chosen": -380.53598257211536, "logits/chosen": -58404365.12820513, "rewards/rejected": 0.01155715890047027, "logps/rejected": -187.11123285060975, "logits/rejected": -50561935.6097561, "rewards/margins": 0.013783786392271555, "regularization/w2": 0.0869247317314148, "regularization/grad_norm_sq": 3960.31103515625, "kl": 0.16934870183467865, "stability/response_length_mean": 35.625, "stability/response_length_std": 31.057262420654297, "stability/response_length_var": 964.5535888671875, "stability/token_entropy_mean": 3.80078125, "stability/repetition_rate_mean": 0.3175307810306549}
+{"current_steps": 95, "total_steps": 196, "loss": 0.5842, "lr": 9.399999999999999e-07, "epoch": 0.48624440179142675, "percentage": 48.47, "elapsed_time": "0:34:45", "remaining_time": "0:36:57", "rewards/chosen": 0.02496972416020647, "logps/chosen": -314.7436956091772, "logits/chosen": -50380346.32911392, "rewards/rejected": 0.008138612464622216, "logps/rejected": -207.06233121141975, "logits/rejected": -48968169.87654321, "rewards/margins": 0.016831111695584253, "regularization/w2": 0.08464869111776352, "regularization/grad_norm_sq": 3773.901123046875, "kl": 0.16499601304531097, "stability/response_length_mean": 21.875, "stability/response_length_std": 27.304826736450195, "stability/response_length_var": 745.5535888671875, "stability/token_entropy_mean": 4.09375, "stability/repetition_rate_mean": 0.09133065491914749}
+{"current_steps": 100, "total_steps": 196, "loss": 0.5846, "lr": 9.9e-07, "epoch": 0.5118362124120281, "percentage": 51.02, "elapsed_time": "0:36:35", "remaining_time": "0:35:07", "rewards/chosen": 0.029381449158127244, "logps/chosen": -359.57635663006755, "logits/chosen": -54697513.51351351, "rewards/rejected": 0.020084953585336376, "logps/rejected": -216.37804324127907, "logits/rejected": -50905552.37209302, "rewards/margins": 0.009296495572790869, "regularization/w2": 0.08535917103290558, "regularization/grad_norm_sq": 3816.7734375, "kl": 0.20538730919361115, "stability/response_length_mean": 41.25, "stability/response_length_std": 31.476749420166016, "stability/response_length_var": 990.7857055664062, "stability/token_entropy_mean": 4.166015625, "stability/repetition_rate_mean": 0.2265625}
+{"current_steps": 105, "total_steps": 196, "loss": 0.5847, "lr": 9.999815554351224e-07, "epoch": 0.5374280230326296, "percentage": 53.57, "elapsed_time": "0:38:25", "remaining_time": "0:33:18", "rewards/chosen": 0.04043001151946654, "logps/chosen": -399.4981645331325, "logits/chosen": -59375134.84337349, "rewards/rejected": 0.01399923919083236, "logps/rejected": -209.86383928571428, "logits/rejected": -53807054.12987013, "rewards/margins": 0.02643077232863418, "regularization/w2": 0.08724039793014526, "regularization/grad_norm_sq": 3959.833740234375, "kl": 0.22678242623806, "stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.359375, "stability/repetition_rate_mean": 0.1897321343421936}
+{"current_steps": 110, "total_steps": 196, "loss": 0.5829, "lr": 9.999066267225444e-07, "epoch": 0.5630198336532309, "percentage": 56.12, "elapsed_time": "0:40:16", "remaining_time": "0:31:29", "rewards/chosen": 0.033429114215345265, "logps/chosen": -378.2289627259036, "logits/chosen": -57406969.8313253, "rewards/rejected": 0.018361073035698432, "logps/rejected": -205.90453784496754, "logits/rejected": -53620639.584415585, "rewards/margins": 0.015068041179646833, "regularization/w2": 0.08427540957927704, "regularization/grad_norm_sq": 3633.171875, "kl": 0.2732836902141571, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.911476135253906, "stability/response_length_var": 894.6964111328125, "stability/token_entropy_mean": 4.416015625, "stability/repetition_rate_mean": 0.30853456258773804}
+{"current_steps": 115, "total_steps": 196, "loss": 0.5815, "lr": 9.997740697079592e-07, "epoch": 0.5886116442738324, "percentage": 58.67, "elapsed_time": "0:42:07", "remaining_time": "0:29:40", "rewards/chosen": 0.052460066879851906, "logps/chosen": -364.0679390822785, "logits/chosen": -60599995.94936709, "rewards/rejected": 0.024060776204238705, "logps/rejected": -200.95594618055554, "logits/rejected": -53401410.37037037, "rewards/margins": 0.0283992906756132, "regularization/w2": 0.08032859116792679, "regularization/grad_norm_sq": 3353.943115234375, "kl": 0.31861141324043274, "stability/response_length_mean": 49.0, "stability/response_length_std": 27.795169830322266, "stability/response_length_var": 772.5714111328125, "stability/token_entropy_mean": 4.779296875, "stability/repetition_rate_mean": 0.3053385615348816}
+{"current_steps": 120, "total_steps": 196, "loss": 0.5837, "lr": 9.995838996722914e-07, "epoch": 0.6142034548944337, "percentage": 61.22, "elapsed_time": "0:43:56", "remaining_time": "0:27:49", "rewards/chosen": 0.05835722630320032, "logps/chosen": -383.6613562091503, "logits/chosen": -57745769.4117647, "rewards/rejected": 0.02325931137907291, "logps/rejected": -211.98376684131736, "logits/rejected": -54368427.68862276, "rewards/margins": 0.03509791492412741, "regularization/w2": 0.08935306966304779, "regularization/grad_norm_sq": 4317.544921875, "kl": 0.3490689694881439, "stability/response_length_mean": 27.875, "stability/response_length_std": 30.31943130493164, "stability/response_length_var": 919.2678833007812, "stability/token_entropy_mean": 4.634765625, "stability/repetition_rate_mean": 0.1742897778749466}
+{"current_steps": 125, "total_steps": 196, "loss": 0.582, "lr": 9.993361385379874e-07, "epoch": 0.6397952655150352, "percentage": 63.78, "elapsed_time": "0:45:47", "remaining_time": "0:26:00", "rewards/chosen": 0.057012499195255646, "logps/chosen": -416.8091823630137, "logits/chosen": -60688299.83561644, "rewards/rejected": 0.02436457009151064, "logps/rejected": -216.12354076867817, "logits/rejected": -55668524.137931034, "rewards/margins": 0.032647929103745005, "regularization/w2": 0.08550175279378891, "regularization/grad_norm_sq": 3812.632568359375, "kl": 0.39925938844680786, "stability/response_length_mean": 15.25, "stability/response_length_std": 24.93276596069336, "stability/response_length_var": 621.6428833007812, "stability/token_entropy_mean": 3.64453125, "stability/repetition_rate_mean": 0.16215278208255768}
+{"current_steps": 130, "total_steps": 196, "loss": 0.5824, "lr": 9.99030814866488e-07, "epoch": 0.6653870761356366, "percentage": 66.33, "elapsed_time": "0:47:36", "remaining_time": "0:24:10", "rewards/chosen": 0.06892158754410282, "logps/chosen": -391.9687247983871, "logits/chosen": -57376213.05806451, "rewards/rejected": 0.025686379634972775, "logps/rejected": -183.66205018939394, "logits/rejected": -53835838.06060606, "rewards/margins": 0.04323520790913005, "regularization/w2": 0.08550234884023666, "regularization/grad_norm_sq": 3791.43017578125, "kl": 0.4498286247253418, "stability/response_length_mean": 43.25, "stability/response_length_std": 29.620214462280273, "stability/response_length_var": 877.3571166992188, "stability/token_entropy_mean": 3.6328125, "stability/repetition_rate_mean": 0.24460937082767487}
+{"current_steps": 135, "total_steps": 196, "loss": 0.5789, "lr": 9.986679638549358e-07, "epoch": 0.690978886756238, "percentage": 68.88, "elapsed_time": "0:49:26", "remaining_time": "0:22:20", "rewards/chosen": 0.08361249467346565, "logps/chosen": -400.9800373849693, "logits/chosen": -59854006.18404908, "rewards/rejected": 0.0351159344812867, "logps/rejected": -247.51960589171975, "logits/rejected": -55003403.41401274, "rewards/margins": 0.048496560192178954, "regularization/w2": 0.08902523666620255, "regularization/grad_norm_sq": 4160.88037109375, "kl": 0.5017997026443481, "stability/response_length_mean": 29.875, "stability/response_length_std": 29.502723693847656, "stability/response_length_var": 870.4107055664062, "stability/token_entropy_mean": 3.8955078125, "stability/repetition_rate_mean": 0.2270585298538208}
+{"current_steps": 140, "total_steps": 196, "loss": 0.5811, "lr": 9.982476273321175e-07, "epoch": 0.7165706973768394, "percentage": 71.43, "elapsed_time": "0:51:15", "remaining_time": "0:20:30", "rewards/chosen": 0.08169195001775569, "logps/chosen": -366.3728456439394, "logits/chosen": -57086696.72727273, "rewards/rejected": 0.010951737434633317, "logps/rejected": -189.92496219758064, "logits/rejected": -50077061.78064516, "rewards/margins": 0.07074021258312237, "regularization/w2": 0.09640365839004517, "regularization/grad_norm_sq": 5850.53369140625, "kl": 0.48766636848449707, "stability/response_length_mean": 52.75, "stability/response_length_std": 22.72663688659668, "stability/response_length_var": 516.5, "stability/token_entropy_mean": 5.19921875, "stability/repetition_rate_mean": 0.2662760317325592}
+{"current_steps": 145, "total_steps": 196, "loss": 0.5778, "lr": 9.977698537536417e-07, "epoch": 0.7421625079974408, "percentage": 73.98, "elapsed_time": "0:53:04", "remaining_time": "0:18:40", "rewards/chosen": 0.07945383600441806, "logps/chosen": -329.4772684487952, "logits/chosen": -51144241.34939759, "rewards/rejected": 0.033332471723680375, "logps/rejected": -199.15458857548703, "logits/rejected": -51662345.97402597, "rewards/margins": 0.04612136428073769, "regularization/w2": 0.08150183409452438, "regularization/grad_norm_sq": 3395.665771484375, "kl": 0.5717339515686035, "stability/response_length_mean": 22.625, "stability/response_length_std": 26.359804153442383, "stability/response_length_var": 694.8392944335938, "stability/token_entropy_mean": 5.59765625, "stability/repetition_rate_mean": 0.19258596003055573}
+{"current_steps": 150, "total_steps": 196, "loss": 0.5788, "lr": 9.972346981963546e-07, "epoch": 0.7677543186180422, "percentage": 76.53, "elapsed_time": "0:54:54", "remaining_time": "0:16:50", "rewards/chosen": 0.11329341215245864, "logps/chosen": -396.8014705882353, "logits/chosen": -55422674.823529415, "rewards/rejected": 0.035931517283121744, "logps/rejected": -207.36661458333333, "logits/rejected": -50191220.053333335, "rewards/margins": 0.0773618948693369, "regularization/w2": 0.08941753953695297, "regularization/grad_norm_sq": 4066.153076171875, "kl": 0.6380648016929626, "stability/response_length_mean": 35.25, "stability/response_length_std": 31.01957893371582, "stability/response_length_var": 962.2142944335938, "stability/token_entropy_mean": 4.6640625, "stability/repetition_rate_mean": 0.2115885466337204}
+{"current_steps": 155, "total_steps": 196, "loss": 0.5778, "lr": 9.966422223519885e-07, "epoch": 0.7933461292386437, "percentage": 79.08, "elapsed_time": "0:56:45", "remaining_time": "0:15:00", "rewards/chosen": 0.09181964537676643, "logps/chosen": -360.07109375, "logits/chosen": -54019288.847058825, "rewards/rejected": 0.03068399429321289, "logps/rejected": -190.99877604166667, "logits/rejected": -49348276.906666666, "rewards/margins": 0.06113565108355354, "regularization/w2": 0.09407851099967957, "regularization/grad_norm_sq": 5026.2802734375, "kl": 0.6510013341903687, "stability/response_length_mean": 61.875, "stability/response_length_std": 6.010407447814941, "stability/response_length_var": 36.125, "stability/token_entropy_mean": 4.5625, "stability/repetition_rate_mean": 0.3444148898124695}
+{"current_steps": 160, "total_steps": 196, "loss": 0.5777, "lr": 9.959924945200523e-07, "epoch": 0.818937939859245, "percentage": 81.63, "elapsed_time": "0:58:35", "remaining_time": "0:13:10", "rewards/chosen": 0.13318413779849098, "logps/chosen": -415.5634765625, "logits/chosen": -55244385.52380952, "rewards/rejected": 0.04332290197673597, "logps/rejected": -186.464111328125, "logits/rejected": -48958032.84210526, "rewards/margins": 0.08986123582175501, "regularization/w2": 0.08765041083097458, "regularization/grad_norm_sq": 4003.491455078125, "kl": 0.7197233438491821, "stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.44921875, "stability/repetition_rate_mean": 0.30851560831069946}
+{"current_steps": 165, "total_steps": 196, "loss": 0.575, "lr": 9.952855895999568e-07, "epoch": 0.8445297504798465, "percentage": 84.18, "elapsed_time": "1:00:24", "remaining_time": "0:11:21", "rewards/chosen": 0.11679701969541352, "logps/chosen": -369.7487652658046, "logits/chosen": -54423646.16091954, "rewards/rejected": 0.048392984965076186, "logps/rejected": -212.1081843964041, "logits/rejected": -49801538.63013699, "rewards/margins": 0.06840403473033733, "regularization/w2": 0.0829373374581337, "regularization/grad_norm_sq": 3531.582763671875, "kl": 0.7622379660606384, "stability/response_length_mean": 22.5, "stability/response_length_std": 28.33977508544922, "stability/response_length_var": 803.1428833007812, "stability/token_entropy_mean": 4.8671875, "stability/repetition_rate_mean": 0.2411106526851654}
+{"current_steps": 170, "total_steps": 196, "loss": 0.5717, "lr": 9.9452158908238e-07, "epoch": 0.8701215611004478, "percentage": 86.73, "elapsed_time": "1:02:15", "remaining_time": "0:09:31", "rewards/chosen": 0.1664069587662375, "logps/chosen": -439.6777921597633, "logits/chosen": -59135054.76923077, "rewards/rejected": 0.0662732661165149, "logps/rejected": -190.59510813327816, "logits/rejected": -52582671.25827815, "rewards/margins": 0.1001336926497226, "regularization/w2": 0.08665412664413452, "regularization/grad_norm_sq": 4036.350830078125, "kl": 0.9274574518203735, "stability/response_length_mean": 36.25, "stability/response_length_std": 30.76988410949707, "stability/response_length_var": 946.7857055664062, "stability/token_entropy_mean": 3.7744140625, "stability/repetition_rate_mean": 0.22428728640079498}
+{"current_steps": 175, "total_steps": 196, "loss": 0.5718, "lr": 9.937005810398745e-07, "epoch": 0.8957133717210493, "percentage": 89.29, "elapsed_time": "1:04:05", "remaining_time": "0:07:41", "rewards/chosen": 0.15614137304834572, "logps/chosen": -354.87575301204816, "logits/chosen": -56650924.72289157, "rewards/rejected": 0.04930761882237026, "logps/rejected": -194.314453125, "logits/rejected": -48021038.54545455, "rewards/margins": 0.10683375422597546, "regularization/w2": 0.08717496693134308, "regularization/grad_norm_sq": 4183.99560546875, "kl": 0.9340354204177856, "stability/response_length_mean": 28.125, "stability/response_length_std": 25.357091903686523, "stability/response_length_var": 642.9821166992188, "stability/token_entropy_mean": 4.91015625, "stability/repetition_rate_mean": 0.2011384218931198}
+{"current_steps": 180, "total_steps": 196, "loss": 0.5699, "lr": 9.928226601167138e-07, "epoch": 0.9213051823416507, "percentage": 91.84, "elapsed_time": "1:05:55", "remaining_time": "0:05:51", "rewards/chosen": 0.1738798925481691, "logps/chosen": -374.4070408358896, "logits/chosen": -52652798.42944785, "rewards/rejected": 0.0784439159806367, "logps/rejected": -210.5137838375796, "logits/rejected": -47983420.33121019, "rewards/margins": 0.0954359765675324, "regularization/w2": 0.08313455432653427, "regularization/grad_norm_sq": 3658.589111328125, "kl": 1.0286387205123901, "stability/response_length_mean": 35.25, "stability/response_length_std": 31.129911422729492, "stability/response_length_var": 969.0714111328125, "stability/token_entropy_mean": 3.41015625, "stability/repetition_rate_mean": 0.1989397406578064}
+{"current_steps": 185, "total_steps": 196, "loss": 0.5768, "lr": 9.918879275179817e-07, "epoch": 0.946896992962252, "percentage": 94.39, "elapsed_time": "1:07:43", "remaining_time": "0:04:01", "rewards/chosen": 0.16850583347273462, "logps/chosen": -310.31630979938274, "logits/chosen": -54617227.061728396, "rewards/rejected": 0.0477899237524105, "logps/rejected": -196.56126384493672, "logits/rejected": -49519227.1392405, "rewards/margins": 0.12071590972032412, "regularization/w2": 0.09475024789571762, "regularization/grad_norm_sq": 6933.6611328125, "kl": 1.0774658918380737, "stability/response_length_mean": 32.875, "stability/response_length_std": 33.30996322631836, "stability/response_length_var": 1109.5535888671875, "stability/token_entropy_mean": 3.6953125, "stability/repetition_rate_mean": 0.15234375}
+{"current_steps": 190, "total_steps": 196, "loss": 0.572, "lr": 9.90896490997906e-07, "epoch": 0.9724888035828535, "percentage": 96.94, "elapsed_time": "1:09:32", "remaining_time": "0:02:11", "rewards/chosen": 0.21311228860638098, "logps/chosen": -390.274186002994, "logits/chosen": -53099575.18562874, "rewards/rejected": 0.07406752093944674, "logps/rejected": -210.4579248366013, "logits/rejected": -49682532.39215686, "rewards/margins": 0.13904476766693424, "regularization/w2": 0.09194418042898178, "regularization/grad_norm_sq": 5340.63427734375, "kl": 1.1172609329223633, "stability/response_length_mean": 16.75, "stability/response_length_std": 27.12800407409668, "stability/response_length_var": 735.9285888671875, "stability/token_entropy_mean": 5.107421875, "stability/repetition_rate_mean": 0.13363486528396606}
+{"current_steps": 195, "total_steps": 196, "loss": 0.5705, "lr": 9.898484648474362e-07, "epoch": 0.9980806142034548, "percentage": 99.49, "elapsed_time": "1:11:21", "remaining_time": "0:00:21", "rewards/chosen": 0.19513650709582914, "logps/chosen": -390.8335685483871, "logits/chosen": -60035851.561290324, "rewards/rejected": 0.08466331019546046, "logps/rejected": -197.07270359848485, "logits/rejected": -53796069.624242425, "rewards/margins": 0.11047319690036868, "regularization/w2": 0.08576133102178574, "regularization/grad_norm_sq": 3933.094482421875, "kl": 1.1507939100265503, "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": 0.3953683078289032}
+{"current_steps": 196, "total_steps": 196, "epoch": 1.0, "percentage": 100.0, "elapsed_time": "1:11:36", "remaining_time": "0:00:00"}

mistral7b_kto_w2_delta0.001/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": 0.5704261660575867,
+      "kl": 0.009683904238045216,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59713832.421052635,
+      "logits/rejected": -50416451.047619045,
+      "logps/chosen": -439.0024157072368,
+      "logps/rejected": -198.41252790178572,
+      "loss": 0.5885,
+      "loss/base_kto": 0.4994261860847473,
+      "metrics/advantage_var": 0.04821223020553589,
+      "regularization/grad_norm_sq": 4308.1396484375,
+      "regularization/w2": 0.09046166390180588,
+      "rewards/chosen": 0.004319707029744198,
+      "rewards/margins": 0.004632244073321347,
+      "rewards/rejected": -0.0003125370435771488,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.6019414663314819,
+      "kl": 0.0205225832760334,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59306230.02597403,
+      "logits/rejected": -50024096.38554217,
+      "logps/chosen": -429.9021915584416,
+      "logps/rejected": -192.85753953313252,
+      "loss": 0.586,
+      "loss/base_kto": 0.4996527135372162,
+      "metrics/advantage_var": 0.06569447368383408,
+      "regularization/grad_norm_sq": 3819.715576171875,
+      "regularization/w2": 0.08616509288549423,
+      "rewards/chosen": 0.003629611297087236,
+      "rewards/margins": 0.002590525050740383,
+      "rewards/rejected": 0.001039086246346853,
+      "stability/repetition_rate_mean": 0.3551284670829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.4365234375,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.6072039604187012,
+      "kl": 0.011671505868434906,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56229478.4,
+      "logits/rejected": -51918392.15483871,
+      "logps/chosen": -404.1024621212121,
+      "logps/rejected": -200.67379032258066,
+      "loss": 0.5856,
+      "loss/base_kto": 0.49987754225730896,
+      "metrics/advantage_var": 0.05594036728143692,
+      "regularization/grad_norm_sq": 4186.52001953125,
+      "regularization/w2": 0.08851373195648193,
+      "rewards/chosen": 0.002835910970514471,
+      "rewards/margins": 0.0008086012605348293,
+      "rewards/rejected": 0.0020273097099796416,
+      "stability/repetition_rate_mean": 0.2763020694255829,
+      "stability/response_length_mean": 28.5,
+      "stability/response_length_std": 30.10457992553711,
+      "stability/response_length_var": 906.2857055664062,
+      "stability/token_entropy_mean": 3.75390625,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.6190371513366699,
+      "kl": 0.01878364384174347,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51667321.6,
+      "logits/rejected": -48058720.0,
+      "logps/chosen": -343.685205078125,
+      "logps/rejected": -225.618603515625,
+      "loss": 0.5864,
+      "loss/base_kto": 0.5004865527153015,
+      "metrics/advantage_var": 0.055938970297575,
+      "regularization/grad_norm_sq": 3616.344970703125,
+      "regularization/w2": 0.08394529670476913,
+      "rewards/chosen": 0.0011490656994283198,
+      "rewards/margins": -0.003325168509036303,
+      "rewards/rejected": 0.004474234208464623,
+      "stability/repetition_rate_mean": 0.1925223171710968,
+      "stability/response_length_mean": 49.75,
+      "stability/response_length_std": 26.649845123291016,
+      "stability/response_length_var": 710.2142944335938,
+      "stability/token_entropy_mean": 4.154296875,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.5828595161437988,
+      "kl": 0.018041405826807022,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56484976.993103445,
+      "logits/rejected": -52025847.22285714,
+      "logps/chosen": -378.47941810344827,
+      "logps/rejected": -202.76964285714286,
+      "loss": 0.5883,
+      "loss/base_kto": 0.5001233220100403,
+      "metrics/advantage_var": 0.07194186747074127,
+      "regularization/grad_norm_sq": 4000.53515625,
+      "regularization/w2": 0.08617646992206573,
+      "rewards/chosen": 0.0012034501495032475,
+      "rewards/margins": -0.001115835549796156,
+      "rewards/rejected": 0.0023192856992994035,
+      "stability/repetition_rate_mean": 0.18387754261493683,
+      "stability/response_length_mean": 43.375,
+      "stability/response_length_std": 28.54538917541504,
+      "stability/response_length_var": 814.8392944335938,
+      "stability/token_entropy_mean": 4.73828125,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.6027837991714478,
+      "kl": 0.01663057506084442,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57733023.7852349,
+      "logits/rejected": -51524224.74853801,
+      "logps/chosen": -468.56889681208054,
+      "logps/rejected": -198.22884685672514,
+      "loss": 0.5872,
+      "loss/base_kto": 0.5000787377357483,
+      "metrics/advantage_var": 0.08475451171398163,
+      "regularization/grad_norm_sq": 4545.20458984375,
+      "regularization/w2": 0.09177713841199875,
+      "rewards/chosen": 0.001230826113848078,
+      "rewards/margins": -0.0006731302826898365,
+      "rewards/rejected": 0.0019039563965379145,
+      "stability/repetition_rate_mean": 0.1875,
+      "stability/response_length_mean": 48.375,
+      "stability/response_length_std": 28.93311309814453,
+      "stability/response_length_var": 837.125,
+      "stability/token_entropy_mean": 4.52734375,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6801803708076477,
+      "kl": 0.016775427386164665,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57625172.25316456,
+      "logits/rejected": -53896609.18518519,
+      "logps/chosen": -385.9157189477848,
+      "logps/rejected": -195.12061149691357,
+      "loss": 0.5878,
+      "loss/base_kto": 0.4998288154602051,
+      "metrics/advantage_var": 0.062246084213256836,
+      "regularization/grad_norm_sq": 3760.884521484375,
+      "regularization/w2": 0.08562777936458588,
+      "rewards/chosen": 0.002080178713496727,
+      "rewards/margins": 0.0014936748957853799,
+      "rewards/rejected": 0.0005865038177113475,
+      "stability/repetition_rate_mean": 0.2854352593421936,
+      "stability/response_length_mean": 50.5,
+      "stability/response_length_std": 25.048524856567383,
+      "stability/response_length_var": 627.4285888671875,
+      "stability/token_entropy_mean": 5.5234375,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.5847268104553223,
+      "kl": 0.017838044092059135,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56634124.8,
+      "logits/rejected": -50225545.6,
+      "logps/chosen": -421.87763671875,
+      "logps/rejected": -219.2741943359375,
+      "loss": 0.5859,
+      "loss/base_kto": 0.5001553297042847,
+      "metrics/advantage_var": 0.06722141057252884,
+      "regularization/grad_norm_sq": 4253.48828125,
+      "regularization/w2": 0.0906253382563591,
+      "rewards/chosen": 0.002502601221203804,
+      "rewards/margins": -0.0010038755834102628,
+      "rewards/rejected": 0.003506476804614067,
+      "stability/repetition_rate_mean": 0.16303454339504242,
+      "stability/response_length_mean": 35.0,
+      "stability/response_length_std": 31.518701553344727,
+      "stability/response_length_var": 993.4285888671875,
+      "stability/token_entropy_mean": 4.8115234375,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.6392990946769714,
+      "kl": 0.044592004269361496,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52689789.422818795,
+      "logits/rejected": -50767135.438596494,
+      "logps/chosen": -352.29244442114094,
+      "logps/rejected": -243.81542397660817,
+      "loss": 0.5871,
+      "loss/base_kto": 0.5003380179405212,
+      "metrics/advantage_var": 0.0578509084880352,
+      "regularization/grad_norm_sq": 3660.001953125,
+      "regularization/w2": 0.08328904211521149,
+      "rewards/chosen": 0.0036000581395706074,
+      "rewards/margins": -0.0022749988076132804,
+      "rewards/rejected": 0.005875056947183888,
+      "stability/repetition_rate_mean": 0.0915892943739891,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.51875877380371,
+      "stability/response_length_var": 871.3571166992188,
+      "stability/token_entropy_mean": 4.705078125,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.5789488554000854,
+      "kl": 0.04660215601325035,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54035864.912751675,
+      "logits/rejected": -48294564.67836257,
+      "logps/chosen": -375.1398385067114,
+      "logps/rejected": -180.18667763157896,
+      "loss": 0.5843,
+      "loss/base_kto": 0.4998607635498047,
+      "metrics/advantage_var": 0.06304984539747238,
+      "regularization/grad_norm_sq": 3861.982421875,
+      "regularization/w2": 0.08533728122711182,
+      "rewards/chosen": 0.005930147715062903,
+      "rewards/margins": 0.0005367304570512354,
+      "rewards/rejected": 0.005393417258011668,
+      "stability/repetition_rate_mean": 0.4206075370311737,
+      "stability/response_length_mean": 55.375,
+      "stability/response_length_std": 16.361431121826172,
+      "stability/response_length_var": 267.6964416503906,
+      "stability/token_entropy_mean": 3.806640625,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.6636475324630737,
+      "kl": 0.04949667677283287,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56956520.94409938,
+      "logits/rejected": -49569669.635220125,
+      "logps/chosen": -344.4948078416149,
+      "logps/rejected": -230.40541470125785,
+      "loss": 0.5859,
+      "loss/base_kto": 0.5003251433372498,
+      "metrics/advantage_var": 0.05528384447097778,
+      "regularization/grad_norm_sq": 3821.798583984375,
+      "regularization/w2": 0.08618461340665817,
+      "rewards/chosen": 0.0017190375683470542,
+      "rewards/margins": -0.0025420666225273456,
+      "rewards/rejected": 0.0042611041908743996,
+      "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": 4.3310546875,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.5860218405723572,
+      "kl": 0.052175115793943405,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52664307.587878786,
+      "logits/rejected": -47816974.86451613,
+      "logps/chosen": -359.62769886363634,
+      "logps/rejected": -178.1703377016129,
+      "loss": 0.5865,
+      "loss/base_kto": 0.4995388090610504,
+      "metrics/advantage_var": 0.05257711559534073,
+      "regularization/grad_norm_sq": 3147.68408203125,
+      "regularization/w2": 0.07853565365076065,
+      "rewards/chosen": 0.009310866124702222,
+      "rewards/margins": 0.003652031144322071,
+      "rewards/rejected": 0.005658834980380151,
+      "stability/repetition_rate_mean": 0.19273173809051514,
+      "stability/response_length_mean": 51.125,
+      "stability/response_length_std": 21.90523338317871,
+      "stability/response_length_var": 479.83929443359375,
+      "stability/token_entropy_mean": 5.384765625,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.5740265846252441,
+      "kl": 0.08242836594581604,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57557335.535483874,
+      "logits/rejected": -47273723.345454544,
+      "logps/chosen": -381.8203629032258,
+      "logps/rejected": -183.4941287878788,
+      "loss": 0.5876,
+      "loss/base_kto": 0.4996063709259033,
+      "metrics/advantage_var": 0.060703542083501816,
+      "regularization/grad_norm_sq": 3963.318359375,
+      "regularization/w2": 0.08724621683359146,
+      "rewards/chosen": 0.012640624661599436,
+      "rewards/margins": 0.003818599709317831,
+      "rewards/rejected": 0.008822024952281605,
+      "stability/repetition_rate_mean": 0.18141163885593414,
+      "stability/response_length_mean": 20.625,
+      "stability/response_length_std": 27.176342010498047,
+      "stability/response_length_var": 738.5535888671875,
+      "stability/token_entropy_mean": 5.541015625,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5525412559509277,
+      "kl": 0.08330817520618439,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52759789.83225807,
+      "logits/rejected": -49030733.57575758,
+      "logps/chosen": -379.01844758064516,
+      "logps/rejected": -220.32575757575756,
+      "loss": 0.5896,
+      "loss/base_kto": 0.4998234808444977,
+      "metrics/advantage_var": 0.07535155117511749,
+      "regularization/grad_norm_sq": 4505.4248046875,
+      "regularization/w2": 0.09144462645053864,
+      "rewards/chosen": 0.01100011948616274,
+      "rewards/margins": 0.001358813158111722,
+      "rewards/rejected": 0.009641306328051017,
+      "stability/repetition_rate_mean": 0.1875,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.734375,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.6087732315063477,
+      "kl": 0.08941509574651718,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49510995.027027026,
+      "logits/rejected": -47899594.41860465,
+      "logps/chosen": -314.11993243243245,
+      "logps/rejected": -228.82019894622093,
+      "loss": 0.5871,
+      "loss/base_kto": 0.4989282786846161,
+      "metrics/advantage_var": 0.0716477707028389,
+      "regularization/grad_norm_sq": 3729.305419921875,
+      "regularization/w2": 0.08471358567476273,
+      "rewards/chosen": 0.01508830366907893,
+      "rewards/margins": 0.008622753590176797,
+      "rewards/rejected": 0.0064655500789021336,
+      "stability/repetition_rate_mean": 0.07421875,
+      "stability/response_length_mean": 24.25,
+      "stability/response_length_std": 32.92306900024414,
+      "stability/response_length_var": 1083.9285888671875,
+      "stability/token_entropy_mean": 1.901123046875,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.6029742360115051,
+      "kl": 0.12230757623910904,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58619103.78807947,
+      "logits/rejected": -53444892.78106509,
+      "logps/chosen": -403.8790873344371,
+      "logps/rejected": -200.96717825443787,
+      "loss": 0.586,
+      "loss/base_kto": 0.49925336241722107,
+      "metrics/advantage_var": 0.08379612118005753,
+      "regularization/grad_norm_sq": 3586.81494140625,
+      "regularization/w2": 0.08361315727233887,
+      "rewards/chosen": 0.015101601745908624,
+      "rewards/margins": 0.006237886015433373,
+      "rewards/rejected": 0.008863715730475251,
+      "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": 5.22265625,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.6115427613258362,
+      "kl": 0.12014111131429672,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51479532.18064516,
+      "logits/rejected": -47144202.86060606,
+      "logps/chosen": -358.09873991935484,
+      "logps/rejected": -230.32935606060607,
+      "loss": 0.5852,
+      "loss/base_kto": 0.4983179271221161,
+      "metrics/advantage_var": 0.0890185683965683,
+      "regularization/grad_norm_sq": 3845.35009765625,
+      "regularization/w2": 0.08649836480617523,
+      "rewards/chosen": 0.023378246061263545,
+      "rewards/margins": 0.013833501425190287,
+      "rewards/rejected": 0.009544744636073257,
+      "stability/repetition_rate_mean": 0.1291852593421936,
+      "stability/response_length_mean": 28.0,
+      "stability/response_length_std": 31.181495666503906,
+      "stability/response_length_var": 972.2857055664062,
+      "stability/token_entropy_mean": 4.955078125,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.6375041604042053,
+      "kl": 0.16934870183467865,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58404365.12820513,
+      "logits/rejected": -50561935.6097561,
+      "logps/chosen": -380.53598257211536,
+      "logps/rejected": -187.11123285060975,
+      "loss": 0.5845,
+      "loss/base_kto": 0.4982176721096039,
+      "metrics/advantage_var": 0.09395573288202286,
+      "regularization/grad_norm_sq": 3960.31103515625,
+      "regularization/w2": 0.0869247317314148,
+      "rewards/chosen": 0.025340945292741824,
+      "rewards/margins": 0.013783786392271555,
+      "rewards/rejected": 0.01155715890047027,
+      "stability/repetition_rate_mean": 0.3175307810306549,
+      "stability/response_length_mean": 35.625,
+      "stability/response_length_std": 31.057262420654297,
+      "stability/response_length_var": 964.5535888671875,
+      "stability/token_entropy_mean": 3.80078125,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.6220579147338867,
+      "kl": 0.16499601304531097,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50380346.32911392,
+      "logits/rejected": -48968169.87654321,
+      "logps/chosen": -314.7436956091772,
+      "logps/rejected": -207.06233121141975,
+      "loss": 0.5842,
+      "loss/base_kto": 0.49793797731399536,
+      "metrics/advantage_var": 0.13462769985198975,
+      "regularization/grad_norm_sq": 3773.901123046875,
+      "regularization/w2": 0.08464869111776352,
+      "rewards/chosen": 0.02496972416020647,
+      "rewards/margins": 0.016831111695584253,
+      "rewards/rejected": 0.008138612464622216,
+      "stability/repetition_rate_mean": 0.09133065491914749,
+      "stability/response_length_mean": 21.875,
+      "stability/response_length_std": 27.304826736450195,
+      "stability/response_length_var": 745.5535888671875,
+      "stability/token_entropy_mean": 4.09375,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.6105950474739075,
+      "kl": 0.20538730919361115,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54697513.51351351,
+      "logits/rejected": -50905552.37209302,
+      "logps/chosen": -359.57635663006755,
+      "logps/rejected": -216.37804324127907,
+      "loss": 0.5846,
+      "loss/base_kto": 0.4987795054912567,
+      "metrics/advantage_var": 0.12774451076984406,
+      "regularization/grad_norm_sq": 3816.7734375,
+      "regularization/w2": 0.08535917103290558,
+      "rewards/chosen": 0.029381449158127244,
+      "rewards/margins": 0.009296495572790869,
+      "rewards/rejected": 0.020084953585336376,
+      "stability/repetition_rate_mean": 0.2265625,
+      "stability/response_length_mean": 41.25,
+      "stability/response_length_std": 31.476749420166016,
+      "stability/response_length_var": 990.7857055664062,
+      "stability/token_entropy_mean": 4.166015625,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6735162138938904,
+      "kl": 0.22678242623806,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59375134.84337349,
+      "logits/rejected": -53807054.12987013,
+      "logps/chosen": -399.4981645331325,
+      "logps/rejected": -209.86383928571428,
+      "loss": 0.5847,
+      "loss/base_kto": 0.49670204520225525,
+      "metrics/advantage_var": 0.2160111963748932,
+      "regularization/grad_norm_sq": 3959.833740234375,
+      "regularization/w2": 0.08724039793014526,
+      "rewards/chosen": 0.04043001151946654,
+      "rewards/margins": 0.02643077232863418,
+      "rewards/rejected": 0.01399923919083236,
+      "stability/repetition_rate_mean": 0.1897321343421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.359375,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.6461487412452698,
+      "kl": 0.2732836902141571,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57406969.8313253,
+      "logits/rejected": -53620639.584415585,
+      "logps/chosen": -378.2289627259036,
+      "logps/rejected": -205.90453784496754,
+      "loss": 0.5829,
+      "loss/base_kto": 0.4979163706302643,
+      "metrics/advantage_var": 0.1939479112625122,
+      "regularization/grad_norm_sq": 3633.171875,
+      "regularization/w2": 0.08427540957927704,
+      "rewards/chosen": 0.033429114215345265,
+      "rewards/margins": 0.015068041179646833,
+      "rewards/rejected": 0.018361073035698432,
+      "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": 4.416015625,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.6481019854545593,
+      "kl": 0.31861141324043274,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60599995.94936709,
+      "logits/rejected": -53401410.37037037,
+      "logps/chosen": -364.0679390822785,
+      "logps/rejected": -200.95594618055554,
+      "loss": 0.5815,
+      "loss/base_kto": 0.4965081214904785,
+      "metrics/advantage_var": 0.2759661376476288,
+      "regularization/grad_norm_sq": 3353.943115234375,
+      "regularization/w2": 0.08032859116792679,
+      "rewards/chosen": 0.052460066879851906,
+      "rewards/margins": 0.0283992906756132,
+      "rewards/rejected": 0.024060776204238705,
+      "stability/repetition_rate_mean": 0.3053385615348816,
+      "stability/response_length_mean": 49.0,
+      "stability/response_length_std": 27.795169830322266,
+      "stability/response_length_var": 772.5714111328125,
+      "stability/token_entropy_mean": 4.779296875,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.6184019446372986,
+      "kl": 0.3490689694881439,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57745769.4117647,
+      "logits/rejected": -54368427.68862276,
+      "logps/chosen": -383.6613562091503,
+      "logps/rejected": -211.98376684131736,
+      "loss": 0.5837,
+      "loss/base_kto": 0.49555841088294983,
+      "metrics/advantage_var": 0.4176083505153656,
+      "regularization/grad_norm_sq": 4317.544921875,
+      "regularization/w2": 0.08935306966304779,
+      "rewards/chosen": 0.05835722630320032,
+      "rewards/margins": 0.03509791492412741,
+      "rewards/rejected": 0.02325931137907291,
+      "stability/repetition_rate_mean": 0.1742897778749466,
+      "stability/response_length_mean": 27.875,
+      "stability/response_length_std": 30.31943130493164,
+      "stability/response_length_var": 919.2678833007812,
+      "stability/token_entropy_mean": 4.634765625,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.6115716099739075,
+      "kl": 0.39925938844680786,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60688299.83561644,
+      "logits/rejected": -55668524.137931034,
+      "logps/chosen": -416.8091823630137,
+      "logps/rejected": -216.12354076867817,
+      "loss": 0.582,
+      "loss/base_kto": 0.49605321884155273,
+      "metrics/advantage_var": 0.6236904263496399,
+      "regularization/grad_norm_sq": 3812.632568359375,
+      "regularization/w2": 0.08550175279378891,
+      "rewards/chosen": 0.057012499195255646,
+      "rewards/margins": 0.032647929103745005,
+      "rewards/rejected": 0.02436457009151064,
+      "stability/repetition_rate_mean": 0.16215278208255768,
+      "stability/response_length_mean": 15.25,
+      "stability/response_length_std": 24.93276596069336,
+      "stability/response_length_var": 621.6428833007812,
+      "stability/token_entropy_mean": 3.64453125,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.593185305595398,
+      "kl": 0.4498286247253418,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57376213.05806451,
+      "logits/rejected": -53835838.06060606,
+      "logps/chosen": -391.9687247983871,
+      "logps/rejected": -183.66205018939394,
+      "loss": 0.5824,
+      "loss/base_kto": 0.4946207106113434,
+      "metrics/advantage_var": 0.780052661895752,
+      "regularization/grad_norm_sq": 3791.43017578125,
+      "regularization/w2": 0.08550234884023666,
+      "rewards/chosen": 0.06892158754410282,
+      "rewards/margins": 0.04323520790913005,
+      "rewards/rejected": 0.025686379634972775,
+      "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": 3.6328125,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5935289263725281,
+      "kl": 0.5017997026443481,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59854006.18404908,
+      "logits/rejected": -55003403.41401274,
+      "logps/chosen": -400.9800373849693,
+      "logps/rejected": -247.51960589171975,
+      "loss": 0.5789,
+      "loss/base_kto": 0.49386605620384216,
+      "metrics/advantage_var": 1.076611042022705,
+      "regularization/grad_norm_sq": 4160.88037109375,
+      "regularization/w2": 0.08902523666620255,
+      "rewards/chosen": 0.08361249467346565,
+      "rewards/margins": 0.048496560192178954,
+      "rewards/rejected": 0.0351159344812867,
+      "stability/repetition_rate_mean": 0.2270585298538208,
+      "stability/response_length_mean": 29.875,
+      "stability/response_length_std": 29.502723693847656,
+      "stability/response_length_var": 870.4107055664062,
+      "stability/token_entropy_mean": 3.8955078125,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.7262139916419983,
+      "kl": 0.48766636848449707,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57086696.72727273,
+      "logits/rejected": -50077061.78064516,
+      "logps/chosen": -366.3728456439394,
+      "logps/rejected": -189.92496219758064,
+      "loss": 0.5811,
+      "loss/base_kto": 0.4916563630104065,
+      "metrics/advantage_var": 1.9005340337753296,
+      "regularization/grad_norm_sq": 5850.53369140625,
+      "regularization/w2": 0.09640365839004517,
+      "rewards/chosen": 0.08169195001775569,
+      "rewards/margins": 0.07074021258312237,
+      "rewards/rejected": 0.010951737434633317,
+      "stability/repetition_rate_mean": 0.2662760317325592,
+      "stability/response_length_mean": 52.75,
+      "stability/response_length_std": 22.72663688659668,
+      "stability/response_length_var": 516.5,
+      "stability/token_entropy_mean": 5.19921875,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.6740468144416809,
+      "kl": 0.5717339515686035,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51144241.34939759,
+      "logits/rejected": -51662345.97402597,
+      "logps/chosen": -329.4772684487952,
+      "logps/rejected": -199.15458857548703,
+      "loss": 0.5778,
+      "loss/base_kto": 0.4942333400249481,
+      "metrics/advantage_var": 0.8122803568840027,
+      "regularization/grad_norm_sq": 3395.665771484375,
+      "regularization/w2": 0.08150183409452438,
+      "rewards/chosen": 0.07945383600441806,
+      "rewards/margins": 0.04612136428073769,
+      "rewards/rejected": 0.033332471723680375,
+      "stability/repetition_rate_mean": 0.19258596003055573,
+      "stability/response_length_mean": 22.625,
+      "stability/response_length_std": 26.359804153442383,
+      "stability/response_length_var": 694.8392944335938,
+      "stability/token_entropy_mean": 5.59765625,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.6410212516784668,
+      "kl": 0.6380648016929626,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55422674.823529415,
+      "logits/rejected": -50191220.053333335,
+      "logps/chosen": -396.8014705882353,
+      "logps/rejected": -207.36661458333333,
+      "loss": 0.5788,
+      "loss/base_kto": 0.48983898758888245,
+      "metrics/advantage_var": 1.331944465637207,
+      "regularization/grad_norm_sq": 4066.153076171875,
+      "regularization/w2": 0.08941753953695297,
+      "rewards/chosen": 0.11329341215245864,
+      "rewards/margins": 0.0773618948693369,
+      "rewards/rejected": 0.035931517283121744,
+      "stability/repetition_rate_mean": 0.2115885466337204,
+      "stability/response_length_mean": 35.25,
+      "stability/response_length_std": 31.01957893371582,
+      "stability/response_length_var": 962.2142944335938,
+      "stability/token_entropy_mean": 4.6640625,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.6439304947853088,
+      "kl": 0.6510013341903687,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54019288.847058825,
+      "logits/rejected": -49348276.906666666,
+      "logps/chosen": -360.07109375,
+      "logps/rejected": -190.99877604166667,
+      "loss": 0.5778,
+      "loss/base_kto": 0.4926762580871582,
+      "metrics/advantage_var": 2.249819755554199,
+      "regularization/grad_norm_sq": 5026.2802734375,
+      "regularization/w2": 0.09407851099967957,
+      "rewards/chosen": 0.09181964537676643,
+      "rewards/margins": 0.06113565108355354,
+      "rewards/rejected": 0.03068399429321289,
+      "stability/repetition_rate_mean": 0.3444148898124695,
+      "stability/response_length_mean": 61.875,
+      "stability/response_length_std": 6.010407447814941,
+      "stability/response_length_var": 36.125,
+      "stability/token_entropy_mean": 4.5625,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.615806519985199,
+      "kl": 0.7197233438491821,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55244385.52380952,
+      "logits/rejected": -48958032.84210526,
+      "logps/chosen": -415.5634765625,
+      "logps/rejected": -186.464111328125,
+      "loss": 0.5777,
+      "loss/base_kto": 0.4885575473308563,
+      "metrics/advantage_var": 1.9970428943634033,
+      "regularization/grad_norm_sq": 4003.491455078125,
+      "regularization/w2": 0.08765041083097458,
+      "rewards/chosen": 0.13318413779849098,
+      "rewards/margins": 0.08986123582175501,
+      "rewards/rejected": 0.04332290197673597,
+      "stability/repetition_rate_mean": 0.30851560831069946,
+      "stability/response_length_mean": 32.5,
+      "stability/response_length_std": 30.528675079345703,
+      "stability/response_length_var": 932.0,
+      "stability/token_entropy_mean": 4.44921875,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.7225887179374695,
+      "kl": 0.7622379660606384,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54423646.16091954,
+      "logits/rejected": -49801538.63013699,
+      "logps/chosen": -369.7487652658046,
+      "logps/rejected": -212.1081843964041,
+      "loss": 0.575,
+      "loss/base_kto": 0.4914153218269348,
+      "metrics/advantage_var": 1.7091789245605469,
+      "regularization/grad_norm_sq": 3531.582763671875,
+      "regularization/w2": 0.0829373374581337,
+      "rewards/chosen": 0.11679701969541352,
+      "rewards/margins": 0.06840403473033733,
+      "rewards/rejected": 0.048392984965076186,
+      "stability/repetition_rate_mean": 0.2411106526851654,
+      "stability/response_length_mean": 22.5,
+      "stability/response_length_std": 28.33977508544922,
+      "stability/response_length_var": 803.1428833007812,
+      "stability/token_entropy_mean": 4.8671875,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.6213456988334656,
+      "kl": 0.9274574518203735,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59135054.76923077,
+      "logits/rejected": -52582671.25827815,
+      "logps/chosen": -439.6777921597633,
+      "logps/rejected": -190.59510813327816,
+      "loss": 0.5717,
+      "loss/base_kto": 0.4875049293041229,
+      "metrics/advantage_var": 2.3025033473968506,
+      "regularization/grad_norm_sq": 4036.350830078125,
+      "regularization/w2": 0.08665412664413452,
+      "rewards/chosen": 0.1664069587662375,
+      "rewards/margins": 0.1001336926497226,
+      "rewards/rejected": 0.0662732661165149,
+      "stability/repetition_rate_mean": 0.22428728640079498,
+      "stability/response_length_mean": 36.25,
+      "stability/response_length_std": 30.76988410949707,
+      "stability/response_length_var": 946.7857055664062,
+      "stability/token_entropy_mean": 3.7744140625,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.7222792506217957,
+      "kl": 0.9340354204177856,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56650924.72289157,
+      "logits/rejected": -48021038.54545455,
+      "logps/chosen": -354.87575301204816,
+      "logps/rejected": -194.314453125,
+      "loss": 0.5718,
+      "loss/base_kto": 0.48661375045776367,
+      "metrics/advantage_var": 4.268935680389404,
+      "regularization/grad_norm_sq": 4183.99560546875,
+      "regularization/w2": 0.08717496693134308,
+      "rewards/chosen": 0.15614137304834572,
+      "rewards/margins": 0.10683375422597546,
+      "rewards/rejected": 0.04930761882237026,
+      "stability/repetition_rate_mean": 0.2011384218931198,
+      "stability/response_length_mean": 28.125,
+      "stability/response_length_std": 25.357091903686523,
+      "stability/response_length_var": 642.9821166992188,
+      "stability/token_entropy_mean": 4.91015625,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.6457726955413818,
+      "kl": 1.0286387205123901,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52652798.42944785,
+      "logits/rejected": -47983420.33121019,
+      "logps/chosen": -374.4070408358896,
+      "logps/rejected": -210.5137838375796,
+      "loss": 0.5699,
+      "loss/base_kto": 0.48794251680374146,
+      "metrics/advantage_var": 4.224320411682129,
+      "regularization/grad_norm_sq": 3658.589111328125,
+      "regularization/w2": 0.08313455432653427,
+      "rewards/chosen": 0.1738798925481691,
+      "rewards/margins": 0.0954359765675324,
+      "rewards/rejected": 0.0784439159806367,
+      "stability/repetition_rate_mean": 0.1989397406578064,
+      "stability/response_length_mean": 35.25,
+      "stability/response_length_std": 31.129911422729492,
+      "stability/response_length_var": 969.0714111328125,
+      "stability/token_entropy_mean": 3.41015625,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.6506476402282715,
+      "kl": 1.0774658918380737,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54617227.061728396,
+      "logits/rejected": -49519227.1392405,
+      "logps/chosen": -310.31630979938274,
+      "logps/rejected": -196.56126384493672,
+      "loss": 0.5768,
+      "loss/base_kto": 0.486668199300766,
+      "metrics/advantage_var": 7.448278903961182,
+      "regularization/grad_norm_sq": 6933.6611328125,
+      "regularization/w2": 0.09475024789571762,
+      "rewards/chosen": 0.16850583347273462,
+      "rewards/margins": 0.12071590972032412,
+      "rewards/rejected": 0.0477899237524105,
+      "stability/repetition_rate_mean": 0.15234375,
+      "stability/response_length_mean": 32.875,
+      "stability/response_length_std": 33.30996322631836,
+      "stability/response_length_var": 1109.5535888671875,
+      "stability/token_entropy_mean": 3.6953125,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.6974467635154724,
+      "kl": 1.1172609329223633,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53099575.18562874,
+      "logits/rejected": -49682532.39215686,
+      "logps/chosen": -390.274186002994,
+      "logps/rejected": -210.4579248366013,
+      "loss": 0.572,
+      "loss/base_kto": 0.4829126298427582,
+      "metrics/advantage_var": 6.995049953460693,
+      "regularization/grad_norm_sq": 5340.63427734375,
+      "regularization/w2": 0.09194418042898178,
+      "rewards/chosen": 0.21311228860638098,
+      "rewards/margins": 0.13904476766693424,
+      "rewards/rejected": 0.07406752093944674,
+      "stability/repetition_rate_mean": 0.13363486528396606,
+      "stability/response_length_mean": 16.75,
+      "stability/response_length_std": 27.12800407409668,
+      "stability/response_length_var": 735.9285888671875,
+      "stability/token_entropy_mean": 5.107421875,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.6366767287254333,
+      "kl": 1.1507939100265503,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60035851.561290324,
+      "logits/rejected": -53796069.624242425,
+      "logps/chosen": -390.8335685483871,
+      "logps/rejected": -197.07270359848485,
+      "loss": 0.5705,
+      "loss/base_kto": 0.48604080080986023,
+      "metrics/advantage_var": 4.371114253997803,
+      "regularization/grad_norm_sq": 3933.094482421875,
+      "regularization/w2": 0.08576133102178574,
+      "rewards/chosen": 0.19513650709582914,
+      "rewards/margins": 0.11047319690036868,
+      "rewards/rejected": 0.08466331019546046,
+      "stability/repetition_rate_mean": 0.3953683078289032,
+      "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": 0.5821207132266493,
+      "train_runtime": 4296.589,
+      "train_samples_per_second": 23.274,
+      "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_delta0.001/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc269180f5d75bd1ef95cbfdbfbd583fa9ef80416dbb4748e6a57fe1cde09ea8
+size 8081