Upload folder using huggingface_hub

mistral7b_kto_tv_rho0.003/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_tv_rho0.003
+  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_tv_rho0.003
+
+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_tv_rho0.003/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": [
+    "v_proj",
+    "o_proj",
+    "q_proj",
+    "up_proj",
+    "gate_proj",
+    "k_proj",
+    "down_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

mistral7b_kto_tv_rho0.003/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:868c70b2b05dafffac10b6a4adef75bca20826fabb22be5ff4b8004938a81aa3
+size 42002584

mistral7b_kto_tv_rho0.003/all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.5009068980508921,
+    "train_runtime": 3152.3184,
+    "train_samples_per_second": 31.723,
+    "train_steps_per_second": 0.062
+}

mistral7b_kto_tv_rho0.003/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_tv_rho0.003/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_tv_rho0.003/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": [
+    "v_proj",
+    "o_proj",
+    "q_proj",
+    "up_proj",
+    "gate_proj",
+    "k_proj",
+    "down_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

mistral7b_kto_tv_rho0.003/checkpoint-196/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:868c70b2b05dafffac10b6a4adef75bca20826fabb22be5ff4b8004938a81aa3
+size 42002584

mistral7b_kto_tv_rho0.003/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_tv_rho0.003/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 sha256:6d101387d88b3f7cee5136f6b1ca52c4de26791ea35815f60b35c938522ed41b
+size 31467333

mistral7b_kto_tv_rho0.003/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:a11d668ccff324dd03a9c97fb487fa871d38257d8416c000f94730b1c8d22fc0
+size 31467333

mistral7b_kto_tv_rho0.003/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:c120240de69530b76ab7187a58367042b8946aa17058f70ca5f0ac7644e66737
+size 31467333

mistral7b_kto_tv_rho0.003/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 sha256:1ed354a040e1661b8dbcfa52213a9b3107050944cec8bb5476e677ca798f99f5
+size 31467397

mistral7b_kto_tv_rho0.003/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:ceef0ed834edc0bba33c94a1fd2e90ff4ef682048e70fbb2527e5d8812a6a5df
+size 31467397

mistral7b_kto_tv_rho0.003/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:b9a7c9ec2f6d940d57293b388b8d1b52236c552594d641b29d1f3afd745f2b81
+size 31467397

mistral7b_kto_tv_rho0.003/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:d4491eed7fdd183ed5745209c6c3a2fd6132abb0ea5027cbbaa57eb9cd30809d
+size 31467397

mistral7b_kto_tv_rho0.003/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:b38cb142c34e01b7baffd78b1eef11bd48caf1828d199fc14b270571628ca3db
+size 31467397

mistral7b_kto_tv_rho0.003/checkpoint-196/global_step196/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3aef6051ab41c4b66720f28d63cc7a989bbb8998918ba35aeb8ec43679b13b9b
+size 42288409

mistral7b_kto_tv_rho0.003/checkpoint-196/latest

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

mistral7b_kto_tv_rho0.003/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_tv_rho0.003/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_tv_rho0.003/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_tv_rho0.003/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_tv_rho0.003/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_tv_rho0.003/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_tv_rho0.003/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_tv_rho0.003/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_tv_rho0.003/checkpoint-196/scheduler.pt

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

mistral7b_kto_tv_rho0.003/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_tv_rho0.003/checkpoint-196/tokenizer.model

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

mistral7b_kto_tv_rho0.003/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_tv_rho0.003/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": 0.6752910017967224,
+      "kl": 0.009316612966358662,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59711191.578947365,
+      "logits/rejected": -50416146.28571428,
+      "logps/chosen": -439.0247224506579,
+      "logps/rejected": -198.40827287946428,
+      "loss": 0.5011,
+      "loss/base_kto": 0.49975714087486267,
+      "metrics/advantage_var": 0.05327611044049263,
+      "regularization/advantage_range": 0.46398472785949707,
+      "regularization/tv": 0.001391954137943685,
+      "rewards/chosen": 0.002090120393978922,
+      "rewards/margins": 0.0019776263836780586,
+      "rewards/rejected": 0.00011249401030086336,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.7144330739974976,
+      "kl": 0.01776806451380253,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59301914.597402595,
+      "logits/rejected": -50024296.86746988,
+      "logps/chosen": -429.9064021915584,
+      "logps/rejected": -192.8597985692771,
+      "loss": 0.5019,
+      "loss/base_kto": 0.49971505999565125,
+      "metrics/advantage_var": 0.06181863695383072,
+      "regularization/advantage_range": 0.5836383700370789,
+      "regularization/tv": 0.0017509151948615909,
+      "rewards/chosen": 0.0032109109225211207,
+      "rewards/margins": 0.0023968832870731706,
+      "rewards/rejected": 0.0008140276354479503,
+      "stability/repetition_rate_mean": 0.3531753420829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.3056640625,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.6771201491355896,
+      "kl": 0.01534965168684721,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56231693.96363636,
+      "logits/rejected": -51925735.22580645,
+      "logps/chosen": -404.1009943181818,
+      "logps/rejected": -200.66796875,
+      "loss": 0.5018,
+      "loss/base_kto": 0.49993228912353516,
+      "metrics/advantage_var": 0.05437135323882103,
+      "regularization/advantage_range": 0.526348888874054,
+      "regularization/tv": 0.001579046598635614,
+      "rewards/chosen": 0.002984200824390758,
+      "rewards/margins": 0.0003753908918050727,
+      "rewards/rejected": 0.0026088099325856855,
+      "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": 3.7392578125,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.7161915302276611,
+      "kl": 0.021325770765542984,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51672825.6,
+      "logits/rejected": -48064236.8,
+      "logps/chosen": -343.6939697265625,
+      "logps/rejected": -225.6203369140625,
+      "loss": 0.5019,
+      "loss/base_kto": 0.5005207061767578,
+      "metrics/advantage_var": 0.052058931440114975,
+      "regularization/advantage_range": 0.529035210609436,
+      "regularization/tv": 0.0015871055657044053,
+      "rewards/chosen": 0.0002750195562839508,
+      "rewards/margins": -0.004027188569307327,
+      "rewards/rejected": 0.004302208125591278,
+      "stability/repetition_rate_mean": 0.2194320410490036,
+      "stability/response_length_mean": 44.0,
+      "stability/response_length_std": 28.0611572265625,
+      "stability/response_length_var": 787.4285888671875,
+      "stability/token_entropy_mean": 4.4296875,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.6439714431762695,
+      "kl": 0.012540136463940144,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56468112.77241379,
+      "logits/rejected": -52019515.977142856,
+      "logps/chosen": -378.4657058189655,
+      "logps/rejected": -202.78388392857144,
+      "loss": 0.5019,
+      "loss/base_kto": 0.499775230884552,
+      "metrics/advantage_var": 0.05616157129406929,
+      "regularization/advantage_range": 0.5223817825317383,
+      "regularization/tv": 0.0015671454602852464,
+      "rewards/chosen": 0.002574328924047536,
+      "rewards/margins": 0.0016784496554012956,
+      "rewards/rejected": 0.0008958792686462403,
+      "stability/repetition_rate_mean": 0.16044004261493683,
+      "stability/response_length_mean": 43.375,
+      "stability/response_length_std": 28.54538917541504,
+      "stability/response_length_var": 814.8392944335938,
+      "stability/token_entropy_mean": 4.806640625,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.6126549243927002,
+      "kl": 0.01668705977499485,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57719072.644295305,
+      "logits/rejected": -51523212.7251462,
+      "logps/chosen": -468.5598259228188,
+      "logps/rejected": -198.2305144371345,
+      "loss": 0.502,
+      "loss/base_kto": 0.4999951422214508,
+      "metrics/advantage_var": 0.05605809763073921,
+      "regularization/advantage_range": 0.542860746383667,
+      "regularization/tv": 0.0016285822493955493,
+      "rewards/chosen": 0.0021430586008417524,
+      "rewards/margins": 0.0004062107972351279,
+      "rewards/rejected": 0.0017368478036066245,
+      "stability/repetition_rate_mean": 0.17223010957241058,
+      "stability/response_length_mean": 45.875,
+      "stability/response_length_std": 28.246049880981445,
+      "stability/response_length_var": 797.8392944335938,
+      "stability/token_entropy_mean": 4.57421875,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6238469481468201,
+      "kl": 0.018097203224897385,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57647505.82278481,
+      "logits/rejected": -53915256.09876543,
+      "logps/chosen": -385.8966327136076,
+      "logps/rejected": -195.1175250771605,
+      "loss": 0.5017,
+      "loss/base_kto": 0.4996470510959625,
+      "metrics/advantage_var": 0.07482800632715225,
+      "regularization/advantage_range": 0.5923306345939636,
+      "regularization/tv": 0.001776991761289537,
+      "rewards/chosen": 0.003989132899272291,
+      "rewards/margins": 0.0030932442741685405,
+      "rewards/rejected": 0.0008958886251037503,
+      "stability/repetition_rate_mean": 0.3147321343421936,
+      "stability/response_length_mean": 56.875,
+      "stability/response_length_std": 20.152544021606445,
+      "stability/response_length_var": 406.125,
+      "stability/token_entropy_mean": 5.421875,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.5992166996002197,
+      "kl": 0.028051286935806274,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56631635.2,
+      "logits/rejected": -50227708.8,
+      "logps/chosen": -421.901123046875,
+      "logps/rejected": -219.2706787109375,
+      "loss": 0.5019,
+      "loss/base_kto": 0.5005082488059998,
+      "metrics/advantage_var": 0.07147374749183655,
+      "regularization/advantage_range": 0.6278894543647766,
+      "regularization/tv": 0.0018836684757843614,
+      "rewards/chosen": 0.00015102624893188478,
+      "rewards/margins": -0.0037047527730464935,
+      "rewards/rejected": 0.0038557790219783785,
+      "stability/repetition_rate_mean": 0.13671875,
+      "stability/response_length_mean": 33.125,
+      "stability/response_length_std": 33.01704025268555,
+      "stability/response_length_var": 1090.125,
+      "stability/token_entropy_mean": 4.8896484375,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.6807482838630676,
+      "kl": 0.03983687236905098,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52712994.36241611,
+      "logits/rejected": -50780387.55555555,
+      "logps/chosen": -352.2607487416107,
+      "logps/rejected": -243.8278965643275,
+      "loss": 0.5015,
+      "loss/base_kto": 0.4997825622558594,
+      "metrics/advantage_var": 0.08648205548524857,
+      "regularization/advantage_range": 0.661689281463623,
+      "regularization/tv": 0.001985067967325449,
+      "rewards/chosen": 0.006769456319360925,
+      "rewards/margins": 0.002141746121956811,
+      "rewards/rejected": 0.004627710197404114,
+      "stability/repetition_rate_mean": 0.0994017943739891,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.51875877380371,
+      "stability/response_length_var": 871.3571166992188,
+      "stability/token_entropy_mean": 4.70703125,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.6231204271316528,
+      "kl": 0.03668095916509628,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54021566.7114094,
+      "logits/rejected": -48286264.88888889,
+      "logps/chosen": -375.1455012583893,
+      "logps/rejected": -180.17639802631578,
+      "loss": 0.5017,
+      "loss/base_kto": 0.5000646710395813,
+      "metrics/advantage_var": 0.05680838227272034,
+      "regularization/advantage_range": 0.5559825301170349,
+      "regularization/tv": 0.0016679477412253618,
+      "rewards/chosen": 0.005363105927537751,
+      "rewards/margins": -0.0010589264621867125,
+      "rewards/rejected": 0.006422032389724464,
+      "stability/repetition_rate_mean": 0.3288106620311737,
+      "stability/response_length_mean": 55.375,
+      "stability/response_length_std": 16.361431121826172,
+      "stability/response_length_var": 267.6964416503906,
+      "stability/token_entropy_mean": 4.068359375,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.7771005034446716,
+      "kl": 0.043028466403484344,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56951623.55279503,
+      "logits/rejected": -49562733.48427673,
+      "logps/chosen": -344.49801048136646,
+      "logps/rejected": -230.3948506289308,
+      "loss": 0.5017,
+      "loss/base_kto": 0.5004634857177734,
+      "metrics/advantage_var": 0.06657923012971878,
+      "regularization/advantage_range": 0.5866410136222839,
+      "regularization/tv": 0.0017599230632185936,
+      "rewards/chosen": 0.0014025861432093271,
+      "rewards/margins": -0.003914588270459707,
+      "rewards/rejected": 0.005317174413669034,
+      "stability/repetition_rate_mean": 0.111328125,
+      "stability/response_length_mean": 24.5,
+      "stability/response_length_std": 32.710853576660156,
+      "stability/response_length_var": 1070.0,
+      "stability/token_entropy_mean": 4.388671875,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.6669939160346985,
+      "kl": 0.03918389976024628,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52672884.36363637,
+      "logits/rejected": -47833497.6,
+      "logps/chosen": -359.6323863636364,
+      "logps/rejected": -178.17522681451612,
+      "loss": 0.5016,
+      "loss/base_kto": 0.49955543875694275,
+      "metrics/advantage_var": 0.06515418738126755,
+      "regularization/advantage_range": 0.5985627770423889,
+      "regularization/tv": 0.0017956882948055863,
+      "rewards/chosen": 0.008838949781475645,
+      "rewards/margins": 0.0036711697820810745,
+      "rewards/rejected": 0.0051677799993945705,
+      "stability/repetition_rate_mean": 0.21596631407737732,
+      "stability/response_length_mean": 52.625,
+      "stability/response_length_std": 21.75800323486328,
+      "stability/response_length_var": 473.41070556640625,
+      "stability/token_entropy_mean": 5.29296875,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.6477825045585632,
+      "kl": 0.04982197284698486,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57567218.787096776,
+      "logits/rejected": -47273661.28484848,
+      "logps/chosen": -381.8376512096774,
+      "logps/rejected": -183.54024621212122,
+      "loss": 0.5013,
+      "loss/base_kto": 0.49914589524269104,
+      "metrics/advantage_var": 0.0746908187866211,
+      "regularization/advantage_range": 0.6157255172729492,
+      "regularization/tv": 0.001847176463343203,
+      "rewards/chosen": 0.01091364276024603,
+      "rewards/margins": 0.006703484932348525,
+      "rewards/rejected": 0.004210157827897506,
+      "stability/repetition_rate_mean": 0.1857219785451889,
+      "stability/response_length_mean": 20.625,
+      "stability/response_length_std": 27.176342010498047,
+      "stability/response_length_var": 738.5535888671875,
+      "stability/token_entropy_mean": 5.544921875,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5892289280891418,
+      "kl": 0.06117330864071846,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52787001.80645161,
+      "logits/rejected": -49036244.557575755,
+      "logps/chosen": -379.0451612903226,
+      "logps/rejected": -220.39483901515152,
+      "loss": 0.5011,
+      "loss/base_kto": 0.4993252456188202,
+      "metrics/advantage_var": 0.07192724943161011,
+      "regularization/advantage_range": 0.6263827681541443,
+      "regularization/tv": 0.001879148418083787,
+      "rewards/chosen": 0.008328685452861171,
+      "rewards/margins": 0.005595458474443572,
+      "rewards/rejected": 0.0027332269784175987,
+      "stability/repetition_rate_mean": 0.11328125,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.4404296875,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.6382232308387756,
+      "kl": 0.07038548588752747,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49505947.675675675,
+      "logits/rejected": -47897421.39534884,
+      "logps/chosen": -314.17166385135135,
+      "logps/rejected": -228.84547601744185,
+      "loss": 0.5015,
+      "loss/base_kto": 0.4992240369319916,
+      "metrics/advantage_var": 0.08677860349416733,
+      "regularization/advantage_range": 0.7051105499267578,
+      "regularization/tv": 0.0021153318230062723,
+      "rewards/chosen": 0.009913306783985448,
+      "rewards/margins": 0.005977991895807532,
+      "rewards/rejected": 0.003935314888177916,
+      "stability/repetition_rate_mean": 0.08203125,
+      "stability/response_length_mean": 24.25,
+      "stability/response_length_std": 32.92306900024414,
+      "stability/response_length_var": 1083.9285888671875,
+      "stability/token_entropy_mean": 1.883056640625,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.6527356505393982,
+      "kl": 0.08504229038953781,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58622535.205298014,
+      "logits/rejected": -53446995.313609466,
+      "logps/chosen": -403.9296616307947,
+      "logps/rejected": -201.01654955621302,
+      "loss": 0.5014,
+      "loss/base_kto": 0.49918365478515625,
+      "metrics/advantage_var": 0.06651351600885391,
+      "regularization/advantage_range": 0.5964646935462952,
+      "regularization/tv": 0.0017893938347697258,
+      "rewards/chosen": 0.010046138668691876,
+      "rewards/margins": 0.006118507530347388,
+      "rewards/rejected": 0.003927631138344488,
+      "stability/repetition_rate_mean": 0.19088542461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.23828125,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.6321583986282349,
+      "kl": 0.0836004987359047,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51481600.0,
+      "logits/rejected": -47146399.806060605,
+      "logps/chosen": -358.2110635080645,
+      "logps/rejected": -230.3817234848485,
+      "loss": 0.5012,
+      "loss/base_kto": 0.49900388717651367,
+      "metrics/advantage_var": 0.08175607770681381,
+      "regularization/advantage_range": 0.68501877784729,
+      "regularization/tv": 0.0020550561603158712,
+      "rewards/chosen": 0.012145080104950936,
+      "rewards/margins": 0.007835714500087913,
+      "rewards/rejected": 0.004309365604863022,
+      "stability/repetition_rate_mean": 0.1369977593421936,
+      "stability/response_length_mean": 28.0,
+      "stability/response_length_std": 31.181495666503906,
+      "stability/response_length_var": 972.2857055664062,
+      "stability/token_entropy_mean": 4.96484375,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.5965731143951416,
+      "kl": 0.09354694932699203,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58432328.20512821,
+      "logits/rejected": -50573452.487804875,
+      "logps/chosen": -380.6665665064103,
+      "logps/rejected": -187.1869521722561,
+      "loss": 0.5008,
+      "loss/base_kto": 0.49893808364868164,
+      "metrics/advantage_var": 0.07857906073331833,
+      "regularization/advantage_range": 0.639580249786377,
+      "regularization/tv": 0.001918740919791162,
+      "rewards/chosen": 0.012280461115714831,
+      "rewards/margins": 0.008296260597409122,
+      "rewards/rejected": 0.0039842005183057085,
+      "stability/repetition_rate_mean": 0.2582859992980957,
+      "stability/response_length_mean": 35.125,
+      "stability/response_length_std": 31.638301849365234,
+      "stability/response_length_var": 1000.9821166992188,
+      "stability/token_entropy_mean": 2.94384765625,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.5790473818778992,
+      "kl": 0.08265768736600876,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50400852.25316456,
+      "logits/rejected": -48991902.02469136,
+      "logps/chosen": -314.8856309335443,
+      "logps/rejected": -207.14375964506172,
+      "loss": 0.5009,
+      "loss/base_kto": 0.49863290786743164,
+      "metrics/advantage_var": 0.07834162563085556,
+      "regularization/advantage_range": 0.6240283250808716,
+      "regularization/tv": 0.0018720850348472595,
+      "rewards/chosen": 0.010773800596406188,
+      "rewards/margins": 0.010777551641798071,
+      "rewards/rejected": -3.7510453918833793e-06,
+      "stability/repetition_rate_mean": 0.19374233484268188,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 31.1264705657959,
+      "stability/response_length_var": 968.8571166992188,
+      "stability/token_entropy_mean": 3.736328125,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.5900207161903381,
+      "kl": 0.08847284317016602,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54724199.783783786,
+      "logits/rejected": -50932414.511627905,
+      "logps/chosen": -359.7025971283784,
+      "logps/rejected": -216.51346747819767,
+      "loss": 0.501,
+      "loss/base_kto": 0.4986952841281891,
+      "metrics/advantage_var": 0.08461682498455048,
+      "regularization/advantage_range": 0.6794241070747375,
+      "regularization/tv": 0.002038272563368082,
+      "rewards/chosen": 0.01675953736176362,
+      "rewards/margins": 0.01021829631476939,
+      "rewards/rejected": 0.006541241046994231,
+      "stability/repetition_rate_mean": 0.2421875,
+      "stability/response_length_mean": 41.25,
+      "stability/response_length_std": 31.476749420166016,
+      "stability/response_length_var": 990.7857055664062,
+      "stability/token_entropy_mean": 4.107421875,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6134894490242004,
+      "kl": 0.09840010851621628,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59393332.43373494,
+      "logits/rejected": -53833864.31168831,
+      "logps/chosen": -399.74868222891564,
+      "logps/rejected": -209.9587307224026,
+      "loss": 0.5009,
+      "loss/base_kto": 0.49860039353370667,
+      "metrics/advantage_var": 0.10574636608362198,
+      "regularization/advantage_range": 0.7764409184455872,
+      "regularization/tv": 0.0023293232079595327,
+      "rewards/chosen": 0.015377698174442154,
+      "rewards/margins": 0.010868746132717095,
+      "rewards/rejected": 0.004508952041725059,
+      "stability/repetition_rate_mean": 0.1916852593421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.388671875,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.6036498546600342,
+      "kl": 0.10348369181156158,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57405316.62650602,
+      "logits/rejected": -53642688.83116883,
+      "logps/chosen": -378.44039439006025,
+      "logps/rejected": -206.05346996753246,
+      "loss": 0.5008,
+      "loss/base_kto": 0.4989023208618164,
+      "metrics/advantage_var": 0.1012769564986229,
+      "regularization/advantage_range": 0.7773522734642029,
+      "regularization/tv": 0.002332056639716029,
+      "rewards/chosen": 0.01228764545486634,
+      "rewards/margins": 0.008821119819704383,
+      "rewards/rejected": 0.003466525635161957,
+      "stability/repetition_rate_mean": 0.3186553120613098,
+      "stability/response_length_mean": 31.125,
+      "stability/response_length_std": 28.412460327148438,
+      "stability/response_length_var": 807.2678833007812,
+      "stability/token_entropy_mean": 4.0673828125,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.5774539709091187,
+      "kl": 0.11748156696557999,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60635421.164556965,
+      "logits/rejected": -53458703.802469134,
+      "logps/chosen": -364.39875890031647,
+      "logps/rejected": -201.12389081790124,
+      "loss": 0.5008,
+      "loss/base_kto": 0.4984841048717499,
+      "metrics/advantage_var": 0.08921544253826141,
+      "regularization/advantage_range": 0.6971777081489563,
+      "regularization/tv": 0.0020915332715958357,
+      "rewards/chosen": 0.019377651093881343,
+      "rewards/margins": 0.01210962627842792,
+      "rewards/rejected": 0.007268024815453423,
+      "stability/repetition_rate_mean": 0.2916666865348816,
+      "stability/response_length_mean": 41.375,
+      "stability/response_length_std": 31.25442886352539,
+      "stability/response_length_var": 976.8392944335938,
+      "stability/token_entropy_mean": 4.935546875,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.5666119456291199,
+      "kl": 0.1038687452673912,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57774535.11111111,
+      "logits/rejected": -54427586.68263473,
+      "logps/chosen": -384.06045751633985,
+      "logps/rejected": -212.17372286676647,
+      "loss": 0.5004,
+      "loss/base_kto": 0.4982244670391083,
+      "metrics/advantage_var": 0.11172550171613693,
+      "regularization/advantage_range": 0.8190727233886719,
+      "regularization/tv": 0.0024572184775024652,
+      "rewards/chosen": 0.018445407643037683,
+      "rewards/margins": 0.014180560395851147,
+      "rewards/rejected": 0.0042648472471865356,
+      "stability/repetition_rate_mean": 0.1664772778749466,
+      "stability/response_length_mean": 27.875,
+      "stability/response_length_std": 30.31943130493164,
+      "stability/response_length_var": 919.2678833007812,
+      "stability/token_entropy_mean": 4.623046875,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.5878152251243591,
+      "kl": 0.11783448606729507,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60747207.89041096,
+      "logits/rejected": -55728210.390804596,
+      "logps/chosen": -417.2116866438356,
+      "logps/rejected": -216.30994073275863,
+      "loss": 0.5008,
+      "loss/base_kto": 0.4985935389995575,
+      "metrics/advantage_var": 0.12825100123882294,
+      "regularization/advantage_range": 0.9061002731323242,
+      "regularization/tv": 0.002718301024287939,
+      "rewards/chosen": 0.01676147114740659,
+      "rewards/margins": 0.011036303883893267,
+      "rewards/rejected": 0.005725167263513324,
+      "stability/repetition_rate_mean": 0.16328126192092896,
+      "stability/response_length_mean": 15.25,
+      "stability/response_length_std": 24.93276596069336,
+      "stability/response_length_var": 621.6428833007812,
+      "stability/token_entropy_mean": 3.6328125,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.5727063417434692,
+      "kl": 0.11419554054737091,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57436979.2,
+      "logits/rejected": -53890463.806060605,
+      "logps/chosen": -392.4932459677419,
+      "logps/rejected": -183.90649857954546,
+      "loss": 0.5003,
+      "loss/base_kto": 0.4980848431587219,
+      "metrics/advantage_var": 0.12817950546741486,
+      "regularization/advantage_range": 0.8711000680923462,
+      "regularization/tv": 0.002613300224766135,
+      "rewards/chosen": 0.016469930833385837,
+      "rewards/margins": 0.015228715582676997,
+      "rewards/rejected": 0.0012412152507088401,
+      "stability/repetition_rate_mean": 0.19791193306446075,
+      "stability/response_length_mean": 38.0,
+      "stability/response_length_std": 29.13515281677246,
+      "stability/response_length_var": 848.8571166992188,
+      "stability/token_entropy_mean": 3.99609375,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5919101238250732,
+      "kl": 0.12443824112415314,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59916112.09815951,
+      "logits/rejected": -55112397.4522293,
+      "logps/chosen": -401.59190471625766,
+      "logps/rejected": -247.84937300955414,
+      "loss": 0.5005,
+      "loss/base_kto": 0.4973863661289215,
+      "metrics/advantage_var": 0.18036113679409027,
+      "regularization/advantage_range": 1.082304835319519,
+      "regularization/tv": 0.0032469145953655243,
+      "rewards/chosen": 0.022427939198499808,
+      "rewards/margins": 0.020287977934563772,
+      "rewards/rejected": 0.0021399612639360368,
+      "stability/repetition_rate_mean": 0.2230902761220932,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.936838150024414,
+      "stability/response_length_var": 896.2142944335938,
+      "stability/token_entropy_mean": 4.07421875,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.5441713333129883,
+      "kl": 0.10719374567270279,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57166928.67878788,
+      "logits/rejected": -50186933.67741936,
+      "logps/chosen": -367.0013494318182,
+      "logps/rejected": -190.0579133064516,
+      "loss": 0.5003,
+      "loss/base_kto": 0.49737364053726196,
+      "metrics/advantage_var": 0.17915482819080353,
+      "regularization/advantage_range": 1.0351334810256958,
+      "regularization/tv": 0.0031054005958139896,
+      "rewards/chosen": 0.018839573137687915,
+      "rewards/margins": 0.02118338370369793,
+      "rewards/rejected": -0.0023438105660100136,
+      "stability/repetition_rate_mean": 0.35546875,
+      "stability/response_length_mean": 56.25,
+      "stability/response_length_std": 21.920310974121094,
+      "stability/response_length_var": 480.5,
+      "stability/token_entropy_mean": 4.7451171875,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.5883694887161255,
+      "kl": 0.1362670212984085,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51238381.4939759,
+      "logits/rejected": -51789099.22077922,
+      "logps/chosen": -330.10217432228916,
+      "logps/rejected": -199.48333502435065,
+      "loss": 0.5005,
+      "loss/base_kto": 0.49796611070632935,
+      "metrics/advantage_var": 0.1358262300491333,
+      "regularization/advantage_range": 0.8813913464546204,
+      "regularization/tv": 0.0026441742666065693,
+      "rewards/chosen": 0.016965202538363904,
+      "rewards/margins": 0.016506232452996924,
+      "rewards/rejected": 0.0004589700853669798,
+      "stability/repetition_rate_mean": 0.20039846003055573,
+      "stability/response_length_mean": 22.625,
+      "stability/response_length_std": 26.359804153442383,
+      "stability/response_length_var": 694.8392944335938,
+      "stability/token_entropy_mean": 5.703125,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.5959792733192444,
+      "kl": 0.14728893339633942,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55525086.870588236,
+      "logits/rejected": -50325459.626666665,
+      "logps/chosen": -397.62251838235295,
+      "logps/rejected": -207.71174479166666,
+      "loss": 0.5003,
+      "loss/base_kto": 0.49606361985206604,
+      "metrics/advantage_var": 0.18760840594768524,
+      "regularization/advantage_range": 1.0953562259674072,
+      "regularization/tv": 0.0032860690262168646,
+      "rewards/chosen": 0.031186995786779068,
+      "rewards/margins": 0.02976833888128692,
+      "rewards/rejected": 0.0014186569054921469,
+      "stability/repetition_rate_mean": 0.2408854216337204,
+      "stability/response_length_mean": 35.25,
+      "stability/response_length_std": 31.01957893371582,
+      "stability/response_length_var": 962.2142944335938,
+      "stability/token_entropy_mean": 4.6015625,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.541877269744873,
+      "kl": 0.14850066602230072,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54157408.37647059,
+      "logits/rejected": -49500801.70666666,
+      "logps/chosen": -360.8025505514706,
+      "logps/rejected": -191.32903645833332,
+      "loss": 0.5003,
+      "loss/base_kto": 0.4973774552345276,
+      "metrics/advantage_var": 0.16561715304851532,
+      "regularization/advantage_range": 1.0279508829116821,
+      "regularization/tv": 0.003083852818235755,
+      "rewards/chosen": 0.01867367520051844,
+      "rewards/margins": 0.02101714923101313,
+      "rewards/rejected": -0.00234347403049469,
+      "stability/repetition_rate_mean": 0.3210846781730652,
+      "stability/response_length_mean": 61.375,
+      "stability/response_length_std": 7.424621105194092,
+      "stability/response_length_var": 55.125,
+      "stability/token_entropy_mean": 4.5986328125,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.5596888065338135,
+      "kl": 0.15647390484809875,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55398857.14285714,
+      "logits/rejected": -49154202.94736842,
+      "logps/chosen": -416.60086495535717,
+      "logps/rejected": -186.86446340460526,
+      "loss": 0.5004,
+      "loss/base_kto": 0.49669837951660156,
+      "metrics/advantage_var": 0.21187877655029297,
+      "regularization/advantage_range": 1.114825963973999,
+      "regularization/tv": 0.003344477852806449,
+      "rewards/chosen": 0.02944736821310861,
+      "rewards/margins": 0.026160127742398055,
+      "rewards/rejected": 0.0032872404707105537,
+      "stability/repetition_rate_mean": 0.31046873331069946,
+      "stability/response_length_mean": 32.5,
+      "stability/response_length_std": 30.528675079345703,
+      "stability/response_length_var": 932.0,
+      "stability/token_entropy_mean": 4.3857421875,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.6522524952888489,
+      "kl": 0.14636269211769104,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54598391.172413796,
+      "logits/rejected": -50009463.23287671,
+      "logps/chosen": -370.67030352011494,
+      "logps/rejected": -212.60320794092465,
+      "loss": 0.5,
+      "loss/base_kto": 0.49680137634277344,
+      "metrics/advantage_var": 0.15778213739395142,
+      "regularization/advantage_range": 1.0048089027404785,
+      "regularization/tv": 0.003014426678419113,
+      "rewards/chosen": 0.02464700841355598,
+      "rewards/margins": 0.02575588618262774,
+      "rewards/rejected": -0.0011088777690717619,
+      "stability/repetition_rate_mean": 0.20468750596046448,
+      "stability/response_length_mean": 12.625,
+      "stability/response_length_std": 21.460512161254883,
+      "stability/response_length_var": 460.5535583496094,
+      "stability/token_entropy_mean": 4.828125,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.5037974715232849,
+      "kl": 0.19872891902923584,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59318084.165680476,
+      "logits/rejected": -52844435.49668874,
+      "logps/chosen": -441.0094767011834,
+      "logps/rejected": -191.21860771937085,
+      "loss": 0.4998,
+      "loss/base_kto": 0.49641498923301697,
+      "metrics/advantage_var": 0.20756979286670685,
+      "regularization/advantage_range": 1.1968477964401245,
+      "regularization/tv": 0.0035905432887375355,
+      "rewards/chosen": 0.03323545399502184,
+      "rewards/margins": 0.02931231930177504,
+      "rewards/rejected": 0.003923134693246804,
+      "stability/repetition_rate_mean": 0.20866228640079498,
+      "stability/response_length_mean": 28.25,
+      "stability/response_length_std": 30.844078063964844,
+      "stability/response_length_var": 951.3571166992188,
+      "stability/token_entropy_mean": 3.31689453125,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.5411744117736816,
+      "kl": 0.2043342888355255,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56856687.03614458,
+      "logits/rejected": -48328421.402597405,
+      "logps/chosen": -356.0929028614458,
+      "logps/rejected": -194.7725497159091,
+      "loss": 0.4997,
+      "loss/base_kto": 0.49607935547828674,
+      "metrics/advantage_var": 0.22021155059337616,
+      "regularization/advantage_range": 1.1863764524459839,
+      "regularization/tv": 0.0035591290798038244,
+      "rewards/chosen": 0.034427757722785675,
+      "rewards/margins": 0.030930112270360598,
+      "rewards/rejected": 0.003497645452425077,
+      "stability/repetition_rate_mean": 0.2897879481315613,
+      "stability/response_length_mean": 32.25,
+      "stability/response_length_std": 27.932315826416016,
+      "stability/response_length_var": 780.2142944335938,
+      "stability/token_entropy_mean": 4.69921875,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.5480484366416931,
+      "kl": 0.2203982174396515,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52875565.54601227,
+      "logits/rejected": -48260839.541401275,
+      "logps/chosen": -375.78383819018404,
+      "logps/rejected": -211.20056230095543,
+      "loss": 0.4997,
+      "loss/base_kto": 0.4966592788696289,
+      "metrics/advantage_var": 0.2269420623779297,
+      "regularization/advantage_range": 1.2233505249023438,
+      "regularization/tv": 0.003670051693916321,
+      "rewards/chosen": 0.0362014126923918,
+      "rewards/margins": 0.026436805906769428,
+      "rewards/rejected": 0.009764606785622371,
+      "stability/repetition_rate_mean": 0.265625,
+      "stability/response_length_mean": 40.375,
+      "stability/response_length_std": 32.61874008178711,
+      "stability/response_length_var": 1063.982177734375,
+      "stability/token_entropy_mean": 3.080078125,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.5584241151809692,
+      "kl": 0.23888865113258362,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54872038.71604938,
+      "logits/rejected": -49803879.69620253,
+      "logps/chosen": -311.7043065200617,
+      "logps/rejected": -196.933222903481,
+      "loss": 0.5001,
+      "loss/base_kto": 0.49761316180229187,
+      "metrics/advantage_var": 0.19539952278137207,
+      "regularization/advantage_range": 1.1465425491333008,
+      "regularization/tv": 0.0034396275877952576,
+      "rewards/chosen": 0.02970717277055905,
+      "rewards/margins": 0.01911422172027149,
+      "rewards/rejected": 0.01059295105028756,
+      "stability/repetition_rate_mean": 0.20561079680919647,
+      "stability/response_length_mean": 26.25,
+      "stability/response_length_std": 31.45404815673828,
+      "stability/response_length_var": 989.3571166992188,
+      "stability/token_entropy_mean": 3.796875,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.5378943085670471,
+      "kl": 0.2414882481098175,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53337394.58682635,
+      "logits/rejected": -49988594.614379086,
+      "logps/chosen": -391.9390203967066,
+      "logps/rejected": -211.10026041666666,
+      "loss": 0.5,
+      "loss/base_kto": 0.4952877163887024,
+      "metrics/advantage_var": 0.31047913432121277,
+      "regularization/advantage_range": 1.421489953994751,
+      "regularization/tv": 0.004264469724148512,
+      "rewards/chosen": 0.04663035112940623,
+      "rewards/margins": 0.036798228382770345,
+      "rewards/rejected": 0.009832122746635885,
+      "stability/repetition_rate_mean": 0.1086309552192688,
+      "stability/response_length_mean": 11.375,
+      "stability/response_length_std": 21.738296508789062,
+      "stability/response_length_var": 472.5535583496094,
+      "stability/token_entropy_mean": 5.33984375,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.546999454498291,
+      "kl": 0.22969751060009003,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60425077.26451613,
+      "logits/rejected": -54190259.97575758,
+      "logps/chosen": -392.32464717741937,
+      "logps/rejected": -197.82759232954547,
+      "loss": 0.4997,
+      "loss/base_kto": 0.4953397214412689,
+      "metrics/advantage_var": 0.2615717351436615,
+      "regularization/advantage_range": 1.2991220951080322,
+      "regularization/tv": 0.0038973658811300993,
+      "rewards/chosen": 0.04602757423154769,
+      "rewards/margins": 0.03685314927982095,
+      "rewards/rejected": 0.00917442495172674,
+      "stability/repetition_rate_mean": 0.4031808078289032,
+      "stability/response_length_mean": 41.0,
+      "stability/response_length_std": 31.807456970214844,
+      "stability/response_length_var": 1011.7142944335938,
+      "stability/token_entropy_mean": 4.40234375,
+      "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_tv_rho0.003/checkpoint-196/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:50ed4a0ba877bcff5b82e539a8621186c048c171ba078ef5bb91bed25f698cba
+size 8017

mistral7b_kto_tv_rho0.003/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_tv_rho0.003/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_tv_rho0.003/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": 3.3056640625, "stability/repetition_rate_mean": 0.3531753420829773, "step": 10}
+{"stability/response_length_mean": 28.5, "stability/response_length_std": 30.10457992553711, "stability/response_length_var": 906.2857055664062, "stability/token_entropy_mean": 3.7392578125, "stability/repetition_rate_mean": 0.2763020694255829, "step": 15}
+{"stability/response_length_mean": 44.0, "stability/response_length_std": 28.0611572265625, "stability/response_length_var": 787.4285888671875, "stability/token_entropy_mean": 4.4296875, "stability/repetition_rate_mean": 0.2194320410490036, "step": 20}
+{"stability/response_length_mean": 43.375, "stability/response_length_std": 28.54538917541504, "stability/response_length_var": 814.8392944335938, "stability/token_entropy_mean": 4.806640625, "stability/repetition_rate_mean": 0.16044004261493683, "step": 25}
+{"stability/response_length_mean": 45.875, "stability/response_length_std": 28.246049880981445, "stability/response_length_var": 797.8392944335938, "stability/token_entropy_mean": 4.57421875, "stability/repetition_rate_mean": 0.17223010957241058, "step": 30}
+{"stability/response_length_mean": 56.875, "stability/response_length_std": 20.152544021606445, "stability/response_length_var": 406.125, "stability/token_entropy_mean": 5.421875, "stability/repetition_rate_mean": 0.3147321343421936, "step": 35}
+{"stability/response_length_mean": 33.125, "stability/response_length_std": 33.01704025268555, "stability/response_length_var": 1090.125, "stability/token_entropy_mean": 4.8896484375, "stability/repetition_rate_mean": 0.13671875, "step": 40}
+{"stability/response_length_mean": 29.25, "stability/response_length_std": 29.51875877380371, "stability/response_length_var": 871.3571166992188, "stability/token_entropy_mean": 4.70703125, "stability/repetition_rate_mean": 0.0994017943739891, "step": 45}
+{"stability/response_length_mean": 55.375, "stability/response_length_std": 16.361431121826172, "stability/response_length_var": 267.6964416503906, "stability/token_entropy_mean": 4.068359375, "stability/repetition_rate_mean": 0.3288106620311737, "step": 50}
+{"stability/response_length_mean": 24.5, "stability/response_length_std": 32.710853576660156, "stability/response_length_var": 1070.0, "stability/token_entropy_mean": 4.388671875, "stability/repetition_rate_mean": 0.111328125, "step": 55}
+{"stability/response_length_mean": 52.625, "stability/response_length_std": 21.75800323486328, "stability/response_length_var": 473.41070556640625, "stability/token_entropy_mean": 5.29296875, "stability/repetition_rate_mean": 0.21596631407737732, "step": 60}
+{"stability/response_length_mean": 20.625, "stability/response_length_std": 27.176342010498047, "stability/response_length_var": 738.5535888671875, "stability/token_entropy_mean": 5.544921875, "stability/repetition_rate_mean": 0.1857219785451889, "step": 65}
+{"stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.4404296875, "stability/repetition_rate_mean": 0.11328125, "step": 70}
+{"stability/response_length_mean": 24.25, "stability/response_length_std": 32.92306900024414, "stability/response_length_var": 1083.9285888671875, "stability/token_entropy_mean": 1.883056640625, "stability/repetition_rate_mean": 0.08203125, "step": 75}
+{"stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.23828125, "stability/repetition_rate_mean": 0.19088542461395264, "step": 80}
+{"stability/response_length_mean": 28.0, "stability/response_length_std": 31.181495666503906, "stability/response_length_var": 972.2857055664062, "stability/token_entropy_mean": 4.96484375, "stability/repetition_rate_mean": 0.1369977593421936, "step": 85}
+{"stability/response_length_mean": 35.125, "stability/response_length_std": 31.638301849365234, "stability/response_length_var": 1000.9821166992188, "stability/token_entropy_mean": 2.94384765625, "stability/repetition_rate_mean": 0.2582859992980957, "step": 90}
+{"stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.736328125, "stability/repetition_rate_mean": 0.19374233484268188, "step": 95}
+{"stability/response_length_mean": 41.25, "stability/response_length_std": 31.476749420166016, "stability/response_length_var": 990.7857055664062, "stability/token_entropy_mean": 4.107421875, "stability/repetition_rate_mean": 0.2421875, "step": 100}
+{"stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.388671875, "stability/repetition_rate_mean": 0.1916852593421936, "step": 105}
+{"stability/response_length_mean": 31.125, "stability/response_length_std": 28.412460327148438, "stability/response_length_var": 807.2678833007812, "stability/token_entropy_mean": 4.0673828125, "stability/repetition_rate_mean": 0.3186553120613098, "step": 110}
+{"stability/response_length_mean": 41.375, "stability/response_length_std": 31.25442886352539, "stability/response_length_var": 976.8392944335938, "stability/token_entropy_mean": 4.935546875, "stability/repetition_rate_mean": 0.2916666865348816, "step": 115}
+{"stability/response_length_mean": 27.875, "stability/response_length_std": 30.31943130493164, "stability/response_length_var": 919.2678833007812, "stability/token_entropy_mean": 4.623046875, "stability/repetition_rate_mean": 0.1664772778749466, "step": 120}
+{"stability/response_length_mean": 15.25, "stability/response_length_std": 24.93276596069336, "stability/response_length_var": 621.6428833007812, "stability/token_entropy_mean": 3.6328125, "stability/repetition_rate_mean": 0.16328126192092896, "step": 125}
+{"stability/response_length_mean": 38.0, "stability/response_length_std": 29.13515281677246, "stability/response_length_var": 848.8571166992188, "stability/token_entropy_mean": 3.99609375, "stability/repetition_rate_mean": 0.19791193306446075, "step": 130}
+{"stability/response_length_mean": 29.25, "stability/response_length_std": 29.936838150024414, "stability/response_length_var": 896.2142944335938, "stability/token_entropy_mean": 4.07421875, "stability/repetition_rate_mean": 0.2230902761220932, "step": 135}
+{"stability/response_length_mean": 56.25, "stability/response_length_std": 21.920310974121094, "stability/response_length_var": 480.5, "stability/token_entropy_mean": 4.7451171875, "stability/repetition_rate_mean": 0.35546875, "step": 140}
+{"stability/response_length_mean": 22.625, "stability/response_length_std": 26.359804153442383, "stability/response_length_var": 694.8392944335938, "stability/token_entropy_mean": 5.703125, "stability/repetition_rate_mean": 0.20039846003055573, "step": 145}
+{"stability/response_length_mean": 35.25, "stability/response_length_std": 31.01957893371582, "stability/response_length_var": 962.2142944335938, "stability/token_entropy_mean": 4.6015625, "stability/repetition_rate_mean": 0.2408854216337204, "step": 150}
+{"stability/response_length_mean": 61.375, "stability/response_length_std": 7.424621105194092, "stability/response_length_var": 55.125, "stability/token_entropy_mean": 4.5986328125, "stability/repetition_rate_mean": 0.3210846781730652, "step": 155}
+{"stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.3857421875, "stability/repetition_rate_mean": 0.31046873331069946, "step": 160}
+{"stability/response_length_mean": 12.625, "stability/response_length_std": 21.460512161254883, "stability/response_length_var": 460.5535583496094, "stability/token_entropy_mean": 4.828125, "stability/repetition_rate_mean": 0.20468750596046448, "step": 165}
+{"stability/response_length_mean": 28.25, "stability/response_length_std": 30.844078063964844, "stability/response_length_var": 951.3571166992188, "stability/token_entropy_mean": 3.31689453125, "stability/repetition_rate_mean": 0.20866228640079498, "step": 170}
+{"stability/response_length_mean": 32.25, "stability/response_length_std": 27.932315826416016, "stability/response_length_var": 780.2142944335938, "stability/token_entropy_mean": 4.69921875, "stability/repetition_rate_mean": 0.2897879481315613, "step": 175}
+{"stability/response_length_mean": 40.375, "stability/response_length_std": 32.61874008178711, "stability/response_length_var": 1063.982177734375, "stability/token_entropy_mean": 3.080078125, "stability/repetition_rate_mean": 0.265625, "step": 180}
+{"stability/response_length_mean": 26.25, "stability/response_length_std": 31.45404815673828, "stability/response_length_var": 989.3571166992188, "stability/token_entropy_mean": 3.796875, "stability/repetition_rate_mean": 0.20561079680919647, "step": 185}
+{"stability/response_length_mean": 11.375, "stability/response_length_std": 21.738296508789062, "stability/response_length_var": 472.5535583496094, "stability/token_entropy_mean": 5.33984375, "stability/repetition_rate_mean": 0.1086309552192688, "step": 190}
+{"stability/response_length_mean": 41.0, "stability/response_length_std": 31.807456970214844, "stability/response_length_var": 1011.7142944335938, "stability/token_entropy_mean": 4.40234375, "stability/repetition_rate_mean": 0.4031808078289032, "step": 195}

mistral7b_kto_tv_rho0.003/tokenizer.model

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

mistral7b_kto_tv_rho0.003/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_tv_rho0.003/train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.5009068980508921,
+    "train_runtime": 3152.3184,
+    "train_samples_per_second": 31.723,
+    "train_steps_per_second": 0.062
+}

mistral7b_kto_tv_rho0.003/trainer_log.jsonl

@@ -0,0 +1,40 @@
+{"current_steps": 5, "total_steps": 196, "loss": 0.5011, "lr": 4e-08, "epoch": 0.025591810620601407, "percentage": 2.55, "elapsed_time": "0:01:17", "remaining_time": "0:49:38", "rewards/chosen": 0.002090120393978922, "logps/chosen": -439.0247224506579, "logits/chosen": -59711191.578947365, "rewards/rejected": 0.00011249401030086336, "logps/rejected": -198.40827287946428, "logits/rejected": -50416146.28571428, "rewards/margins": 0.0019776263836780586, "regularization/tv": 0.001391954137943685, "regularization/advantage_range": 0.46398472785949707, "kl": 0.009316612966358662}
+{"current_steps": 10, "total_steps": 196, "loss": 0.5019, "lr": 9e-08, "epoch": 0.05118362124120281, "percentage": 5.1, "elapsed_time": "0:02:38", "remaining_time": "0:49:12", "rewards/chosen": 0.0032109109225211207, "logps/chosen": -429.9064021915584, "logits/chosen": -59301914.597402595, "rewards/rejected": 0.0008140276354479503, "logps/rejected": -192.8597985692771, "logits/rejected": -50024296.86746988, "rewards/margins": 0.0023968832870731706, "regularization/tv": 0.0017509151948615909, "regularization/advantage_range": 0.5836383700370789, "kl": 0.01776806451380253, "stability/response_length_mean": 46.125, "stability/response_length_std": 27.231481552124023, "stability/response_length_var": 741.5535888671875, "stability/token_entropy_mean": 3.3056640625, "stability/repetition_rate_mean": 0.3531753420829773}
+{"current_steps": 15, "total_steps": 196, "loss": 0.5018, "lr": 1.4e-07, "epoch": 0.07677543186180422, "percentage": 7.65, "elapsed_time": "0:03:58", "remaining_time": "0:47:58", "rewards/chosen": 0.002984200824390758, "logps/chosen": -404.1009943181818, "logits/chosen": -56231693.96363636, "rewards/rejected": 0.0026088099325856855, "logps/rejected": -200.66796875, "logits/rejected": -51925735.22580645, "rewards/margins": 0.0003753908918050727, "regularization/tv": 0.001579046598635614, "regularization/advantage_range": 0.526348888874054, "kl": 0.01534965168684721, "stability/response_length_mean": 28.5, "stability/response_length_std": 30.10457992553711, "stability/response_length_var": 906.2857055664062, "stability/token_entropy_mean": 3.7392578125, "stability/repetition_rate_mean": 0.2763020694255829}
+{"current_steps": 20, "total_steps": 196, "loss": 0.5019, "lr": 1.8999999999999998e-07, "epoch": 0.10236724248240563, "percentage": 10.2, "elapsed_time": "0:05:19", "remaining_time": "0:46:54", "rewards/chosen": 0.0002750195562839508, "logps/chosen": -343.6939697265625, "logits/chosen": -51672825.6, "rewards/rejected": 0.004302208125591278, "logps/rejected": -225.6203369140625, "logits/rejected": -48064236.8, "rewards/margins": -0.004027188569307327, "regularization/tv": 0.0015871055657044053, "regularization/advantage_range": 0.529035210609436, "kl": 0.021325770765542984, "stability/response_length_mean": 44.0, "stability/response_length_std": 28.0611572265625, "stability/response_length_var": 787.4285888671875, "stability/token_entropy_mean": 4.4296875, "stability/repetition_rate_mean": 0.2194320410490036}
+{"current_steps": 25, "total_steps": 196, "loss": 0.5019, "lr": 2.4e-07, "epoch": 0.12795905310300704, "percentage": 12.76, "elapsed_time": "0:06:39", "remaining_time": "0:45:34", "rewards/chosen": 0.002574328924047536, "logps/chosen": -378.4657058189655, "logits/chosen": -56468112.77241379, "rewards/rejected": 0.0008958792686462403, "logps/rejected": -202.78388392857144, "logits/rejected": -52019515.977142856, "rewards/margins": 0.0016784496554012956, "regularization/tv": 0.0015671454602852464, "regularization/advantage_range": 0.5223817825317383, "kl": 0.012540136463940144, "stability/response_length_mean": 43.375, "stability/response_length_std": 28.54538917541504, "stability/response_length_var": 814.8392944335938, "stability/token_entropy_mean": 4.806640625, "stability/repetition_rate_mean": 0.16044004261493683}
+{"current_steps": 30, "total_steps": 196, "loss": 0.502, "lr": 2.9e-07, "epoch": 0.15355086372360843, "percentage": 15.31, "elapsed_time": "0:07:59", "remaining_time": "0:44:14", "rewards/chosen": 0.0021430586008417524, "logps/chosen": -468.5598259228188, "logits/chosen": -57719072.644295305, "rewards/rejected": 0.0017368478036066245, "logps/rejected": -198.2305144371345, "logits/rejected": -51523212.7251462, "rewards/margins": 0.0004062107972351279, "regularization/tv": 0.0016285822493955493, "regularization/advantage_range": 0.542860746383667, "kl": 0.01668705977499485, "stability/response_length_mean": 45.875, "stability/response_length_std": 28.246049880981445, "stability/response_length_var": 797.8392944335938, "stability/token_entropy_mean": 4.57421875, "stability/repetition_rate_mean": 0.17223010957241058}
+{"current_steps": 35, "total_steps": 196, "loss": 0.5017, "lr": 3.4000000000000003e-07, "epoch": 0.17914267434420986, "percentage": 17.86, "elapsed_time": "0:09:21", "remaining_time": "0:43:00", "rewards/chosen": 0.003989132899272291, "logps/chosen": -385.8966327136076, "logits/chosen": -57647505.82278481, "rewards/rejected": 0.0008958886251037503, "logps/rejected": -195.1175250771605, "logits/rejected": -53915256.09876543, "rewards/margins": 0.0030932442741685405, "regularization/tv": 0.001776991761289537, "regularization/advantage_range": 0.5923306345939636, "kl": 0.018097203224897385, "stability/response_length_mean": 56.875, "stability/response_length_std": 20.152544021606445, "stability/response_length_var": 406.125, "stability/token_entropy_mean": 5.421875, "stability/repetition_rate_mean": 0.3147321343421936}
+{"current_steps": 40, "total_steps": 196, "loss": 0.5019, "lr": 3.8999999999999997e-07, "epoch": 0.20473448496481125, "percentage": 20.41, "elapsed_time": "0:10:41", "remaining_time": "0:41:42", "rewards/chosen": 0.00015102624893188478, "logps/chosen": -421.901123046875, "logits/chosen": -56631635.2, "rewards/rejected": 0.0038557790219783785, "logps/rejected": -219.2706787109375, "logits/rejected": -50227708.8, "rewards/margins": -0.0037047527730464935, "regularization/tv": 0.0018836684757843614, "regularization/advantage_range": 0.6278894543647766, "kl": 0.028051286935806274, "stability/response_length_mean": 33.125, "stability/response_length_std": 33.01704025268555, "stability/response_length_var": 1090.125, "stability/token_entropy_mean": 4.8896484375, "stability/repetition_rate_mean": 0.13671875}
+{"current_steps": 45, "total_steps": 196, "loss": 0.5015, "lr": 4.3999999999999997e-07, "epoch": 0.23032629558541268, "percentage": 22.96, "elapsed_time": "0:12:02", "remaining_time": "0:40:23", "rewards/chosen": 0.006769456319360925, "logps/chosen": -352.2607487416107, "logits/chosen": -52712994.36241611, "rewards/rejected": 0.004627710197404114, "logps/rejected": -243.8278965643275, "logits/rejected": -50780387.55555555, "rewards/margins": 0.002141746121956811, "regularization/tv": 0.001985067967325449, "regularization/advantage_range": 0.661689281463623, "kl": 0.03983687236905098, "stability/response_length_mean": 29.25, "stability/response_length_std": 29.51875877380371, "stability/response_length_var": 871.3571166992188, "stability/token_entropy_mean": 4.70703125, "stability/repetition_rate_mean": 0.0994017943739891}
+{"current_steps": 50, "total_steps": 196, "loss": 0.5017, "lr": 4.9e-07, "epoch": 0.2559181062060141, "percentage": 25.51, "elapsed_time": "0:13:22", "remaining_time": "0:39:02", "rewards/chosen": 0.005363105927537751, "logps/chosen": -375.1455012583893, "logits/chosen": -54021566.7114094, "rewards/rejected": 0.006422032389724464, "logps/rejected": -180.17639802631578, "logits/rejected": -48286264.88888889, "rewards/margins": -0.0010589264621867125, "regularization/tv": 0.0016679477412253618, "regularization/advantage_range": 0.5559825301170349, "kl": 0.03668095916509628, "stability/response_length_mean": 55.375, "stability/response_length_std": 16.361431121826172, "stability/response_length_var": 267.6964416503906, "stability/token_entropy_mean": 4.068359375, "stability/repetition_rate_mean": 0.3288106620311737}
+{"current_steps": 55, "total_steps": 196, "loss": 0.5017, "lr": 5.4e-07, "epoch": 0.28150991682661547, "percentage": 28.06, "elapsed_time": "0:14:42", "remaining_time": "0:37:41", "rewards/chosen": 0.0014025861432093271, "logps/chosen": -344.49801048136646, "logits/chosen": -56951623.55279503, "rewards/rejected": 0.005317174413669034, "logps/rejected": -230.3948506289308, "logits/rejected": -49562733.48427673, "rewards/margins": -0.003914588270459707, "regularization/tv": 0.0017599230632185936, "regularization/advantage_range": 0.5866410136222839, "kl": 0.043028466403484344, "stability/response_length_mean": 24.5, "stability/response_length_std": 32.710853576660156, "stability/response_length_var": 1070.0, "stability/token_entropy_mean": 4.388671875, "stability/repetition_rate_mean": 0.111328125}
+{"current_steps": 60, "total_steps": 196, "loss": 0.5016, "lr": 5.9e-07, "epoch": 0.30710172744721687, "percentage": 30.61, "elapsed_time": "0:16:03", "remaining_time": "0:36:24", "rewards/chosen": 0.008838949781475645, "logps/chosen": -359.6323863636364, "logits/chosen": -52672884.36363637, "rewards/rejected": 0.0051677799993945705, "logps/rejected": -178.17522681451612, "logits/rejected": -47833497.6, "rewards/margins": 0.0036711697820810745, "regularization/tv": 0.0017956882948055863, "regularization/advantage_range": 0.5985627770423889, "kl": 0.03918389976024628, "stability/response_length_mean": 52.625, "stability/response_length_std": 21.75800323486328, "stability/response_length_var": 473.41070556640625, "stability/token_entropy_mean": 5.29296875, "stability/repetition_rate_mean": 0.21596631407737732}
+{"current_steps": 65, "total_steps": 196, "loss": 0.5013, "lr": 6.4e-07, "epoch": 0.3326935380678183, "percentage": 33.16, "elapsed_time": "0:17:24", "remaining_time": "0:35:04", "rewards/chosen": 0.01091364276024603, "logps/chosen": -381.8376512096774, "logits/chosen": -57567218.787096776, "rewards/rejected": 0.004210157827897506, "logps/rejected": -183.54024621212122, "logits/rejected": -47273661.28484848, "rewards/margins": 0.006703484932348525, "regularization/tv": 0.001847176463343203, "regularization/advantage_range": 0.6157255172729492, "kl": 0.04982197284698486, "stability/response_length_mean": 20.625, "stability/response_length_std": 27.176342010498047, "stability/response_length_var": 738.5535888671875, "stability/token_entropy_mean": 5.544921875, "stability/repetition_rate_mean": 0.1857219785451889}
+{"current_steps": 70, "total_steps": 196, "loss": 0.5011, "lr": 6.9e-07, "epoch": 0.3582853486884197, "percentage": 35.71, "elapsed_time": "0:18:44", "remaining_time": "0:33:44", "rewards/chosen": 0.008328685452861171, "logps/chosen": -379.0451612903226, "logits/chosen": -52787001.80645161, "rewards/rejected": 0.0027332269784175987, "logps/rejected": -220.39483901515152, "logits/rejected": -49036244.557575755, "rewards/margins": 0.005595458474443572, "regularization/tv": 0.001879148418083787, "regularization/advantage_range": 0.6263827681541443, "kl": 0.06117330864071846, "stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.4404296875, "stability/repetition_rate_mean": 0.11328125}
+{"current_steps": 75, "total_steps": 196, "loss": 0.5015, "lr": 7.4e-07, "epoch": 0.3838771593090211, "percentage": 38.27, "elapsed_time": "0:20:05", "remaining_time": "0:32:24", "rewards/chosen": 0.009913306783985448, "logps/chosen": -314.17166385135135, "logits/chosen": -49505947.675675675, "rewards/rejected": 0.003935314888177916, "logps/rejected": -228.84547601744185, "logits/rejected": -47897421.39534884, "rewards/margins": 0.005977991895807532, "regularization/tv": 0.0021153318230062723, "regularization/advantage_range": 0.7051105499267578, "kl": 0.07038548588752747, "stability/response_length_mean": 24.25, "stability/response_length_std": 32.92306900024414, "stability/response_length_var": 1083.9285888671875, "stability/token_entropy_mean": 1.883056640625, "stability/repetition_rate_mean": 0.08203125}
+{"current_steps": 80, "total_steps": 196, "loss": 0.5014, "lr": 7.9e-07, "epoch": 0.4094689699296225, "percentage": 40.82, "elapsed_time": "0:21:26", "remaining_time": "0:31:05", "rewards/chosen": 0.010046138668691876, "logps/chosen": -403.9296616307947, "logits/chosen": -58622535.205298014, "rewards/rejected": 0.003927631138344488, "logps/rejected": -201.01654955621302, "logits/rejected": -53446995.313609466, "rewards/margins": 0.006118507530347388, "regularization/tv": 0.0017893938347697258, "regularization/advantage_range": 0.5964646935462952, "kl": 0.08504229038953781, "stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.23828125, "stability/repetition_rate_mean": 0.19088542461395264}
+{"current_steps": 85, "total_steps": 196, "loss": 0.5012, "lr": 8.399999999999999e-07, "epoch": 0.4350607805502239, "percentage": 43.37, "elapsed_time": "0:22:46", "remaining_time": "0:29:44", "rewards/chosen": 0.012145080104950936, "logps/chosen": -358.2110635080645, "logits/chosen": -51481600.0, "rewards/rejected": 0.004309365604863022, "logps/rejected": -230.3817234848485, "logits/rejected": -47146399.806060605, "rewards/margins": 0.007835714500087913, "regularization/tv": 0.0020550561603158712, "regularization/advantage_range": 0.68501877784729, "kl": 0.0836004987359047, "stability/response_length_mean": 28.0, "stability/response_length_std": 31.181495666503906, "stability/response_length_var": 972.2857055664062, "stability/token_entropy_mean": 4.96484375, "stability/repetition_rate_mean": 0.1369977593421936}
+{"current_steps": 90, "total_steps": 196, "loss": 0.5008, "lr": 8.9e-07, "epoch": 0.46065259117082535, "percentage": 45.92, "elapsed_time": "0:24:07", "remaining_time": "0:28:25", "rewards/chosen": 0.012280461115714831, "logps/chosen": -380.6665665064103, "logits/chosen": -58432328.20512821, "rewards/rejected": 0.0039842005183057085, "logps/rejected": -187.1869521722561, "logits/rejected": -50573452.487804875, "rewards/margins": 0.008296260597409122, "regularization/tv": 0.001918740919791162, "regularization/advantage_range": 0.639580249786377, "kl": 0.09354694932699203, "stability/response_length_mean": 35.125, "stability/response_length_std": 31.638301849365234, "stability/response_length_var": 1000.9821166992188, "stability/token_entropy_mean": 2.94384765625, "stability/repetition_rate_mean": 0.2582859992980957}
+{"current_steps": 95, "total_steps": 196, "loss": 0.5009, "lr": 9.399999999999999e-07, "epoch": 0.48624440179142675, "percentage": 48.47, "elapsed_time": "0:25:27", "remaining_time": "0:27:04", "rewards/chosen": 0.010773800596406188, "logps/chosen": -314.8856309335443, "logits/chosen": -50400852.25316456, "rewards/rejected": -3.7510453918833793e-06, "logps/rejected": -207.14375964506172, "logits/rejected": -48991902.02469136, "rewards/margins": 0.010777551641798071, "regularization/tv": 0.0018720850348472595, "regularization/advantage_range": 0.6240283250808716, "kl": 0.08265768736600876, "stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.736328125, "stability/repetition_rate_mean": 0.19374233484268188}
+{"current_steps": 100, "total_steps": 196, "loss": 0.501, "lr": 9.9e-07, "epoch": 0.5118362124120281, "percentage": 51.02, "elapsed_time": "0:26:48", "remaining_time": "0:25:43", "rewards/chosen": 0.01675953736176362, "logps/chosen": -359.7025971283784, "logits/chosen": -54724199.783783786, "rewards/rejected": 0.006541241046994231, "logps/rejected": -216.51346747819767, "logits/rejected": -50932414.511627905, "rewards/margins": 0.01021829631476939, "regularization/tv": 0.002038272563368082, "regularization/advantage_range": 0.6794241070747375, "kl": 0.08847284317016602, "stability/response_length_mean": 41.25, "stability/response_length_std": 31.476749420166016, "stability/response_length_var": 990.7857055664062, "stability/token_entropy_mean": 4.107421875, "stability/repetition_rate_mean": 0.2421875}
+{"current_steps": 105, "total_steps": 196, "loss": 0.5009, "lr": 9.999815554351224e-07, "epoch": 0.5374280230326296, "percentage": 53.57, "elapsed_time": "0:28:08", "remaining_time": "0:24:23", "rewards/chosen": 0.015377698174442154, "logps/chosen": -399.74868222891564, "logits/chosen": -59393332.43373494, "rewards/rejected": 0.004508952041725059, "logps/rejected": -209.9587307224026, "logits/rejected": -53833864.31168831, "rewards/margins": 0.010868746132717095, "regularization/tv": 0.0023293232079595327, "regularization/advantage_range": 0.7764409184455872, "kl": 0.09840010851621628, "stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.388671875, "stability/repetition_rate_mean": 0.1916852593421936}
+{"current_steps": 110, "total_steps": 196, "loss": 0.5008, "lr": 9.999066267225444e-07, "epoch": 0.5630198336532309, "percentage": 56.12, "elapsed_time": "0:29:30", "remaining_time": "0:23:04", "rewards/chosen": 0.01228764545486634, "logps/chosen": -378.44039439006025, "logits/chosen": -57405316.62650602, "rewards/rejected": 0.003466525635161957, "logps/rejected": -206.05346996753246, "logits/rejected": -53642688.83116883, "rewards/margins": 0.008821119819704383, "regularization/tv": 0.002332056639716029, "regularization/advantage_range": 0.7773522734642029, "kl": 0.10348369181156158, "stability/response_length_mean": 31.125, "stability/response_length_std": 28.412460327148438, "stability/response_length_var": 807.2678833007812, "stability/token_entropy_mean": 4.0673828125, "stability/repetition_rate_mean": 0.3186553120613098}
+{"current_steps": 115, "total_steps": 196, "loss": 0.5008, "lr": 9.997740697079592e-07, "epoch": 0.5886116442738324, "percentage": 58.67, "elapsed_time": "0:30:51", "remaining_time": "0:21:44", "rewards/chosen": 0.019377651093881343, "logps/chosen": -364.39875890031647, "logits/chosen": -60635421.164556965, "rewards/rejected": 0.007268024815453423, "logps/rejected": -201.12389081790124, "logits/rejected": -53458703.802469134, "rewards/margins": 0.01210962627842792, "regularization/tv": 0.0020915332715958357, "regularization/advantage_range": 0.6971777081489563, "kl": 0.11748156696557999, "stability/response_length_mean": 41.375, "stability/response_length_std": 31.25442886352539, "stability/response_length_var": 976.8392944335938, "stability/token_entropy_mean": 4.935546875, "stability/repetition_rate_mean": 0.2916666865348816}
+{"current_steps": 120, "total_steps": 196, "loss": 0.5004, "lr": 9.995838996722914e-07, "epoch": 0.6142034548944337, "percentage": 61.22, "elapsed_time": "0:32:11", "remaining_time": "0:20:22", "rewards/chosen": 0.018445407643037683, "logps/chosen": -384.06045751633985, "logits/chosen": -57774535.11111111, "rewards/rejected": 0.0042648472471865356, "logps/rejected": -212.17372286676647, "logits/rejected": -54427586.68263473, "rewards/margins": 0.014180560395851147, "regularization/tv": 0.0024572184775024652, "regularization/advantage_range": 0.8190727233886719, "kl": 0.1038687452673912, "stability/response_length_mean": 27.875, "stability/response_length_std": 30.31943130493164, "stability/response_length_var": 919.2678833007812, "stability/token_entropy_mean": 4.623046875, "stability/repetition_rate_mean": 0.1664772778749466}
+{"current_steps": 125, "total_steps": 196, "loss": 0.5008, "lr": 9.993361385379874e-07, "epoch": 0.6397952655150352, "percentage": 63.78, "elapsed_time": "0:33:31", "remaining_time": "0:19:02", "rewards/chosen": 0.01676147114740659, "logps/chosen": -417.2116866438356, "logits/chosen": -60747207.89041096, "rewards/rejected": 0.005725167263513324, "logps/rejected": -216.30994073275863, "logits/rejected": -55728210.390804596, "rewards/margins": 0.011036303883893267, "regularization/tv": 0.002718301024287939, "regularization/advantage_range": 0.9061002731323242, "kl": 0.11783448606729507, "stability/response_length_mean": 15.25, "stability/response_length_std": 24.93276596069336, "stability/response_length_var": 621.6428833007812, "stability/token_entropy_mean": 3.6328125, "stability/repetition_rate_mean": 0.16328126192092896}
+{"current_steps": 130, "total_steps": 196, "loss": 0.5003, "lr": 9.99030814866488e-07, "epoch": 0.6653870761356366, "percentage": 66.33, "elapsed_time": "0:34:51", "remaining_time": "0:17:41", "rewards/chosen": 0.016469930833385837, "logps/chosen": -392.4932459677419, "logits/chosen": -57436979.2, "rewards/rejected": 0.0012412152507088401, "logps/rejected": -183.90649857954546, "logits/rejected": -53890463.806060605, "rewards/margins": 0.015228715582676997, "regularization/tv": 0.002613300224766135, "regularization/advantage_range": 0.8711000680923462, "kl": 0.11419554054737091, "stability/response_length_mean": 38.0, "stability/response_length_std": 29.13515281677246, "stability/response_length_var": 848.8571166992188, "stability/token_entropy_mean": 3.99609375, "stability/repetition_rate_mean": 0.19791193306446075}
+{"current_steps": 135, "total_steps": 196, "loss": 0.5005, "lr": 9.986679638549358e-07, "epoch": 0.690978886756238, "percentage": 68.88, "elapsed_time": "0:36:11", "remaining_time": "0:16:21", "rewards/chosen": 0.022427939198499808, "logps/chosen": -401.59190471625766, "logits/chosen": -59916112.09815951, "rewards/rejected": 0.0021399612639360368, "logps/rejected": -247.84937300955414, "logits/rejected": -55112397.4522293, "rewards/margins": 0.020287977934563772, "regularization/tv": 0.0032469145953655243, "regularization/advantage_range": 1.082304835319519, "kl": 0.12443824112415314, "stability/response_length_mean": 29.25, "stability/response_length_std": 29.936838150024414, "stability/response_length_var": 896.2142944335938, "stability/token_entropy_mean": 4.07421875, "stability/repetition_rate_mean": 0.2230902761220932}
+{"current_steps": 140, "total_steps": 196, "loss": 0.5003, "lr": 9.982476273321175e-07, "epoch": 0.7165706973768394, "percentage": 71.43, "elapsed_time": "0:37:32", "remaining_time": "0:15:00", "rewards/chosen": 0.018839573137687915, "logps/chosen": -367.0013494318182, "logits/chosen": -57166928.67878788, "rewards/rejected": -0.0023438105660100136, "logps/rejected": -190.0579133064516, "logits/rejected": -50186933.67741936, "rewards/margins": 0.02118338370369793, "regularization/tv": 0.0031054005958139896, "regularization/advantage_range": 1.0351334810256958, "kl": 0.10719374567270279, "stability/response_length_mean": 56.25, "stability/response_length_std": 21.920310974121094, "stability/response_length_var": 480.5, "stability/token_entropy_mean": 4.7451171875, "stability/repetition_rate_mean": 0.35546875}
+{"current_steps": 145, "total_steps": 196, "loss": 0.5005, "lr": 9.977698537536417e-07, "epoch": 0.7421625079974408, "percentage": 73.98, "elapsed_time": "0:38:52", "remaining_time": "0:13:40", "rewards/chosen": 0.016965202538363904, "logps/chosen": -330.10217432228916, "logits/chosen": -51238381.4939759, "rewards/rejected": 0.0004589700853669798, "logps/rejected": -199.48333502435065, "logits/rejected": -51789099.22077922, "rewards/margins": 0.016506232452996924, "regularization/tv": 0.0026441742666065693, "regularization/advantage_range": 0.8813913464546204, "kl": 0.1362670212984085, "stability/response_length_mean": 22.625, "stability/response_length_std": 26.359804153442383, "stability/response_length_var": 694.8392944335938, "stability/token_entropy_mean": 5.703125, "stability/repetition_rate_mean": 0.20039846003055573}
+{"current_steps": 150, "total_steps": 196, "loss": 0.5003, "lr": 9.972346981963546e-07, "epoch": 0.7677543186180422, "percentage": 76.53, "elapsed_time": "0:40:13", "remaining_time": "0:12:20", "rewards/chosen": 0.031186995786779068, "logps/chosen": -397.62251838235295, "logits/chosen": -55525086.870588236, "rewards/rejected": 0.0014186569054921469, "logps/rejected": -207.71174479166666, "logits/rejected": -50325459.626666665, "rewards/margins": 0.02976833888128692, "regularization/tv": 0.0032860690262168646, "regularization/advantage_range": 1.0953562259674072, "kl": 0.14728893339633942, "stability/response_length_mean": 35.25, "stability/response_length_std": 31.01957893371582, "stability/response_length_var": 962.2142944335938, "stability/token_entropy_mean": 4.6015625, "stability/repetition_rate_mean": 0.2408854216337204}
+{"current_steps": 155, "total_steps": 196, "loss": 0.5003, "lr": 9.966422223519885e-07, "epoch": 0.7933461292386437, "percentage": 79.08, "elapsed_time": "0:41:34", "remaining_time": "0:10:59", "rewards/chosen": 0.01867367520051844, "logps/chosen": -360.8025505514706, "logits/chosen": -54157408.37647059, "rewards/rejected": -0.00234347403049469, "logps/rejected": -191.32903645833332, "logits/rejected": -49500801.70666666, "rewards/margins": 0.02101714923101313, "regularization/tv": 0.003083852818235755, "regularization/advantage_range": 1.0279508829116821, "kl": 0.14850066602230072, "stability/response_length_mean": 61.375, "stability/response_length_std": 7.424621105194092, "stability/response_length_var": 55.125, "stability/token_entropy_mean": 4.5986328125, "stability/repetition_rate_mean": 0.3210846781730652}
+{"current_steps": 160, "total_steps": 196, "loss": 0.5004, "lr": 9.959924945200523e-07, "epoch": 0.818937939859245, "percentage": 81.63, "elapsed_time": "0:42:55", "remaining_time": "0:09:39", "rewards/chosen": 0.02944736821310861, "logps/chosen": -416.60086495535717, "logits/chosen": -55398857.14285714, "rewards/rejected": 0.0032872404707105537, "logps/rejected": -186.86446340460526, "logits/rejected": -49154202.94736842, "rewards/margins": 0.026160127742398055, "regularization/tv": 0.003344477852806449, "regularization/advantage_range": 1.114825963973999, "kl": 0.15647390484809875, "stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.3857421875, "stability/repetition_rate_mean": 0.31046873331069946}
+{"current_steps": 165, "total_steps": 196, "loss": 0.5, "lr": 9.952855895999568e-07, "epoch": 0.8445297504798465, "percentage": 84.18, "elapsed_time": "0:44:15", "remaining_time": "0:08:18", "rewards/chosen": 0.02464700841355598, "logps/chosen": -370.67030352011494, "logits/chosen": -54598391.172413796, "rewards/rejected": -0.0011088777690717619, "logps/rejected": -212.60320794092465, "logits/rejected": -50009463.23287671, "rewards/margins": 0.02575588618262774, "regularization/tv": 0.003014426678419113, "regularization/advantage_range": 1.0048089027404785, "kl": 0.14636269211769104, "stability/response_length_mean": 12.625, "stability/response_length_std": 21.460512161254883, "stability/response_length_var": 460.5535583496094, "stability/token_entropy_mean": 4.828125, "stability/repetition_rate_mean": 0.20468750596046448}
+{"current_steps": 170, "total_steps": 196, "loss": 0.4998, "lr": 9.9452158908238e-07, "epoch": 0.8701215611004478, "percentage": 86.73, "elapsed_time": "0:45:37", "remaining_time": "0:06:58", "rewards/chosen": 0.03323545399502184, "logps/chosen": -441.0094767011834, "logits/chosen": -59318084.165680476, "rewards/rejected": 0.003923134693246804, "logps/rejected": -191.21860771937085, "logits/rejected": -52844435.49668874, "rewards/margins": 0.02931231930177504, "regularization/tv": 0.0035905432887375355, "regularization/advantage_range": 1.1968477964401245, "kl": 0.19872891902923584, "stability/response_length_mean": 28.25, "stability/response_length_std": 30.844078063964844, "stability/response_length_var": 951.3571166992188, "stability/token_entropy_mean": 3.31689453125, "stability/repetition_rate_mean": 0.20866228640079498}
+{"current_steps": 175, "total_steps": 196, "loss": 0.4997, "lr": 9.937005810398745e-07, "epoch": 0.8957133717210493, "percentage": 89.29, "elapsed_time": "0:46:57", "remaining_time": "0:05:38", "rewards/chosen": 0.034427757722785675, "logps/chosen": -356.0929028614458, "logits/chosen": -56856687.03614458, "rewards/rejected": 0.003497645452425077, "logps/rejected": -194.7725497159091, "logits/rejected": -48328421.402597405, "rewards/margins": 0.030930112270360598, "regularization/tv": 0.0035591290798038244, "regularization/advantage_range": 1.1863764524459839, "kl": 0.2043342888355255, "stability/response_length_mean": 32.25, "stability/response_length_std": 27.932315826416016, "stability/response_length_var": 780.2142944335938, "stability/token_entropy_mean": 4.69921875, "stability/repetition_rate_mean": 0.2897879481315613}
+{"current_steps": 180, "total_steps": 196, "loss": 0.4997, "lr": 9.928226601167138e-07, "epoch": 0.9213051823416507, "percentage": 91.84, "elapsed_time": "0:48:18", "remaining_time": "0:04:17", "rewards/chosen": 0.0362014126923918, "logps/chosen": -375.78383819018404, "logits/chosen": -52875565.54601227, "rewards/rejected": 0.009764606785622371, "logps/rejected": -211.20056230095543, "logits/rejected": -48260839.541401275, "rewards/margins": 0.026436805906769428, "regularization/tv": 0.003670051693916321, "regularization/advantage_range": 1.2233505249023438, "kl": 0.2203982174396515, "stability/response_length_mean": 40.375, "stability/response_length_std": 32.61874008178711, "stability/response_length_var": 1063.982177734375, "stability/token_entropy_mean": 3.080078125, "stability/repetition_rate_mean": 0.265625}
+{"current_steps": 185, "total_steps": 196, "loss": 0.5001, "lr": 9.918879275179817e-07, "epoch": 0.946896992962252, "percentage": 94.39, "elapsed_time": "0:49:37", "remaining_time": "0:02:57", "rewards/chosen": 0.02970717277055905, "logps/chosen": -311.7043065200617, "logits/chosen": -54872038.71604938, "rewards/rejected": 0.01059295105028756, "logps/rejected": -196.933222903481, "logits/rejected": -49803879.69620253, "rewards/margins": 0.01911422172027149, "regularization/tv": 0.0034396275877952576, "regularization/advantage_range": 1.1465425491333008, "kl": 0.23888865113258362, "stability/response_length_mean": 26.25, "stability/response_length_std": 31.45404815673828, "stability/response_length_var": 989.3571166992188, "stability/token_entropy_mean": 3.796875, "stability/repetition_rate_mean": 0.20561079680919647}
+{"current_steps": 190, "total_steps": 196, "loss": 0.5, "lr": 9.90896490997906e-07, "epoch": 0.9724888035828535, "percentage": 96.94, "elapsed_time": "0:50:58", "remaining_time": "0:01:36", "rewards/chosen": 0.04663035112940623, "logps/chosen": -391.9390203967066, "logits/chosen": -53337394.58682635, "rewards/rejected": 0.009832122746635885, "logps/rejected": -211.10026041666666, "logits/rejected": -49988594.614379086, "rewards/margins": 0.036798228382770345, "regularization/tv": 0.004264469724148512, "regularization/advantage_range": 1.421489953994751, "kl": 0.2414882481098175, "stability/response_length_mean": 11.375, "stability/response_length_std": 21.738296508789062, "stability/response_length_var": 472.5535583496094, "stability/token_entropy_mean": 5.33984375, "stability/repetition_rate_mean": 0.1086309552192688}
+{"current_steps": 195, "total_steps": 196, "loss": 0.4997, "lr": 9.898484648474362e-07, "epoch": 0.9980806142034548, "percentage": 99.49, "elapsed_time": "0:52:18", "remaining_time": "0:00:16", "rewards/chosen": 0.04602757423154769, "logps/chosen": -392.32464717741937, "logits/chosen": -60425077.26451613, "rewards/rejected": 0.00917442495172674, "logps/rejected": -197.82759232954547, "logits/rejected": -54190259.97575758, "rewards/margins": 0.03685314927982095, "regularization/tv": 0.0038973658811300993, "regularization/advantage_range": 1.2991220951080322, "kl": 0.22969751060009003, "stability/response_length_mean": 41.0, "stability/response_length_std": 31.807456970214844, "stability/response_length_var": 1011.7142944335938, "stability/token_entropy_mean": 4.40234375, "stability/repetition_rate_mean": 0.4031808078289032}
+{"current_steps": 196, "total_steps": 196, "epoch": 1.0, "percentage": 100.0, "elapsed_time": "0:52:32", "remaining_time": "0:00:00"}

mistral7b_kto_tv_rho0.003/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": 0.6752910017967224,
+      "kl": 0.009316612966358662,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59711191.578947365,
+      "logits/rejected": -50416146.28571428,
+      "logps/chosen": -439.0247224506579,
+      "logps/rejected": -198.40827287946428,
+      "loss": 0.5011,
+      "loss/base_kto": 0.49975714087486267,
+      "metrics/advantage_var": 0.05327611044049263,
+      "regularization/advantage_range": 0.46398472785949707,
+      "regularization/tv": 0.001391954137943685,
+      "rewards/chosen": 0.002090120393978922,
+      "rewards/margins": 0.0019776263836780586,
+      "rewards/rejected": 0.00011249401030086336,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.7144330739974976,
+      "kl": 0.01776806451380253,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59301914.597402595,
+      "logits/rejected": -50024296.86746988,
+      "logps/chosen": -429.9064021915584,
+      "logps/rejected": -192.8597985692771,
+      "loss": 0.5019,
+      "loss/base_kto": 0.49971505999565125,
+      "metrics/advantage_var": 0.06181863695383072,
+      "regularization/advantage_range": 0.5836383700370789,
+      "regularization/tv": 0.0017509151948615909,
+      "rewards/chosen": 0.0032109109225211207,
+      "rewards/margins": 0.0023968832870731706,
+      "rewards/rejected": 0.0008140276354479503,
+      "stability/repetition_rate_mean": 0.3531753420829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.3056640625,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.6771201491355896,
+      "kl": 0.01534965168684721,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56231693.96363636,
+      "logits/rejected": -51925735.22580645,
+      "logps/chosen": -404.1009943181818,
+      "logps/rejected": -200.66796875,
+      "loss": 0.5018,
+      "loss/base_kto": 0.49993228912353516,
+      "metrics/advantage_var": 0.05437135323882103,
+      "regularization/advantage_range": 0.526348888874054,
+      "regularization/tv": 0.001579046598635614,
+      "rewards/chosen": 0.002984200824390758,
+      "rewards/margins": 0.0003753908918050727,
+      "rewards/rejected": 0.0026088099325856855,
+      "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": 3.7392578125,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.7161915302276611,
+      "kl": 0.021325770765542984,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51672825.6,
+      "logits/rejected": -48064236.8,
+      "logps/chosen": -343.6939697265625,
+      "logps/rejected": -225.6203369140625,
+      "loss": 0.5019,
+      "loss/base_kto": 0.5005207061767578,
+      "metrics/advantage_var": 0.052058931440114975,
+      "regularization/advantage_range": 0.529035210609436,
+      "regularization/tv": 0.0015871055657044053,
+      "rewards/chosen": 0.0002750195562839508,
+      "rewards/margins": -0.004027188569307327,
+      "rewards/rejected": 0.004302208125591278,
+      "stability/repetition_rate_mean": 0.2194320410490036,
+      "stability/response_length_mean": 44.0,
+      "stability/response_length_std": 28.0611572265625,
+      "stability/response_length_var": 787.4285888671875,
+      "stability/token_entropy_mean": 4.4296875,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.6439714431762695,
+      "kl": 0.012540136463940144,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56468112.77241379,
+      "logits/rejected": -52019515.977142856,
+      "logps/chosen": -378.4657058189655,
+      "logps/rejected": -202.78388392857144,
+      "loss": 0.5019,
+      "loss/base_kto": 0.499775230884552,
+      "metrics/advantage_var": 0.05616157129406929,
+      "regularization/advantage_range": 0.5223817825317383,
+      "regularization/tv": 0.0015671454602852464,
+      "rewards/chosen": 0.002574328924047536,
+      "rewards/margins": 0.0016784496554012956,
+      "rewards/rejected": 0.0008958792686462403,
+      "stability/repetition_rate_mean": 0.16044004261493683,
+      "stability/response_length_mean": 43.375,
+      "stability/response_length_std": 28.54538917541504,
+      "stability/response_length_var": 814.8392944335938,
+      "stability/token_entropy_mean": 4.806640625,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.6126549243927002,
+      "kl": 0.01668705977499485,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57719072.644295305,
+      "logits/rejected": -51523212.7251462,
+      "logps/chosen": -468.5598259228188,
+      "logps/rejected": -198.2305144371345,
+      "loss": 0.502,
+      "loss/base_kto": 0.4999951422214508,
+      "metrics/advantage_var": 0.05605809763073921,
+      "regularization/advantage_range": 0.542860746383667,
+      "regularization/tv": 0.0016285822493955493,
+      "rewards/chosen": 0.0021430586008417524,
+      "rewards/margins": 0.0004062107972351279,
+      "rewards/rejected": 0.0017368478036066245,
+      "stability/repetition_rate_mean": 0.17223010957241058,
+      "stability/response_length_mean": 45.875,
+      "stability/response_length_std": 28.246049880981445,
+      "stability/response_length_var": 797.8392944335938,
+      "stability/token_entropy_mean": 4.57421875,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6238469481468201,
+      "kl": 0.018097203224897385,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57647505.82278481,
+      "logits/rejected": -53915256.09876543,
+      "logps/chosen": -385.8966327136076,
+      "logps/rejected": -195.1175250771605,
+      "loss": 0.5017,
+      "loss/base_kto": 0.4996470510959625,
+      "metrics/advantage_var": 0.07482800632715225,
+      "regularization/advantage_range": 0.5923306345939636,
+      "regularization/tv": 0.001776991761289537,
+      "rewards/chosen": 0.003989132899272291,
+      "rewards/margins": 0.0030932442741685405,
+      "rewards/rejected": 0.0008958886251037503,
+      "stability/repetition_rate_mean": 0.3147321343421936,
+      "stability/response_length_mean": 56.875,
+      "stability/response_length_std": 20.152544021606445,
+      "stability/response_length_var": 406.125,
+      "stability/token_entropy_mean": 5.421875,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.5992166996002197,
+      "kl": 0.028051286935806274,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56631635.2,
+      "logits/rejected": -50227708.8,
+      "logps/chosen": -421.901123046875,
+      "logps/rejected": -219.2706787109375,
+      "loss": 0.5019,
+      "loss/base_kto": 0.5005082488059998,
+      "metrics/advantage_var": 0.07147374749183655,
+      "regularization/advantage_range": 0.6278894543647766,
+      "regularization/tv": 0.0018836684757843614,
+      "rewards/chosen": 0.00015102624893188478,
+      "rewards/margins": -0.0037047527730464935,
+      "rewards/rejected": 0.0038557790219783785,
+      "stability/repetition_rate_mean": 0.13671875,
+      "stability/response_length_mean": 33.125,
+      "stability/response_length_std": 33.01704025268555,
+      "stability/response_length_var": 1090.125,
+      "stability/token_entropy_mean": 4.8896484375,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.6807482838630676,
+      "kl": 0.03983687236905098,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52712994.36241611,
+      "logits/rejected": -50780387.55555555,
+      "logps/chosen": -352.2607487416107,
+      "logps/rejected": -243.8278965643275,
+      "loss": 0.5015,
+      "loss/base_kto": 0.4997825622558594,
+      "metrics/advantage_var": 0.08648205548524857,
+      "regularization/advantage_range": 0.661689281463623,
+      "regularization/tv": 0.001985067967325449,
+      "rewards/chosen": 0.006769456319360925,
+      "rewards/margins": 0.002141746121956811,
+      "rewards/rejected": 0.004627710197404114,
+      "stability/repetition_rate_mean": 0.0994017943739891,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.51875877380371,
+      "stability/response_length_var": 871.3571166992188,
+      "stability/token_entropy_mean": 4.70703125,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.6231204271316528,
+      "kl": 0.03668095916509628,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54021566.7114094,
+      "logits/rejected": -48286264.88888889,
+      "logps/chosen": -375.1455012583893,
+      "logps/rejected": -180.17639802631578,
+      "loss": 0.5017,
+      "loss/base_kto": 0.5000646710395813,
+      "metrics/advantage_var": 0.05680838227272034,
+      "regularization/advantage_range": 0.5559825301170349,
+      "regularization/tv": 0.0016679477412253618,
+      "rewards/chosen": 0.005363105927537751,
+      "rewards/margins": -0.0010589264621867125,
+      "rewards/rejected": 0.006422032389724464,
+      "stability/repetition_rate_mean": 0.3288106620311737,
+      "stability/response_length_mean": 55.375,
+      "stability/response_length_std": 16.361431121826172,
+      "stability/response_length_var": 267.6964416503906,
+      "stability/token_entropy_mean": 4.068359375,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.7771005034446716,
+      "kl": 0.043028466403484344,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56951623.55279503,
+      "logits/rejected": -49562733.48427673,
+      "logps/chosen": -344.49801048136646,
+      "logps/rejected": -230.3948506289308,
+      "loss": 0.5017,
+      "loss/base_kto": 0.5004634857177734,
+      "metrics/advantage_var": 0.06657923012971878,
+      "regularization/advantage_range": 0.5866410136222839,
+      "regularization/tv": 0.0017599230632185936,
+      "rewards/chosen": 0.0014025861432093271,
+      "rewards/margins": -0.003914588270459707,
+      "rewards/rejected": 0.005317174413669034,
+      "stability/repetition_rate_mean": 0.111328125,
+      "stability/response_length_mean": 24.5,
+      "stability/response_length_std": 32.710853576660156,
+      "stability/response_length_var": 1070.0,
+      "stability/token_entropy_mean": 4.388671875,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.6669939160346985,
+      "kl": 0.03918389976024628,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52672884.36363637,
+      "logits/rejected": -47833497.6,
+      "logps/chosen": -359.6323863636364,
+      "logps/rejected": -178.17522681451612,
+      "loss": 0.5016,
+      "loss/base_kto": 0.49955543875694275,
+      "metrics/advantage_var": 0.06515418738126755,
+      "regularization/advantage_range": 0.5985627770423889,
+      "regularization/tv": 0.0017956882948055863,
+      "rewards/chosen": 0.008838949781475645,
+      "rewards/margins": 0.0036711697820810745,
+      "rewards/rejected": 0.0051677799993945705,
+      "stability/repetition_rate_mean": 0.21596631407737732,
+      "stability/response_length_mean": 52.625,
+      "stability/response_length_std": 21.75800323486328,
+      "stability/response_length_var": 473.41070556640625,
+      "stability/token_entropy_mean": 5.29296875,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.6477825045585632,
+      "kl": 0.04982197284698486,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57567218.787096776,
+      "logits/rejected": -47273661.28484848,
+      "logps/chosen": -381.8376512096774,
+      "logps/rejected": -183.54024621212122,
+      "loss": 0.5013,
+      "loss/base_kto": 0.49914589524269104,
+      "metrics/advantage_var": 0.0746908187866211,
+      "regularization/advantage_range": 0.6157255172729492,
+      "regularization/tv": 0.001847176463343203,
+      "rewards/chosen": 0.01091364276024603,
+      "rewards/margins": 0.006703484932348525,
+      "rewards/rejected": 0.004210157827897506,
+      "stability/repetition_rate_mean": 0.1857219785451889,
+      "stability/response_length_mean": 20.625,
+      "stability/response_length_std": 27.176342010498047,
+      "stability/response_length_var": 738.5535888671875,
+      "stability/token_entropy_mean": 5.544921875,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5892289280891418,
+      "kl": 0.06117330864071846,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52787001.80645161,
+      "logits/rejected": -49036244.557575755,
+      "logps/chosen": -379.0451612903226,
+      "logps/rejected": -220.39483901515152,
+      "loss": 0.5011,
+      "loss/base_kto": 0.4993252456188202,
+      "metrics/advantage_var": 0.07192724943161011,
+      "regularization/advantage_range": 0.6263827681541443,
+      "regularization/tv": 0.001879148418083787,
+      "rewards/chosen": 0.008328685452861171,
+      "rewards/margins": 0.005595458474443572,
+      "rewards/rejected": 0.0027332269784175987,
+      "stability/repetition_rate_mean": 0.11328125,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.4404296875,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.6382232308387756,
+      "kl": 0.07038548588752747,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49505947.675675675,
+      "logits/rejected": -47897421.39534884,
+      "logps/chosen": -314.17166385135135,
+      "logps/rejected": -228.84547601744185,
+      "loss": 0.5015,
+      "loss/base_kto": 0.4992240369319916,
+      "metrics/advantage_var": 0.08677860349416733,
+      "regularization/advantage_range": 0.7051105499267578,
+      "regularization/tv": 0.0021153318230062723,
+      "rewards/chosen": 0.009913306783985448,
+      "rewards/margins": 0.005977991895807532,
+      "rewards/rejected": 0.003935314888177916,
+      "stability/repetition_rate_mean": 0.08203125,
+      "stability/response_length_mean": 24.25,
+      "stability/response_length_std": 32.92306900024414,
+      "stability/response_length_var": 1083.9285888671875,
+      "stability/token_entropy_mean": 1.883056640625,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.6527356505393982,
+      "kl": 0.08504229038953781,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58622535.205298014,
+      "logits/rejected": -53446995.313609466,
+      "logps/chosen": -403.9296616307947,
+      "logps/rejected": -201.01654955621302,
+      "loss": 0.5014,
+      "loss/base_kto": 0.49918365478515625,
+      "metrics/advantage_var": 0.06651351600885391,
+      "regularization/advantage_range": 0.5964646935462952,
+      "regularization/tv": 0.0017893938347697258,
+      "rewards/chosen": 0.010046138668691876,
+      "rewards/margins": 0.006118507530347388,
+      "rewards/rejected": 0.003927631138344488,
+      "stability/repetition_rate_mean": 0.19088542461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.23828125,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.6321583986282349,
+      "kl": 0.0836004987359047,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51481600.0,
+      "logits/rejected": -47146399.806060605,
+      "logps/chosen": -358.2110635080645,
+      "logps/rejected": -230.3817234848485,
+      "loss": 0.5012,
+      "loss/base_kto": 0.49900388717651367,
+      "metrics/advantage_var": 0.08175607770681381,
+      "regularization/advantage_range": 0.68501877784729,
+      "regularization/tv": 0.0020550561603158712,
+      "rewards/chosen": 0.012145080104950936,
+      "rewards/margins": 0.007835714500087913,
+      "rewards/rejected": 0.004309365604863022,
+      "stability/repetition_rate_mean": 0.1369977593421936,
+      "stability/response_length_mean": 28.0,
+      "stability/response_length_std": 31.181495666503906,
+      "stability/response_length_var": 972.2857055664062,
+      "stability/token_entropy_mean": 4.96484375,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.5965731143951416,
+      "kl": 0.09354694932699203,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58432328.20512821,
+      "logits/rejected": -50573452.487804875,
+      "logps/chosen": -380.6665665064103,
+      "logps/rejected": -187.1869521722561,
+      "loss": 0.5008,
+      "loss/base_kto": 0.49893808364868164,
+      "metrics/advantage_var": 0.07857906073331833,
+      "regularization/advantage_range": 0.639580249786377,
+      "regularization/tv": 0.001918740919791162,
+      "rewards/chosen": 0.012280461115714831,
+      "rewards/margins": 0.008296260597409122,
+      "rewards/rejected": 0.0039842005183057085,
+      "stability/repetition_rate_mean": 0.2582859992980957,
+      "stability/response_length_mean": 35.125,
+      "stability/response_length_std": 31.638301849365234,
+      "stability/response_length_var": 1000.9821166992188,
+      "stability/token_entropy_mean": 2.94384765625,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.5790473818778992,
+      "kl": 0.08265768736600876,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50400852.25316456,
+      "logits/rejected": -48991902.02469136,
+      "logps/chosen": -314.8856309335443,
+      "logps/rejected": -207.14375964506172,
+      "loss": 0.5009,
+      "loss/base_kto": 0.49863290786743164,
+      "metrics/advantage_var": 0.07834162563085556,
+      "regularization/advantage_range": 0.6240283250808716,
+      "regularization/tv": 0.0018720850348472595,
+      "rewards/chosen": 0.010773800596406188,
+      "rewards/margins": 0.010777551641798071,
+      "rewards/rejected": -3.7510453918833793e-06,
+      "stability/repetition_rate_mean": 0.19374233484268188,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 31.1264705657959,
+      "stability/response_length_var": 968.8571166992188,
+      "stability/token_entropy_mean": 3.736328125,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.5900207161903381,
+      "kl": 0.08847284317016602,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54724199.783783786,
+      "logits/rejected": -50932414.511627905,
+      "logps/chosen": -359.7025971283784,
+      "logps/rejected": -216.51346747819767,
+      "loss": 0.501,
+      "loss/base_kto": 0.4986952841281891,
+      "metrics/advantage_var": 0.08461682498455048,
+      "regularization/advantage_range": 0.6794241070747375,
+      "regularization/tv": 0.002038272563368082,
+      "rewards/chosen": 0.01675953736176362,
+      "rewards/margins": 0.01021829631476939,
+      "rewards/rejected": 0.006541241046994231,
+      "stability/repetition_rate_mean": 0.2421875,
+      "stability/response_length_mean": 41.25,
+      "stability/response_length_std": 31.476749420166016,
+      "stability/response_length_var": 990.7857055664062,
+      "stability/token_entropy_mean": 4.107421875,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6134894490242004,
+      "kl": 0.09840010851621628,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59393332.43373494,
+      "logits/rejected": -53833864.31168831,
+      "logps/chosen": -399.74868222891564,
+      "logps/rejected": -209.9587307224026,
+      "loss": 0.5009,
+      "loss/base_kto": 0.49860039353370667,
+      "metrics/advantage_var": 0.10574636608362198,
+      "regularization/advantage_range": 0.7764409184455872,
+      "regularization/tv": 0.0023293232079595327,
+      "rewards/chosen": 0.015377698174442154,
+      "rewards/margins": 0.010868746132717095,
+      "rewards/rejected": 0.004508952041725059,
+      "stability/repetition_rate_mean": 0.1916852593421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.388671875,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.6036498546600342,
+      "kl": 0.10348369181156158,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57405316.62650602,
+      "logits/rejected": -53642688.83116883,
+      "logps/chosen": -378.44039439006025,
+      "logps/rejected": -206.05346996753246,
+      "loss": 0.5008,
+      "loss/base_kto": 0.4989023208618164,
+      "metrics/advantage_var": 0.1012769564986229,
+      "regularization/advantage_range": 0.7773522734642029,
+      "regularization/tv": 0.002332056639716029,
+      "rewards/chosen": 0.01228764545486634,
+      "rewards/margins": 0.008821119819704383,
+      "rewards/rejected": 0.003466525635161957,
+      "stability/repetition_rate_mean": 0.3186553120613098,
+      "stability/response_length_mean": 31.125,
+      "stability/response_length_std": 28.412460327148438,
+      "stability/response_length_var": 807.2678833007812,
+      "stability/token_entropy_mean": 4.0673828125,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.5774539709091187,
+      "kl": 0.11748156696557999,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60635421.164556965,
+      "logits/rejected": -53458703.802469134,
+      "logps/chosen": -364.39875890031647,
+      "logps/rejected": -201.12389081790124,
+      "loss": 0.5008,
+      "loss/base_kto": 0.4984841048717499,
+      "metrics/advantage_var": 0.08921544253826141,
+      "regularization/advantage_range": 0.6971777081489563,
+      "regularization/tv": 0.0020915332715958357,
+      "rewards/chosen": 0.019377651093881343,
+      "rewards/margins": 0.01210962627842792,
+      "rewards/rejected": 0.007268024815453423,
+      "stability/repetition_rate_mean": 0.2916666865348816,
+      "stability/response_length_mean": 41.375,
+      "stability/response_length_std": 31.25442886352539,
+      "stability/response_length_var": 976.8392944335938,
+      "stability/token_entropy_mean": 4.935546875,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.5666119456291199,
+      "kl": 0.1038687452673912,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57774535.11111111,
+      "logits/rejected": -54427586.68263473,
+      "logps/chosen": -384.06045751633985,
+      "logps/rejected": -212.17372286676647,
+      "loss": 0.5004,
+      "loss/base_kto": 0.4982244670391083,
+      "metrics/advantage_var": 0.11172550171613693,
+      "regularization/advantage_range": 0.8190727233886719,
+      "regularization/tv": 0.0024572184775024652,
+      "rewards/chosen": 0.018445407643037683,
+      "rewards/margins": 0.014180560395851147,
+      "rewards/rejected": 0.0042648472471865356,
+      "stability/repetition_rate_mean": 0.1664772778749466,
+      "stability/response_length_mean": 27.875,
+      "stability/response_length_std": 30.31943130493164,
+      "stability/response_length_var": 919.2678833007812,
+      "stability/token_entropy_mean": 4.623046875,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.5878152251243591,
+      "kl": 0.11783448606729507,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60747207.89041096,
+      "logits/rejected": -55728210.390804596,
+      "logps/chosen": -417.2116866438356,
+      "logps/rejected": -216.30994073275863,
+      "loss": 0.5008,
+      "loss/base_kto": 0.4985935389995575,
+      "metrics/advantage_var": 0.12825100123882294,
+      "regularization/advantage_range": 0.9061002731323242,
+      "regularization/tv": 0.002718301024287939,
+      "rewards/chosen": 0.01676147114740659,
+      "rewards/margins": 0.011036303883893267,
+      "rewards/rejected": 0.005725167263513324,
+      "stability/repetition_rate_mean": 0.16328126192092896,
+      "stability/response_length_mean": 15.25,
+      "stability/response_length_std": 24.93276596069336,
+      "stability/response_length_var": 621.6428833007812,
+      "stability/token_entropy_mean": 3.6328125,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.5727063417434692,
+      "kl": 0.11419554054737091,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57436979.2,
+      "logits/rejected": -53890463.806060605,
+      "logps/chosen": -392.4932459677419,
+      "logps/rejected": -183.90649857954546,
+      "loss": 0.5003,
+      "loss/base_kto": 0.4980848431587219,
+      "metrics/advantage_var": 0.12817950546741486,
+      "regularization/advantage_range": 0.8711000680923462,
+      "regularization/tv": 0.002613300224766135,
+      "rewards/chosen": 0.016469930833385837,
+      "rewards/margins": 0.015228715582676997,
+      "rewards/rejected": 0.0012412152507088401,
+      "stability/repetition_rate_mean": 0.19791193306446075,
+      "stability/response_length_mean": 38.0,
+      "stability/response_length_std": 29.13515281677246,
+      "stability/response_length_var": 848.8571166992188,
+      "stability/token_entropy_mean": 3.99609375,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5919101238250732,
+      "kl": 0.12443824112415314,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59916112.09815951,
+      "logits/rejected": -55112397.4522293,
+      "logps/chosen": -401.59190471625766,
+      "logps/rejected": -247.84937300955414,
+      "loss": 0.5005,
+      "loss/base_kto": 0.4973863661289215,
+      "metrics/advantage_var": 0.18036113679409027,
+      "regularization/advantage_range": 1.082304835319519,
+      "regularization/tv": 0.0032469145953655243,
+      "rewards/chosen": 0.022427939198499808,
+      "rewards/margins": 0.020287977934563772,
+      "rewards/rejected": 0.0021399612639360368,
+      "stability/repetition_rate_mean": 0.2230902761220932,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.936838150024414,
+      "stability/response_length_var": 896.2142944335938,
+      "stability/token_entropy_mean": 4.07421875,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.5441713333129883,
+      "kl": 0.10719374567270279,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57166928.67878788,
+      "logits/rejected": -50186933.67741936,
+      "logps/chosen": -367.0013494318182,
+      "logps/rejected": -190.0579133064516,
+      "loss": 0.5003,
+      "loss/base_kto": 0.49737364053726196,
+      "metrics/advantage_var": 0.17915482819080353,
+      "regularization/advantage_range": 1.0351334810256958,
+      "regularization/tv": 0.0031054005958139896,
+      "rewards/chosen": 0.018839573137687915,
+      "rewards/margins": 0.02118338370369793,
+      "rewards/rejected": -0.0023438105660100136,
+      "stability/repetition_rate_mean": 0.35546875,
+      "stability/response_length_mean": 56.25,
+      "stability/response_length_std": 21.920310974121094,
+      "stability/response_length_var": 480.5,
+      "stability/token_entropy_mean": 4.7451171875,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.5883694887161255,
+      "kl": 0.1362670212984085,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51238381.4939759,
+      "logits/rejected": -51789099.22077922,
+      "logps/chosen": -330.10217432228916,
+      "logps/rejected": -199.48333502435065,
+      "loss": 0.5005,
+      "loss/base_kto": 0.49796611070632935,
+      "metrics/advantage_var": 0.1358262300491333,
+      "regularization/advantage_range": 0.8813913464546204,
+      "regularization/tv": 0.0026441742666065693,
+      "rewards/chosen": 0.016965202538363904,
+      "rewards/margins": 0.016506232452996924,
+      "rewards/rejected": 0.0004589700853669798,
+      "stability/repetition_rate_mean": 0.20039846003055573,
+      "stability/response_length_mean": 22.625,
+      "stability/response_length_std": 26.359804153442383,
+      "stability/response_length_var": 694.8392944335938,
+      "stability/token_entropy_mean": 5.703125,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.5959792733192444,
+      "kl": 0.14728893339633942,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55525086.870588236,
+      "logits/rejected": -50325459.626666665,
+      "logps/chosen": -397.62251838235295,
+      "logps/rejected": -207.71174479166666,
+      "loss": 0.5003,
+      "loss/base_kto": 0.49606361985206604,
+      "metrics/advantage_var": 0.18760840594768524,
+      "regularization/advantage_range": 1.0953562259674072,
+      "regularization/tv": 0.0032860690262168646,
+      "rewards/chosen": 0.031186995786779068,
+      "rewards/margins": 0.02976833888128692,
+      "rewards/rejected": 0.0014186569054921469,
+      "stability/repetition_rate_mean": 0.2408854216337204,
+      "stability/response_length_mean": 35.25,
+      "stability/response_length_std": 31.01957893371582,
+      "stability/response_length_var": 962.2142944335938,
+      "stability/token_entropy_mean": 4.6015625,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.541877269744873,
+      "kl": 0.14850066602230072,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54157408.37647059,
+      "logits/rejected": -49500801.70666666,
+      "logps/chosen": -360.8025505514706,
+      "logps/rejected": -191.32903645833332,
+      "loss": 0.5003,
+      "loss/base_kto": 0.4973774552345276,
+      "metrics/advantage_var": 0.16561715304851532,
+      "regularization/advantage_range": 1.0279508829116821,
+      "regularization/tv": 0.003083852818235755,
+      "rewards/chosen": 0.01867367520051844,
+      "rewards/margins": 0.02101714923101313,
+      "rewards/rejected": -0.00234347403049469,
+      "stability/repetition_rate_mean": 0.3210846781730652,
+      "stability/response_length_mean": 61.375,
+      "stability/response_length_std": 7.424621105194092,
+      "stability/response_length_var": 55.125,
+      "stability/token_entropy_mean": 4.5986328125,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.5596888065338135,
+      "kl": 0.15647390484809875,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55398857.14285714,
+      "logits/rejected": -49154202.94736842,
+      "logps/chosen": -416.60086495535717,
+      "logps/rejected": -186.86446340460526,
+      "loss": 0.5004,
+      "loss/base_kto": 0.49669837951660156,
+      "metrics/advantage_var": 0.21187877655029297,
+      "regularization/advantage_range": 1.114825963973999,
+      "regularization/tv": 0.003344477852806449,
+      "rewards/chosen": 0.02944736821310861,
+      "rewards/margins": 0.026160127742398055,
+      "rewards/rejected": 0.0032872404707105537,
+      "stability/repetition_rate_mean": 0.31046873331069946,
+      "stability/response_length_mean": 32.5,
+      "stability/response_length_std": 30.528675079345703,
+      "stability/response_length_var": 932.0,
+      "stability/token_entropy_mean": 4.3857421875,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.6522524952888489,
+      "kl": 0.14636269211769104,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54598391.172413796,
+      "logits/rejected": -50009463.23287671,
+      "logps/chosen": -370.67030352011494,
+      "logps/rejected": -212.60320794092465,
+      "loss": 0.5,
+      "loss/base_kto": 0.49680137634277344,
+      "metrics/advantage_var": 0.15778213739395142,
+      "regularization/advantage_range": 1.0048089027404785,
+      "regularization/tv": 0.003014426678419113,
+      "rewards/chosen": 0.02464700841355598,
+      "rewards/margins": 0.02575588618262774,
+      "rewards/rejected": -0.0011088777690717619,
+      "stability/repetition_rate_mean": 0.20468750596046448,
+      "stability/response_length_mean": 12.625,
+      "stability/response_length_std": 21.460512161254883,
+      "stability/response_length_var": 460.5535583496094,
+      "stability/token_entropy_mean": 4.828125,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.5037974715232849,
+      "kl": 0.19872891902923584,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59318084.165680476,
+      "logits/rejected": -52844435.49668874,
+      "logps/chosen": -441.0094767011834,
+      "logps/rejected": -191.21860771937085,
+      "loss": 0.4998,
+      "loss/base_kto": 0.49641498923301697,
+      "metrics/advantage_var": 0.20756979286670685,
+      "regularization/advantage_range": 1.1968477964401245,
+      "regularization/tv": 0.0035905432887375355,
+      "rewards/chosen": 0.03323545399502184,
+      "rewards/margins": 0.02931231930177504,
+      "rewards/rejected": 0.003923134693246804,
+      "stability/repetition_rate_mean": 0.20866228640079498,
+      "stability/response_length_mean": 28.25,
+      "stability/response_length_std": 30.844078063964844,
+      "stability/response_length_var": 951.3571166992188,
+      "stability/token_entropy_mean": 3.31689453125,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.5411744117736816,
+      "kl": 0.2043342888355255,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56856687.03614458,
+      "logits/rejected": -48328421.402597405,
+      "logps/chosen": -356.0929028614458,
+      "logps/rejected": -194.7725497159091,
+      "loss": 0.4997,
+      "loss/base_kto": 0.49607935547828674,
+      "metrics/advantage_var": 0.22021155059337616,
+      "regularization/advantage_range": 1.1863764524459839,
+      "regularization/tv": 0.0035591290798038244,
+      "rewards/chosen": 0.034427757722785675,
+      "rewards/margins": 0.030930112270360598,
+      "rewards/rejected": 0.003497645452425077,
+      "stability/repetition_rate_mean": 0.2897879481315613,
+      "stability/response_length_mean": 32.25,
+      "stability/response_length_std": 27.932315826416016,
+      "stability/response_length_var": 780.2142944335938,
+      "stability/token_entropy_mean": 4.69921875,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.5480484366416931,
+      "kl": 0.2203982174396515,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52875565.54601227,
+      "logits/rejected": -48260839.541401275,
+      "logps/chosen": -375.78383819018404,
+      "logps/rejected": -211.20056230095543,
+      "loss": 0.4997,
+      "loss/base_kto": 0.4966592788696289,
+      "metrics/advantage_var": 0.2269420623779297,
+      "regularization/advantage_range": 1.2233505249023438,
+      "regularization/tv": 0.003670051693916321,
+      "rewards/chosen": 0.0362014126923918,
+      "rewards/margins": 0.026436805906769428,
+      "rewards/rejected": 0.009764606785622371,
+      "stability/repetition_rate_mean": 0.265625,
+      "stability/response_length_mean": 40.375,
+      "stability/response_length_std": 32.61874008178711,
+      "stability/response_length_var": 1063.982177734375,
+      "stability/token_entropy_mean": 3.080078125,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.5584241151809692,
+      "kl": 0.23888865113258362,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54872038.71604938,
+      "logits/rejected": -49803879.69620253,
+      "logps/chosen": -311.7043065200617,
+      "logps/rejected": -196.933222903481,
+      "loss": 0.5001,
+      "loss/base_kto": 0.49761316180229187,
+      "metrics/advantage_var": 0.19539952278137207,
+      "regularization/advantage_range": 1.1465425491333008,
+      "regularization/tv": 0.0034396275877952576,
+      "rewards/chosen": 0.02970717277055905,
+      "rewards/margins": 0.01911422172027149,
+      "rewards/rejected": 0.01059295105028756,
+      "stability/repetition_rate_mean": 0.20561079680919647,
+      "stability/response_length_mean": 26.25,
+      "stability/response_length_std": 31.45404815673828,
+      "stability/response_length_var": 989.3571166992188,
+      "stability/token_entropy_mean": 3.796875,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.5378943085670471,
+      "kl": 0.2414882481098175,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53337394.58682635,
+      "logits/rejected": -49988594.614379086,
+      "logps/chosen": -391.9390203967066,
+      "logps/rejected": -211.10026041666666,
+      "loss": 0.5,
+      "loss/base_kto": 0.4952877163887024,
+      "metrics/advantage_var": 0.31047913432121277,
+      "regularization/advantage_range": 1.421489953994751,
+      "regularization/tv": 0.004264469724148512,
+      "rewards/chosen": 0.04663035112940623,
+      "rewards/margins": 0.036798228382770345,
+      "rewards/rejected": 0.009832122746635885,
+      "stability/repetition_rate_mean": 0.1086309552192688,
+      "stability/response_length_mean": 11.375,
+      "stability/response_length_std": 21.738296508789062,
+      "stability/response_length_var": 472.5535583496094,
+      "stability/token_entropy_mean": 5.33984375,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.546999454498291,
+      "kl": 0.22969751060009003,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60425077.26451613,
+      "logits/rejected": -54190259.97575758,
+      "logps/chosen": -392.32464717741937,
+      "logps/rejected": -197.82759232954547,
+      "loss": 0.4997,
+      "loss/base_kto": 0.4953397214412689,
+      "metrics/advantage_var": 0.2615717351436615,
+      "regularization/advantage_range": 1.2991220951080322,
+      "regularization/tv": 0.0038973658811300993,
+      "rewards/chosen": 0.04602757423154769,
+      "rewards/margins": 0.03685314927982095,
+      "rewards/rejected": 0.00917442495172674,
+      "stability/repetition_rate_mean": 0.4031808078289032,
+      "stability/response_length_mean": 41.0,
+      "stability/response_length_std": 31.807456970214844,
+      "stability/response_length_var": 1011.7142944335938,
+      "stability/token_entropy_mean": 4.40234375,
+      "step": 195
+    },
+    {
+      "epoch": 1.0,
+      "step": 196,
+      "total_flos": 3.1861866257511875e+18,
+      "train_loss": 0.5009068980508921,
+      "train_runtime": 3152.3184,
+      "train_samples_per_second": 31.723,
+      "train_steps_per_second": 0.062
+    }
+  ],
+  "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_tv_rho0.003/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:50ed4a0ba877bcff5b82e539a8621186c048c171ba078ef5bb91bed25f698cba
+size 8017