Upload folder using huggingface_hub

mistral7b_kto_tv_rho0.001/README.md

@@ -0,0 +1,64 @@
+---
+library_name: peft
+license: other
+base_model: mistralai/Mistral-7B-v0.1
+tags:
+- base_model:adapter:mistralai/Mistral-7B-v0.1
+- llama-factory
+- lora
+- transformers
+pipeline_tag: text-generation
+model-index:
+- name: mistral7b_kto_tv_rho0.001
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+# mistral7b_kto_tv_rho0.001
+
+This model is a fine-tuned version of [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) on the shp_kto_train dataset.
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 1e-06
+- train_batch_size: 8
+- eval_batch_size: 8
+- seed: 42
+- distributed_type: multi-GPU
+- num_devices: 8
+- gradient_accumulation_steps: 8
+- total_train_batch_size: 512
+- total_eval_batch_size: 64
+- optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
+- lr_scheduler_type: cosine
+- lr_scheduler_warmup_steps: 100
+- num_epochs: 1.0
+
+### Training results
+
+
+
+### Framework versions
+
+- PEFT 0.17.1
+- Transformers 4.57.1
+- Pytorch 2.10.0+cu128
+- Datasets 4.0.0
+- Tokenizers 0.22.2

mistral7b_kto_tv_rho0.001/adapter_config.json

@@ -0,0 +1,42 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "mistralai/Mistral-7B-v0.1",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "up_proj",
+    "k_proj",
+    "gate_proj",
+    "o_proj",
+    "v_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_rho0.001/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a47706597e696b2090aa243a0021b23a591fc92c657b0a9be998e425f5b2427
+size 42002584

mistral7b_kto_tv_rho0.001/all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.49831690745694296,
+    "train_runtime": 3155.7241,
+    "train_samples_per_second": 31.688,
+    "train_steps_per_second": 0.062
+}

mistral7b_kto_tv_rho0.001/chat_template.jinja

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

mistral7b_kto_tv_rho0.001/checkpoint-196/README.md

@@ -0,0 +1,208 @@
+---
+base_model: mistralai/Mistral-7B-v0.1
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:mistralai/Mistral-7B-v0.1
+- llama-factory
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.17.1

mistral7b_kto_tv_rho0.001/checkpoint-196/adapter_config.json

@@ -0,0 +1,42 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "mistralai/Mistral-7B-v0.1",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "up_proj",
+    "k_proj",
+    "gate_proj",
+    "o_proj",
+    "v_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_rho0.001/checkpoint-196/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a47706597e696b2090aa243a0021b23a591fc92c657b0a9be998e425f5b2427
+size 42002584

mistral7b_kto_tv_rho0.001/checkpoint-196/chat_template.jinja

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

mistral7b_kto_tv_rho0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:542c6bb20f8998b35b98eb202985e6ea0f811e1c643a640dfb3982c1d4221a5e
+size 31467333

mistral7b_kto_tv_rho0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2847103a14f79afdc85164303539907d1a9b33491446b7135f54d7beb3b0117d
+size 31467333

mistral7b_kto_tv_rho0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5271a3466e4bb3c844c5e787e4e27342c299cfa923d24fbbaec75244054341d
+size 31467333

mistral7b_kto_tv_rho0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f52e3ec5a22cb7d0225e950b4e32de9eb7e284a8d9a97cc182a313ce25c0da05
+size 31467397

mistral7b_kto_tv_rho0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f097fee48f5e359112df633b2a47e546336277d5f827e5b0d1849e38b213ff5
+size 31467397

mistral7b_kto_tv_rho0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ede74559f19fddc5dc441dd73f573f86587471d411c5d120ff098bfc7e737472
+size 31467397

mistral7b_kto_tv_rho0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:954b49a7b297beea0c7817751ce6d68f69bc3ebeb71d87b0a41036ac5e7d95bf
+size 31467397

mistral7b_kto_tv_rho0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:806b1d3730a47951d314c3077ef318a390b9d29a5641f6e997f3bc503f41e78b
+size 31467397

mistral7b_kto_tv_rho0.001/checkpoint-196/global_step196/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:536c7370032937eb45212bc83fdac85ee04c48a19063a2090bc97b019cae1d74
+size 42288409

mistral7b_kto_tv_rho0.001/checkpoint-196/latest

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

mistral7b_kto_tv_rho0.001/checkpoint-196/rng_state_0.pth

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

mistral7b_kto_tv_rho0.001/checkpoint-196/rng_state_1.pth

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

mistral7b_kto_tv_rho0.001/checkpoint-196/rng_state_2.pth

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

mistral7b_kto_tv_rho0.001/checkpoint-196/rng_state_3.pth

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

mistral7b_kto_tv_rho0.001/checkpoint-196/rng_state_4.pth

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

mistral7b_kto_tv_rho0.001/checkpoint-196/rng_state_5.pth

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

mistral7b_kto_tv_rho0.001/checkpoint-196/rng_state_6.pth

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

mistral7b_kto_tv_rho0.001/checkpoint-196/rng_state_7.pth

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

mistral7b_kto_tv_rho0.001/checkpoint-196/scheduler.pt

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

mistral7b_kto_tv_rho0.001/checkpoint-196/special_tokens_map.json

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

mistral7b_kto_tv_rho0.001/checkpoint-196/tokenizer.model

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

mistral7b_kto_tv_rho0.001/checkpoint-196/tokenizer_config.json

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

mistral7b_kto_tv_rho0.001/checkpoint-196/trainer_state.json

@@ -0,0 +1,965 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 196,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.025591810620601407,
+      "grad_norm": 0.5810207724571228,
+      "kl": 0.006213414948433638,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59713886.315789476,
+      "logits/rejected": -50416222.47619048,
+      "logps/chosen": -439.0241056743421,
+      "logps/rejected": -198.41524832589286,
+      "loss": 0.5003,
+      "loss/base_kto": 0.4996565878391266,
+      "metrics/advantage_var": 0.04205384477972984,
+      "regularization/advantage_range": 0.39461883902549744,
+      "regularization/tv": 0.0003946188371628523,
+      "rewards/chosen": 0.0021508518410356422,
+      "rewards/margins": 0.0027365540889532944,
+      "rewards/rejected": -0.0005857022479176521,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.6166413426399231,
+      "kl": 0.019698908552527428,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59302034.28571428,
+      "logits/rejected": -50026792.09638554,
+      "logps/chosen": -429.94201501623377,
+      "logps/rejected": -192.86311652861446,
+      "loss": 0.5008,
+      "loss/base_kto": 0.5001104474067688,
+      "metrics/advantage_var": 0.06644176691770554,
+      "regularization/advantage_range": 0.6011344790458679,
+      "regularization/tv": 0.0006011345540173352,
+      "rewards/chosen": -0.0003495565676069879,
+      "rewards/margins": -0.0008308161839905749,
+      "rewards/rejected": 0.000481259616383587,
+      "stability/repetition_rate_mean": 0.3668472170829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.2646484375,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.6122223734855652,
+      "kl": 0.013860845938324928,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56237422.15757576,
+      "logits/rejected": -51927885.62580645,
+      "logps/chosen": -404.140625,
+      "logps/rejected": -200.65655241935485,
+      "loss": 0.5005,
+      "loss/base_kto": 0.5005344748497009,
+      "metrics/advantage_var": 0.06271388381719589,
+      "regularization/advantage_range": 0.579309344291687,
+      "regularization/tv": 0.0005793093587271869,
+      "rewards/chosen": -0.0009836391066059922,
+      "rewards/margins": -0.00473424715089192,
+      "rewards/rejected": 0.003750608044285928,
+      "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": 3.76171875,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.6354861259460449,
+      "kl": 0.016369277611374855,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51669683.2,
+      "logits/rejected": -48070534.4,
+      "logps/chosen": -343.7012939453125,
+      "logps/rejected": -225.653271484375,
+      "loss": 0.5008,
+      "loss/base_kto": 0.5001991391181946,
+      "metrics/advantage_var": 0.06498073786497116,
+      "regularization/advantage_range": 0.5812579989433289,
+      "regularization/tv": 0.0005812579765915871,
+      "rewards/chosen": -0.00045777419582009314,
+      "rewards/margins": -0.0014659445732831955,
+      "rewards/rejected": 0.0010081703774631023,
+      "stability/repetition_rate_mean": 0.2040860652923584,
+      "stability/response_length_mean": 42.125,
+      "stability/response_length_std": 27.01553726196289,
+      "stability/response_length_var": 729.8392944335938,
+      "stability/token_entropy_mean": 4.44921875,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.5954239368438721,
+      "kl": 0.019973892718553543,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56468356.4137931,
+      "logits/rejected": -52007339.154285714,
+      "logps/chosen": -378.46373922413795,
+      "logps/rejected": -202.78017857142856,
+      "loss": 0.5008,
+      "loss/base_kto": 0.4998336732387543,
+      "metrics/advantage_var": 0.060091253370046616,
+      "regularization/advantage_range": 0.5770506262779236,
+      "regularization/tv": 0.0005770506686531007,
+      "rewards/chosen": 0.002771485262903674,
+      "rewards/margins": 0.0015031191985595401,
+      "rewards/rejected": 0.0012683660643441337,
+      "stability/repetition_rate_mean": 0.18011869490146637,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 27.66121482849121,
+      "stability/response_length_var": 765.1428833007812,
+      "stability/token_entropy_mean": 4.8046875,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.5939051508903503,
+      "kl": 0.018949225544929504,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57724185.77181208,
+      "logits/rejected": -51514062.596491225,
+      "logps/chosen": -468.5343435402685,
+      "logps/rejected": -198.2335754751462,
+      "loss": 0.5006,
+      "loss/base_kto": 0.4996163845062256,
+      "metrics/advantage_var": 0.06395137310028076,
+      "regularization/advantage_range": 0.5929219126701355,
+      "regularization/tv": 0.0005929219187237322,
+      "rewards/chosen": 0.004691862419947681,
+      "rewards/margins": 0.0032610034258597036,
+      "rewards/rejected": 0.0014308589940879776,
+      "stability/repetition_rate_mean": 0.185546875,
+      "stability/response_length_mean": 48.375,
+      "stability/response_length_std": 28.93311309814453,
+      "stability/response_length_var": 837.125,
+      "stability/token_entropy_mean": 4.5625,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6523953676223755,
+      "kl": 0.023824213072657585,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57645127.29113924,
+      "logits/rejected": -53893480.2962963,
+      "logps/chosen": -385.89912974683546,
+      "logps/rejected": -195.09599247685185,
+      "loss": 0.5004,
+      "loss/base_kto": 0.49997439980506897,
+      "metrics/advantage_var": 0.06881736218929291,
+      "regularization/advantage_range": 0.5582152605056763,
+      "regularization/tv": 0.0005582152516581118,
+      "rewards/chosen": 0.003739585604848741,
+      "rewards/margins": 0.0006903532632236689,
+      "rewards/rejected": 0.0030492323416250722,
+      "stability/repetition_rate_mean": 0.3566158413887024,
+      "stability/response_length_mean": 53.375,
+      "stability/response_length_std": 21.14533042907715,
+      "stability/response_length_var": 447.125,
+      "stability/token_entropy_mean": 5.337890625,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.57895427942276,
+      "kl": 0.03129922226071358,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56629888.0,
+      "logits/rejected": -50223590.4,
+      "logps/chosen": -421.85048828125,
+      "logps/rejected": -219.295849609375,
+      "loss": 0.5004,
+      "loss/base_kto": 0.49950963258743286,
+      "metrics/advantage_var": 0.0760921984910965,
+      "regularization/advantage_range": 0.6682614088058472,
+      "regularization/tv": 0.0006682613748125732,
+      "rewards/chosen": 0.005213038995862007,
+      "rewards/margins": 0.0038720893673598765,
+      "rewards/rejected": 0.0013409496285021305,
+      "stability/repetition_rate_mean": 0.142578125,
+      "stability/response_length_mean": 33.125,
+      "stability/response_length_std": 33.01704025268555,
+      "stability/response_length_var": 1090.125,
+      "stability/token_entropy_mean": 4.97265625,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.6361792683601379,
+      "kl": 0.041447144001722336,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52700613.58389262,
+      "logits/rejected": -50780142.03508772,
+      "logps/chosen": -352.2842648909396,
+      "logps/rejected": -243.8423793859649,
+      "loss": 0.5004,
+      "loss/base_kto": 0.49984151124954224,
+      "metrics/advantage_var": 0.07464335113763809,
+      "regularization/advantage_range": 0.6375295519828796,
+      "regularization/tv": 0.0006375295924954116,
+      "rewards/chosen": 0.004419918028300241,
+      "rewards/margins": 0.001242195384609741,
+      "rewards/rejected": 0.0031777226436905,
+      "stability/repetition_rate_mean": 0.0935424193739891,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.87324333190918,
+      "stability/response_length_var": 892.4107055664062,
+      "stability/token_entropy_mean": 4.755859375,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.5714625120162964,
+      "kl": 0.05188734456896782,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54044507.060402684,
+      "logits/rejected": -48291307.04093567,
+      "logps/chosen": -375.12659920302013,
+      "logps/rejected": -180.181914747807,
+      "loss": 0.5004,
+      "loss/base_kto": 0.49978047609329224,
+      "metrics/advantage_var": 0.05654274299740791,
+      "regularization/advantage_range": 0.5456675887107849,
+      "regularization/tv": 0.0005456676008179784,
+      "rewards/chosen": 0.00725229154497185,
+      "rewards/margins": 0.0013820036665003416,
+      "rewards/rejected": 0.005870287878471508,
+      "stability/repetition_rate_mean": 0.4206075370311737,
+      "stability/response_length_mean": 55.375,
+      "stability/response_length_std": 16.361431121826172,
+      "stability/response_length_var": 267.6964416503906,
+      "stability/token_entropy_mean": 3.810546875,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.6483849883079529,
+      "kl": 0.0480094812810421,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56957481.34161491,
+      "logits/rejected": -49573807.49685535,
+      "logps/chosen": -344.50655085403724,
+      "logps/rejected": -230.41897602201257,
+      "loss": 0.5002,
+      "loss/base_kto": 0.5002776980400085,
+      "metrics/advantage_var": 0.06638753414154053,
+      "regularization/advantage_range": 0.617785632610321,
+      "regularization/tv": 0.0006177856703288853,
+      "rewards/chosen": 0.0005464234522410802,
+      "rewards/margins": -0.0023582645851753896,
+      "rewards/rejected": 0.00290468803741647,
+      "stability/repetition_rate_mean": 0.10418051481246948,
+      "stability/response_length_mean": 22.375,
+      "stability/response_length_std": 30.236862182617188,
+      "stability/response_length_var": 914.2678833007812,
+      "stability/token_entropy_mean": 4.3671875,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.5852304697036743,
+      "kl": 0.04956696555018425,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52675124.75151515,
+      "logits/rejected": -47822603.561290324,
+      "logps/chosen": -359.6289772727273,
+      "logps/rejected": -178.1511214717742,
+      "loss": 0.5004,
+      "loss/base_kto": 0.49980688095092773,
+      "metrics/advantage_var": 0.05555450916290283,
+      "regularization/advantage_range": 0.5291790962219238,
+      "regularization/tv": 0.0005291791749186814,
+      "rewards/chosen": 0.009183568665475556,
+      "rewards/margins": 0.0016036406057437962,
+      "rewards/rejected": 0.00757992805973176,
+      "stability/repetition_rate_mean": 0.21226298809051514,
+      "stability/response_length_mean": 51.125,
+      "stability/response_length_std": 21.90523338317871,
+      "stability/response_length_var": 479.83929443359375,
+      "stability/token_entropy_mean": 5.322265625,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.5865442752838135,
+      "kl": 0.07576815038919449,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57571228.90322581,
+      "logits/rejected": -47274263.27272727,
+      "logps/chosen": -381.83079637096773,
+      "logps/rejected": -183.50297111742424,
+      "loss": 0.5002,
+      "loss/base_kto": 0.49959665536880493,
+      "metrics/advantage_var": 0.06296278536319733,
+      "regularization/advantage_range": 0.5749912261962891,
+      "regularization/tv": 0.0005749912816099823,
+      "rewards/chosen": 0.011598679327195691,
+      "rewards/margins": 0.0036599966554464614,
+      "rewards/rejected": 0.00793868267174923,
+      "stability/repetition_rate_mean": 0.18097305297851562,
+      "stability/response_length_mean": 17.875,
+      "stability/response_length_std": 23.485177993774414,
+      "stability/response_length_var": 551.5535888671875,
+      "stability/token_entropy_mean": 5.66796875,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5472328662872314,
+      "kl": 0.07580217719078064,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52759284.438709676,
+      "logits/rejected": -49026904.43636364,
+      "logps/chosen": -379.0366179435484,
+      "logps/rejected": -220.35340909090908,
+      "loss": 0.4998,
+      "loss/base_kto": 0.4997020363807678,
+      "metrics/advantage_var": 0.07085214555263519,
+      "regularization/advantage_range": 0.6015222668647766,
+      "regularization/tv": 0.0006015223334543407,
+      "rewards/chosen": 0.009184343584122196,
+      "rewards/margins": 0.002310059706836152,
+      "rewards/rejected": 0.006874283877286044,
+      "stability/repetition_rate_mean": 0.1796875,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.68359375,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.5962035655975342,
+      "kl": 0.09119434654712677,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49505473.72972973,
+      "logits/rejected": -47898463.255813956,
+      "logps/chosen": -314.1518950591216,
+      "logps/rejected": -228.81252271075581,
+      "loss": 0.5,
+      "loss/base_kto": 0.4993656575679779,
+      "metrics/advantage_var": 0.07352752983570099,
+      "regularization/advantage_range": 0.6583485007286072,
+      "regularization/tv": 0.0006583484937436879,
+      "rewards/chosen": 0.011890417820698506,
+      "rewards/margins": 0.004657435761840744,
+      "rewards/rejected": 0.007232982058857762,
+      "stability/repetition_rate_mean": 0.068359375,
+      "stability/response_length_mean": 24.25,
+      "stability/response_length_std": 32.92306900024414,
+      "stability/response_length_var": 1083.9285888671875,
+      "stability/token_entropy_mean": 1.91650390625,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.5844407081604004,
+      "kl": 0.10768461227416992,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58620202.38410596,
+      "logits/rejected": -53447098.31952663,
+      "logps/chosen": -403.8815190397351,
+      "logps/rejected": -200.99170210798817,
+      "loss": 0.4998,
+      "loss/base_kto": 0.49889513850212097,
+      "metrics/advantage_var": 0.07821118086576462,
+      "regularization/advantage_range": 0.6350656747817993,
+      "regularization/tv": 0.0006350657204166055,
+      "rewards/chosen": 0.014854513256755096,
+      "rewards/margins": 0.008444882989177331,
+      "rewards/rejected": 0.006409630267577764,
+      "stability/repetition_rate_mean": 0.19479167461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.203125,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.5908569097518921,
+      "kl": 0.10215399414300919,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51473758.14193548,
+      "logits/rejected": -47141112.24242424,
+      "logps/chosen": -358.18568548387094,
+      "logps/rejected": -230.36178977272726,
+      "loss": 0.4995,
+      "loss/base_kto": 0.49893245100975037,
+      "metrics/advantage_var": 0.07610619813203812,
+      "regularization/advantage_range": 0.6356574892997742,
+      "regularization/tv": 0.0006356575759127736,
+      "rewards/chosen": 0.014681977610434256,
+      "rewards/margins": 0.00838077613685138,
+      "rewards/rejected": 0.0063012014735828745,
+      "stability/repetition_rate_mean": 0.095703125,
+      "stability/response_length_mean": 32.5,
+      "stability/response_length_std": 33.67915725708008,
+      "stability/response_length_var": 1134.2857666015625,
+      "stability/token_entropy_mean": 4.91015625,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.5829842686653137,
+      "kl": 0.14038605988025665,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58412242.05128205,
+      "logits/rejected": -50579849.36585366,
+      "logps/chosen": -380.6251001602564,
+      "logps/rejected": -187.1724823742378,
+      "loss": 0.4994,
+      "loss/base_kto": 0.49847766757011414,
+      "metrics/advantage_var": 0.09148811548948288,
+      "regularization/advantage_range": 0.7223256826400757,
+      "regularization/tv": 0.0007223256980068982,
+      "rewards/chosen": 0.016429907236343775,
+      "rewards/margins": 0.01099776524614736,
+      "rewards/rejected": 0.005432141990196414,
+      "stability/repetition_rate_mean": 0.2647964060306549,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 29.132394790649414,
+      "stability/response_length_var": 848.6964111328125,
+      "stability/token_entropy_mean": 3.72021484375,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.5847001671791077,
+      "kl": 0.12943434715270996,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50392932.4556962,
+      "logits/rejected": -48968514.37037037,
+      "logps/chosen": -314.8338607594937,
+      "logps/rejected": -207.10426311728395,
+      "loss": 0.4994,
+      "loss/base_kto": 0.49842140078544617,
+      "metrics/advantage_var": 0.09590301662683487,
+      "regularization/advantage_range": 0.7024412155151367,
+      "regularization/tv": 0.0007024412043392658,
+      "rewards/chosen": 0.01595103287998634,
+      "rewards/margins": 0.012005401870425297,
+      "rewards/rejected": 0.0039456310095610445,
+      "stability/repetition_rate_mean": 0.20155483484268188,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 31.1264705657959,
+      "stability/response_length_var": 968.8571166992188,
+      "stability/token_entropy_mean": 3.685546875,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.5611183643341064,
+      "kl": 0.15796113014221191,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54706445.83783784,
+      "logits/rejected": -50905885.76744186,
+      "logps/chosen": -359.63999155405406,
+      "logps/rejected": -216.46538880813952,
+      "loss": 0.4991,
+      "loss/base_kto": 0.49846044182777405,
+      "metrics/advantage_var": 0.09960363060235977,
+      "regularization/advantage_range": 0.7743232846260071,
+      "regularization/tv": 0.0007743232999928296,
+      "rewards/chosen": 0.023018472903483623,
+      "rewards/margins": 0.011669156650265083,
+      "rewards/rejected": 0.01134931625321854,
+      "stability/repetition_rate_mean": 0.23383361101150513,
+      "stability/response_length_mean": 37.5,
+      "stability/response_length_std": 29.640705108642578,
+      "stability/response_length_var": 878.5714111328125,
+      "stability/token_entropy_mean": 4.150390625,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6100413799285889,
+      "kl": 0.1611875742673874,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59367004.530120485,
+      "logits/rejected": -53783964.25974026,
+      "logps/chosen": -399.5828313253012,
+      "logps/rejected": -209.8855265827922,
+      "loss": 0.4989,
+      "loss/base_kto": 0.49745798110961914,
+      "metrics/advantage_var": 0.18634693324565887,
+      "regularization/advantage_range": 1.0710378885269165,
+      "regularization/tv": 0.001071037957444787,
+      "rewards/chosen": 0.0319610503782709,
+      "rewards/margins": 0.020131756210715127,
+      "rewards/rejected": 0.011829294167555772,
+      "stability/repetition_rate_mean": 0.2053571343421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.41796875,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.5926874876022339,
+      "kl": 0.19908857345581055,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57380000.38554217,
+      "logits/rejected": -53610043.84415584,
+      "logps/chosen": -378.29080384036143,
+      "logps/rejected": -205.96319500811688,
+      "loss": 0.4988,
+      "loss/base_kto": 0.49805688858032227,
+      "metrics/advantage_var": 0.14875562489032745,
+      "regularization/advantage_range": 0.9510820508003235,
+      "regularization/tv": 0.0009510820964351296,
+      "rewards/chosen": 0.02724501885563494,
+      "rewards/margins": 0.01475072959987346,
+      "rewards/rejected": 0.01249428925576148,
+      "stability/repetition_rate_mean": 0.3234747052192688,
+      "stability/response_length_mean": 32.25,
+      "stability/response_length_std": 30.541540145874023,
+      "stability/response_length_var": 932.7857055664062,
+      "stability/token_entropy_mean": 4.4296875,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.5929023623466492,
+      "kl": 0.2150437831878662,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60607008.405063294,
+      "logits/rejected": -53416820.938271604,
+      "logps/chosen": -364.25111253955697,
+      "logps/rejected": -201.07520736882717,
+      "loss": 0.4986,
+      "loss/base_kto": 0.4972267150878906,
+      "metrics/advantage_var": 0.17768144607543945,
+      "regularization/advantage_range": 1.046006202697754,
+      "regularization/tv": 0.0010460062185302377,
+      "rewards/chosen": 0.03414275072797944,
+      "rewards/margins": 0.022008340011110225,
+      "rewards/rejected": 0.012134410716869213,
+      "stability/repetition_rate_mean": 0.3148462176322937,
+      "stability/response_length_mean": 47.625,
+      "stability/response_length_std": 27.21311378479004,
+      "stability/response_length_var": 740.5535888671875,
+      "stability/token_entropy_mean": 4.78125,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.5756809711456299,
+      "kl": 0.221540167927742,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57751137.04575163,
+      "logits/rejected": -54370886.51497006,
+      "logps/chosen": -383.87048100490193,
+      "logps/rejected": -212.07943488023952,
+      "loss": 0.4981,
+      "loss/base_kto": 0.496986448764801,
+      "metrics/advantage_var": 0.20863568782806396,
+      "regularization/advantage_range": 1.190858244895935,
+      "regularization/tv": 0.0011908583110198379,
+      "rewards/chosen": 0.03744382172628166,
+      "rewards/margins": 0.023750938765853644,
+      "rewards/rejected": 0.01369288296042802,
+      "stability/repetition_rate_mean": 0.1762429028749466,
+      "stability/response_length_mean": 27.875,
+      "stability/response_length_std": 30.31943130493164,
+      "stability/response_length_var": 919.2678833007812,
+      "stability/token_entropy_mean": 4.646484375,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.5690529942512512,
+      "kl": 0.2525067925453186,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60704396.2739726,
+      "logits/rejected": -55673891.31034483,
+      "logps/chosen": -417.02595248287673,
+      "logps/rejected": -216.25145923132183,
+      "loss": 0.4983,
+      "loss/base_kto": 0.4970117211341858,
+      "metrics/advantage_var": 0.32070493698120117,
+      "regularization/advantage_range": 1.4626210927963257,
+      "regularization/tv": 0.0014626210322603583,
+      "rewards/chosen": 0.035337840041069137,
+      "rewards/margins": 0.023764848746646767,
+      "rewards/rejected": 0.01157299129442237,
+      "stability/repetition_rate_mean": 0.15937499701976776,
+      "stability/response_length_mean": 15.875,
+      "stability/response_length_std": 24.584766387939453,
+      "stability/response_length_var": 604.4107055664062,
+      "stability/token_entropy_mean": 3.5,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.5786094069480896,
+      "kl": 0.2731947600841522,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57406563.09677419,
+      "logits/rejected": -53861922.13333333,
+      "logps/chosen": -392.2603326612903,
+      "logps/rejected": -183.81022727272727,
+      "loss": 0.4979,
+      "loss/base_kto": 0.4962916374206543,
+      "metrics/advantage_var": 0.4037338197231293,
+      "regularization/advantage_range": 1.604854941368103,
+      "regularization/tv": 0.0016048549441620708,
+      "rewards/chosen": 0.039759131400815904,
+      "rewards/margins": 0.028890261995011646,
+      "rewards/rejected": 0.010868869405804258,
+      "stability/repetition_rate_mean": 0.23874999582767487,
+      "stability/response_length_mean": 43.25,
+      "stability/response_length_std": 29.620214462280273,
+      "stability/response_length_var": 877.3571166992188,
+      "stability/token_entropy_mean": 3.6943359375,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5664066672325134,
+      "kl": 0.3012661039829254,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59866928.68711656,
+      "logits/rejected": -55031821.04458599,
+      "logps/chosen": -401.296227952454,
+      "logps/rejected": -247.69799960191082,
+      "loss": 0.4972,
+      "loss/base_kto": 0.49553966522216797,
+      "metrics/advantage_var": 0.4732483923435211,
+      "regularization/advantage_range": 1.8137600421905518,
+      "regularization/tv": 0.0018137602601200342,
+      "rewards/chosen": 0.051993937580132044,
+      "rewards/margins": 0.03472005898450045,
+      "rewards/rejected": 0.017273878595631592,
+      "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": 4.078125,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.6262583136558533,
+      "kl": 0.2736550271511078,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57111142.4,
+      "logits/rejected": -50093432.56774194,
+      "logps/chosen": -366.6800899621212,
+      "logps/rejected": -190.03079637096775,
+      "loss": 0.497,
+      "loss/base_kto": 0.49387773871421814,
+      "metrics/advantage_var": 0.812842845916748,
+      "regularization/advantage_range": 2.1453397274017334,
+      "regularization/tv": 0.0021453395020216703,
+      "rewards/chosen": 0.05096650557084517,
+      "rewards/margins": 0.050598360366485684,
+      "rewards/rejected": 0.00036814520435948523,
+      "stability/repetition_rate_mean": 0.3053385317325592,
+      "stability/response_length_mean": 52.75,
+      "stability/response_length_std": 22.72663688659668,
+      "stability/response_length_var": 516.5,
+      "stability/token_entropy_mean": 5.111328125,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.6019709706306458,
+      "kl": 0.3225500285625458,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51165295.03614458,
+      "logits/rejected": -51708233.14285714,
+      "logps/chosen": -329.8277249623494,
+      "logps/rejected": -199.37265371347402,
+      "loss": 0.497,
+      "loss/base_kto": 0.49590760469436646,
+      "metrics/advantage_var": 0.36499539017677307,
+      "regularization/advantage_range": 1.656163215637207,
+      "regularization/tv": 0.001656163134612143,
+      "rewards/chosen": 0.04440879821777344,
+      "rewards/margins": 0.03288166786169077,
+      "rewards/rejected": 0.011527130356082669,
+      "stability/repetition_rate_mean": 0.24838952720165253,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.73423957824707,
+      "stability/response_length_var": 884.125,
+      "stability/token_entropy_mean": 5.26953125,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.6047873497009277,
+      "kl": 0.3542978763580322,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55434553.22352941,
+      "logits/rejected": -50214345.38666666,
+      "logps/chosen": -397.2438419117647,
+      "logps/rejected": -207.591796875,
+      "loss": 0.4967,
+      "loss/base_kto": 0.49276694655418396,
+      "metrics/advantage_var": 0.5820550322532654,
+      "regularization/advantage_range": 2.0140323638916016,
+      "regularization/tv": 0.002014032332226634,
+      "rewards/chosen": 0.06905886706183939,
+      "rewards/margins": 0.055644914870168656,
+      "rewards/rejected": 0.013413952191670736,
+      "stability/repetition_rate_mean": 0.23066024482250214,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.193626403808594,
+      "stability/response_length_var": 852.2678833007812,
+      "stability/token_entropy_mean": 4.59375,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.561375617980957,
+      "kl": 0.351535439491272,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54049448.65882353,
+      "logits/rejected": -49388888.74666667,
+      "logps/chosen": -360.47366727941176,
+      "logps/rejected": -191.245,
+      "loss": 0.4968,
+      "loss/base_kto": 0.4943511188030243,
+      "metrics/advantage_var": 0.7565718293190002,
+      "regularization/advantage_range": 2.201819658279419,
+      "regularization/tv": 0.002201819559559226,
+      "rewards/chosen": 0.051560289719525504,
+      "rewards/margins": 0.04549936201058182,
+      "rewards/rejected": 0.006060927708943685,
+      "stability/repetition_rate_mean": 0.3171784281730652,
+      "stability/response_length_mean": 61.375,
+      "stability/response_length_std": 7.424621105194092,
+      "stability/response_length_var": 55.125,
+      "stability/token_entropy_mean": 4.544921875,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.5659539103507996,
+      "kl": 0.38903898000717163,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55283943.61904762,
+      "logits/rejected": -49011105.684210524,
+      "logps/chosen": -416.15206473214283,
+      "logps/rejected": -186.7262541118421,
+      "loss": 0.4963,
+      "loss/base_kto": 0.4927336871623993,
+      "metrics/advantage_var": 0.7581729888916016,
+      "regularization/advantage_range": 2.441723585128784,
+      "regularization/tv": 0.0024417235981673002,
+      "rewards/chosen": 0.07432699771154494,
+      "rewards/margins": 0.05721738717908548,
+      "rewards/rejected": 0.01710961053245946,
+      "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": 4.3837890625,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.6592298150062561,
+      "kl": 0.4146839678287506,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54472480.36781609,
+      "logits/rejected": -49860436.16438356,
+      "logps/chosen": -370.26210039511494,
+      "logps/rejected": -212.42588827054794,
+      "loss": 0.4954,
+      "loss/base_kto": 0.4939287304878235,
+      "metrics/advantage_var": 0.6743358969688416,
+      "regularization/advantage_range": 2.269331455230713,
+      "regularization/tv": 0.002269331831485033,
+      "rewards/chosen": 0.06546377861636808,
+      "rewards/margins": 0.048841037687927205,
+      "rewards/rejected": 0.016622740928440877,
+      "stability/repetition_rate_mean": 0.2411106526851654,
+      "stability/response_length_mean": 22.5,
+      "stability/response_length_std": 28.33977508544922,
+      "stability/response_length_var": 803.1428833007812,
+      "stability/token_entropy_mean": 4.87109375,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.564885139465332,
+      "kl": 0.5150820016860962,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59181152.94674556,
+      "logits/rejected": -52670104.58278146,
+      "logps/chosen": -440.4078679733728,
+      "logps/rejected": -190.99668874172184,
+      "loss": 0.4948,
+      "loss/base_kto": 0.49169474840164185,
+      "metrics/advantage_var": 0.832883358001709,
+      "regularization/advantage_range": 2.5218818187713623,
+      "regularization/tv": 0.0025218818336725235,
+      "rewards/chosen": 0.0933977894529083,
+      "rewards/margins": 0.06728294196652601,
+      "rewards/rejected": 0.02611484748638229,
+      "stability/repetition_rate_mean": 0.2678554654121399,
+      "stability/response_length_mean": 37.25,
+      "stability/response_length_std": 30.92733383178711,
+      "stability/response_length_var": 956.5,
+      "stability/token_entropy_mean": 3.80126953125,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.6160000562667847,
+      "kl": 0.5376589894294739,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56722012.530120485,
+      "logits/rejected": -48136594.28571428,
+      "logps/chosen": -355.58876129518075,
+      "logps/rejected": -194.6046570616883,
+      "loss": 0.4945,
+      "loss/base_kto": 0.4918275475502014,
+      "metrics/advantage_var": 1.3826684951782227,
+      "regularization/advantage_range": 3.0055034160614014,
+      "regularization/tv": 0.0030055036768317223,
+      "rewards/chosen": 0.08484269911984363,
+      "rewards/margins": 0.06455535258957336,
+      "rewards/rejected": 0.020287346530270266,
+      "stability/repetition_rate_mean": 0.2454427182674408,
+      "stability/response_length_mean": 39.75,
+      "stability/response_length_std": 27.473363876342773,
+      "stability/response_length_var": 754.7857055664062,
+      "stability/token_entropy_mean": 4.73046875,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.578485906124115,
+      "kl": 0.5820572972297668,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52728517.88957055,
+      "logits/rejected": -48089241.27388535,
+      "logps/chosen": -375.1566334355828,
+      "logps/rejected": -210.98556926751593,
+      "loss": 0.4938,
+      "loss/base_kto": 0.49145957827568054,
+      "metrics/advantage_var": 1.5378645658493042,
+      "regularization/advantage_range": 3.2784626483917236,
+      "regularization/tv": 0.0032784626819193363,
+      "rewards/chosen": 0.0989208455466054,
+      "rewards/margins": 0.06765583580624973,
+      "rewards/rejected": 0.03126500974035567,
+      "stability/repetition_rate_mean": 0.26171875,
+      "stability/response_length_mean": 40.375,
+      "stability/response_length_std": 32.61874008178711,
+      "stability/response_length_var": 1063.982177734375,
+      "stability/token_entropy_mean": 3.091796875,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.5877865552902222,
+      "kl": 0.6085312962532043,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54704355.55555555,
+      "logits/rejected": -49614323.037974685,
+      "logps/chosen": -311.0955825617284,
+      "logps/rejected": -196.88338113132912,
+      "loss": 0.4943,
+      "loss/base_kto": 0.49103212356567383,
+      "metrics/advantage_var": 2.168581485748291,
+      "regularization/advantage_range": 3.4313366413116455,
+      "regularization/tv": 0.003431336721405387,
+      "rewards/chosen": 0.09057903289794922,
+      "rewards/margins": 0.075000848951219,
+      "rewards/rejected": 0.015578183946730216,
+      "stability/repetition_rate_mean": 0.14541903138160706,
+      "stability/response_length_mean": 26.25,
+      "stability/response_length_std": 31.45404815673828,
+      "stability/response_length_var": 989.3571166992188,
+      "stability/token_entropy_mean": 3.7890625,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.6331533193588257,
+      "kl": 0.6512568593025208,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53173057.91616766,
+      "logits/rejected": -49808035.97385621,
+      "logps/chosen": -391.2273811751497,
+      "logps/rejected": -210.86637050653596,
+      "loss": 0.4933,
+      "loss/base_kto": 0.4893794655799866,
+      "metrics/advantage_var": 2.1304852962493896,
+      "regularization/advantage_range": 3.9502923488616943,
+      "regularization/tv": 0.003950292710214853,
+      "rewards/chosen": 0.11779353433026525,
+      "rewards/margins": 0.08457047255051998,
+      "rewards/rejected": 0.03322306177974527,
+      "stability/repetition_rate_mean": 0.1901041716337204,
+      "stability/response_length_mean": 17.625,
+      "stability/response_length_std": 28.680194854736328,
+      "stability/response_length_var": 822.5535888671875,
+      "stability/token_entropy_mean": 4.8427734375,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.5744519829750061,
+      "kl": 0.6606561541557312,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60185910.5032258,
+      "logits/rejected": -53942718.836363636,
+      "logps/chosen": -391.6687247983871,
+      "logps/rejected": -197.56677320075758,
+      "loss": 0.4937,
+      "loss/base_kto": 0.4902775287628174,
+      "metrics/advantage_var": 1.463286280632019,
+      "regularization/advantage_range": 3.479656219482422,
+      "regularization/tv": 0.003479656297713518,
+      "rewards/chosen": 0.11162065075289819,
+      "rewards/margins": 0.07636482559457669,
+      "rewards/rejected": 0.0352558251583215,
+      "stability/repetition_rate_mean": 0.42074424028396606,
+      "stability/response_length_mean": 52.875,
+      "stability/response_length_std": 22.850053787231445,
+      "stability/response_length_var": 522.125,
+      "stability/token_entropy_mean": 5.18359375,
+      "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_rho0.001/checkpoint-196/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c494ab7cafd29dea773d452695f132b657423e6fff5ae7d83d91a8b8b8e7981
+size 8017

mistral7b_kto_tv_rho0.001/checkpoint-196/zero_to_fp32.py

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

mistral7b_kto_tv_rho0.001/special_tokens_map.json

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

mistral7b_kto_tv_rho0.001/stability_metrics.jsonl

@@ -0,0 +1,38 @@
+{"stability/response_length_mean": 46.125, "stability/response_length_std": 27.231481552124023, "stability/response_length_var": 741.5535888671875, "stability/token_entropy_mean": 3.2646484375, "stability/repetition_rate_mean": 0.3668472170829773, "step": 10}
+{"stability/response_length_mean": 28.5, "stability/response_length_std": 30.10457992553711, "stability/response_length_var": 906.2857055664062, "stability/token_entropy_mean": 3.76171875, "stability/repetition_rate_mean": 0.2763020694255829, "step": 15}
+{"stability/response_length_mean": 42.125, "stability/response_length_std": 27.01553726196289, "stability/response_length_var": 729.8392944335938, "stability/token_entropy_mean": 4.44921875, "stability/repetition_rate_mean": 0.2040860652923584, "step": 20}
+{"stability/response_length_mean": 42.0, "stability/response_length_std": 27.66121482849121, "stability/response_length_var": 765.1428833007812, "stability/token_entropy_mean": 4.8046875, "stability/repetition_rate_mean": 0.18011869490146637, "step": 25}
+{"stability/response_length_mean": 48.375, "stability/response_length_std": 28.93311309814453, "stability/response_length_var": 837.125, "stability/token_entropy_mean": 4.5625, "stability/repetition_rate_mean": 0.185546875, "step": 30}
+{"stability/response_length_mean": 53.375, "stability/response_length_std": 21.14533042907715, "stability/response_length_var": 447.125, "stability/token_entropy_mean": 5.337890625, "stability/repetition_rate_mean": 0.3566158413887024, "step": 35}
+{"stability/response_length_mean": 33.125, "stability/response_length_std": 33.01704025268555, "stability/response_length_var": 1090.125, "stability/token_entropy_mean": 4.97265625, "stability/repetition_rate_mean": 0.142578125, "step": 40}
+{"stability/response_length_mean": 28.875, "stability/response_length_std": 29.87324333190918, "stability/response_length_var": 892.4107055664062, "stability/token_entropy_mean": 4.755859375, "stability/repetition_rate_mean": 0.0935424193739891, "step": 45}
+{"stability/response_length_mean": 55.375, "stability/response_length_std": 16.361431121826172, "stability/response_length_var": 267.6964416503906, "stability/token_entropy_mean": 3.810546875, "stability/repetition_rate_mean": 0.4206075370311737, "step": 50}
+{"stability/response_length_mean": 22.375, "stability/response_length_std": 30.236862182617188, "stability/response_length_var": 914.2678833007812, "stability/token_entropy_mean": 4.3671875, "stability/repetition_rate_mean": 0.10418051481246948, "step": 55}
+{"stability/response_length_mean": 51.125, "stability/response_length_std": 21.90523338317871, "stability/response_length_var": 479.83929443359375, "stability/token_entropy_mean": 5.322265625, "stability/repetition_rate_mean": 0.21226298809051514, "step": 60}
+{"stability/response_length_mean": 17.875, "stability/response_length_std": 23.485177993774414, "stability/response_length_var": 551.5535888671875, "stability/token_entropy_mean": 5.66796875, "stability/repetition_rate_mean": 0.18097305297851562, "step": 65}
+{"stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.68359375, "stability/repetition_rate_mean": 0.1796875, "step": 70}
+{"stability/response_length_mean": 24.25, "stability/response_length_std": 32.92306900024414, "stability/response_length_var": 1083.9285888671875, "stability/token_entropy_mean": 1.91650390625, "stability/repetition_rate_mean": 0.068359375, "step": 75}
+{"stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.203125, "stability/repetition_rate_mean": 0.19479167461395264, "step": 80}
+{"stability/response_length_mean": 32.5, "stability/response_length_std": 33.67915725708008, "stability/response_length_var": 1134.2857666015625, "stability/token_entropy_mean": 4.91015625, "stability/repetition_rate_mean": 0.095703125, "step": 85}
+{"stability/response_length_mean": 33.875, "stability/response_length_std": 29.132394790649414, "stability/response_length_var": 848.6964111328125, "stability/token_entropy_mean": 3.72021484375, "stability/repetition_rate_mean": 0.2647964060306549, "step": 90}
+{"stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.685546875, "stability/repetition_rate_mean": 0.20155483484268188, "step": 95}
+{"stability/response_length_mean": 37.5, "stability/response_length_std": 29.640705108642578, "stability/response_length_var": 878.5714111328125, "stability/token_entropy_mean": 4.150390625, "stability/repetition_rate_mean": 0.23383361101150513, "step": 100}
+{"stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.41796875, "stability/repetition_rate_mean": 0.2053571343421936, "step": 105}
+{"stability/response_length_mean": 32.25, "stability/response_length_std": 30.541540145874023, "stability/response_length_var": 932.7857055664062, "stability/token_entropy_mean": 4.4296875, "stability/repetition_rate_mean": 0.3234747052192688, "step": 110}
+{"stability/response_length_mean": 47.625, "stability/response_length_std": 27.21311378479004, "stability/response_length_var": 740.5535888671875, "stability/token_entropy_mean": 4.78125, "stability/repetition_rate_mean": 0.3148462176322937, "step": 115}
+{"stability/response_length_mean": 27.875, "stability/response_length_std": 30.31943130493164, "stability/response_length_var": 919.2678833007812, "stability/token_entropy_mean": 4.646484375, "stability/repetition_rate_mean": 0.1762429028749466, "step": 120}
+{"stability/response_length_mean": 15.875, "stability/response_length_std": 24.584766387939453, "stability/response_length_var": 604.4107055664062, "stability/token_entropy_mean": 3.5, "stability/repetition_rate_mean": 0.15937499701976776, "step": 125}
+{"stability/response_length_mean": 43.25, "stability/response_length_std": 29.620214462280273, "stability/response_length_var": 877.3571166992188, "stability/token_entropy_mean": 3.6943359375, "stability/repetition_rate_mean": 0.23874999582767487, "step": 130}
+{"stability/response_length_mean": 29.25, "stability/response_length_std": 29.936838150024414, "stability/response_length_var": 896.2142944335938, "stability/token_entropy_mean": 4.078125, "stability/repetition_rate_mean": 0.2230902761220932, "step": 135}
+{"stability/response_length_mean": 52.75, "stability/response_length_std": 22.72663688659668, "stability/response_length_var": 516.5, "stability/token_entropy_mean": 5.111328125, "stability/repetition_rate_mean": 0.3053385317325592, "step": 140}
+{"stability/response_length_mean": 28.875, "stability/response_length_std": 29.73423957824707, "stability/response_length_var": 884.125, "stability/token_entropy_mean": 5.26953125, "stability/repetition_rate_mean": 0.24838952720165253, "step": 145}
+{"stability/response_length_mean": 29.375, "stability/response_length_std": 29.193626403808594, "stability/response_length_var": 852.2678833007812, "stability/token_entropy_mean": 4.59375, "stability/repetition_rate_mean": 0.23066024482250214, "step": 150}
+{"stability/response_length_mean": 61.375, "stability/response_length_std": 7.424621105194092, "stability/response_length_var": 55.125, "stability/token_entropy_mean": 4.544921875, "stability/repetition_rate_mean": 0.3171784281730652, "step": 155}
+{"stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.3837890625, "stability/repetition_rate_mean": 0.31046873331069946, "step": 160}
+{"stability/response_length_mean": 22.5, "stability/response_length_std": 28.33977508544922, "stability/response_length_var": 803.1428833007812, "stability/token_entropy_mean": 4.87109375, "stability/repetition_rate_mean": 0.2411106526851654, "step": 165}
+{"stability/response_length_mean": 37.25, "stability/response_length_std": 30.92733383178711, "stability/response_length_var": 956.5, "stability/token_entropy_mean": 3.80126953125, "stability/repetition_rate_mean": 0.2678554654121399, "step": 170}
+{"stability/response_length_mean": 39.75, "stability/response_length_std": 27.473363876342773, "stability/response_length_var": 754.7857055664062, "stability/token_entropy_mean": 4.73046875, "stability/repetition_rate_mean": 0.2454427182674408, "step": 175}
+{"stability/response_length_mean": 40.375, "stability/response_length_std": 32.61874008178711, "stability/response_length_var": 1063.982177734375, "stability/token_entropy_mean": 3.091796875, "stability/repetition_rate_mean": 0.26171875, "step": 180}
+{"stability/response_length_mean": 26.25, "stability/response_length_std": 31.45404815673828, "stability/response_length_var": 989.3571166992188, "stability/token_entropy_mean": 3.7890625, "stability/repetition_rate_mean": 0.14541903138160706, "step": 185}
+{"stability/response_length_mean": 17.625, "stability/response_length_std": 28.680194854736328, "stability/response_length_var": 822.5535888671875, "stability/token_entropy_mean": 4.8427734375, "stability/repetition_rate_mean": 0.1901041716337204, "step": 190}
+{"stability/response_length_mean": 52.875, "stability/response_length_std": 22.850053787231445, "stability/response_length_var": 522.125, "stability/token_entropy_mean": 5.18359375, "stability/repetition_rate_mean": 0.42074424028396606, "step": 195}

mistral7b_kto_tv_rho0.001/tokenizer.model

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

mistral7b_kto_tv_rho0.001/tokenizer_config.json

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

mistral7b_kto_tv_rho0.001/train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.49831690745694296,
+    "train_runtime": 3155.7241,
+    "train_samples_per_second": 31.688,
+    "train_steps_per_second": 0.062
+}

mistral7b_kto_tv_rho0.001/trainer_log.jsonl

@@ -0,0 +1,40 @@
+{"current_steps": 5, "total_steps": 196, "loss": 0.5003, "lr": 4e-08, "epoch": 0.025591810620601407, "percentage": 2.55, "elapsed_time": "0:01:18", "remaining_time": "0:49:52", "rewards/chosen": 0.0021508518410356422, "logps/chosen": -439.0241056743421, "logits/chosen": -59713886.315789476, "rewards/rejected": -0.0005857022479176521, "logps/rejected": -198.41524832589286, "logits/rejected": -50416222.47619048, "rewards/margins": 0.0027365540889532944, "regularization/tv": 0.0003946188371628523, "regularization/advantage_range": 0.39461883902549744, "kl": 0.006213414948433638}
+{"current_steps": 10, "total_steps": 196, "loss": 0.5008, "lr": 9e-08, "epoch": 0.05118362124120281, "percentage": 5.1, "elapsed_time": "0:02:39", "remaining_time": "0:49:21", "rewards/chosen": -0.0003495565676069879, "logps/chosen": -429.94201501623377, "logits/chosen": -59302034.28571428, "rewards/rejected": 0.000481259616383587, "logps/rejected": -192.86311652861446, "logits/rejected": -50026792.09638554, "rewards/margins": -0.0008308161839905749, "regularization/tv": 0.0006011345540173352, "regularization/advantage_range": 0.6011344790458679, "kl": 0.019698908552527428, "stability/response_length_mean": 46.125, "stability/response_length_std": 27.231481552124023, "stability/response_length_var": 741.5535888671875, "stability/token_entropy_mean": 3.2646484375, "stability/repetition_rate_mean": 0.3668472170829773}
+{"current_steps": 15, "total_steps": 196, "loss": 0.5005, "lr": 1.4e-07, "epoch": 0.07677543186180422, "percentage": 7.65, "elapsed_time": "0:03:59", "remaining_time": "0:48:04", "rewards/chosen": -0.0009836391066059922, "logps/chosen": -404.140625, "logits/chosen": -56237422.15757576, "rewards/rejected": 0.003750608044285928, "logps/rejected": -200.65655241935485, "logits/rejected": -51927885.62580645, "rewards/margins": -0.00473424715089192, "regularization/tv": 0.0005793093587271869, "regularization/advantage_range": 0.579309344291687, "kl": 0.013860845938324928, "stability/response_length_mean": 28.5, "stability/response_length_std": 30.10457992553711, "stability/response_length_var": 906.2857055664062, "stability/token_entropy_mean": 3.76171875, "stability/repetition_rate_mean": 0.2763020694255829}
+{"current_steps": 20, "total_steps": 196, "loss": 0.5008, "lr": 1.8999999999999998e-07, "epoch": 0.10236724248240563, "percentage": 10.2, "elapsed_time": "0:05:20", "remaining_time": "0:47:01", "rewards/chosen": -0.00045777419582009314, "logps/chosen": -343.7012939453125, "logits/chosen": -51669683.2, "rewards/rejected": 0.0010081703774631023, "logps/rejected": -225.653271484375, "logits/rejected": -48070534.4, "rewards/margins": -0.0014659445732831955, "regularization/tv": 0.0005812579765915871, "regularization/advantage_range": 0.5812579989433289, "kl": 0.016369277611374855, "stability/response_length_mean": 42.125, "stability/response_length_std": 27.01553726196289, "stability/response_length_var": 729.8392944335938, "stability/token_entropy_mean": 4.44921875, "stability/repetition_rate_mean": 0.2040860652923584}
+{"current_steps": 25, "total_steps": 196, "loss": 0.5008, "lr": 2.4e-07, "epoch": 0.12795905310300704, "percentage": 12.76, "elapsed_time": "0:06:40", "remaining_time": "0:45:39", "rewards/chosen": 0.002771485262903674, "logps/chosen": -378.46373922413795, "logits/chosen": -56468356.4137931, "rewards/rejected": 0.0012683660643441337, "logps/rejected": -202.78017857142856, "logits/rejected": -52007339.154285714, "rewards/margins": 0.0015031191985595401, "regularization/tv": 0.0005770506686531007, "regularization/advantage_range": 0.5770506262779236, "kl": 0.019973892718553543, "stability/response_length_mean": 42.0, "stability/response_length_std": 27.66121482849121, "stability/response_length_var": 765.1428833007812, "stability/token_entropy_mean": 4.8046875, "stability/repetition_rate_mean": 0.18011869490146637}
+{"current_steps": 30, "total_steps": 196, "loss": 0.5006, "lr": 2.9e-07, "epoch": 0.15355086372360843, "percentage": 15.31, "elapsed_time": "0:08:00", "remaining_time": "0:44:19", "rewards/chosen": 0.004691862419947681, "logps/chosen": -468.5343435402685, "logits/chosen": -57724185.77181208, "rewards/rejected": 0.0014308589940879776, "logps/rejected": -198.2335754751462, "logits/rejected": -51514062.596491225, "rewards/margins": 0.0032610034258597036, "regularization/tv": 0.0005929219187237322, "regularization/advantage_range": 0.5929219126701355, "kl": 0.018949225544929504, "stability/response_length_mean": 48.375, "stability/response_length_std": 28.93311309814453, "stability/response_length_var": 837.125, "stability/token_entropy_mean": 4.5625, "stability/repetition_rate_mean": 0.185546875}
+{"current_steps": 35, "total_steps": 196, "loss": 0.5004, "lr": 3.4000000000000003e-07, "epoch": 0.17914267434420986, "percentage": 17.86, "elapsed_time": "0:09:21", "remaining_time": "0:43:04", "rewards/chosen": 0.003739585604848741, "logps/chosen": -385.89912974683546, "logits/chosen": -57645127.29113924, "rewards/rejected": 0.0030492323416250722, "logps/rejected": -195.09599247685185, "logits/rejected": -53893480.2962963, "rewards/margins": 0.0006903532632236689, "regularization/tv": 0.0005582152516581118, "regularization/advantage_range": 0.5582152605056763, "kl": 0.023824213072657585, "stability/response_length_mean": 53.375, "stability/response_length_std": 21.14533042907715, "stability/response_length_var": 447.125, "stability/token_entropy_mean": 5.337890625, "stability/repetition_rate_mean": 0.3566158413887024}
+{"current_steps": 40, "total_steps": 196, "loss": 0.5004, "lr": 3.8999999999999997e-07, "epoch": 0.20473448496481125, "percentage": 20.41, "elapsed_time": "0:10:42", "remaining_time": "0:41:46", "rewards/chosen": 0.005213038995862007, "logps/chosen": -421.85048828125, "logits/chosen": -56629888.0, "rewards/rejected": 0.0013409496285021305, "logps/rejected": -219.295849609375, "logits/rejected": -50223590.4, "rewards/margins": 0.0038720893673598765, "regularization/tv": 0.0006682613748125732, "regularization/advantage_range": 0.6682614088058472, "kl": 0.03129922226071358, "stability/response_length_mean": 33.125, "stability/response_length_std": 33.01704025268555, "stability/response_length_var": 1090.125, "stability/token_entropy_mean": 4.97265625, "stability/repetition_rate_mean": 0.142578125}
+{"current_steps": 45, "total_steps": 196, "loss": 0.5004, "lr": 4.3999999999999997e-07, "epoch": 0.23032629558541268, "percentage": 22.96, "elapsed_time": "0:12:03", "remaining_time": "0:40:27", "rewards/chosen": 0.004419918028300241, "logps/chosen": -352.2842648909396, "logits/chosen": -52700613.58389262, "rewards/rejected": 0.0031777226436905, "logps/rejected": -243.8423793859649, "logits/rejected": -50780142.03508772, "rewards/margins": 0.001242195384609741, "regularization/tv": 0.0006375295924954116, "regularization/advantage_range": 0.6375295519828796, "kl": 0.041447144001722336, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.87324333190918, "stability/response_length_var": 892.4107055664062, "stability/token_entropy_mean": 4.755859375, "stability/repetition_rate_mean": 0.0935424193739891}
+{"current_steps": 50, "total_steps": 196, "loss": 0.5004, "lr": 4.9e-07, "epoch": 0.2559181062060141, "percentage": 25.51, "elapsed_time": "0:13:23", "remaining_time": "0:39:05", "rewards/chosen": 0.00725229154497185, "logps/chosen": -375.12659920302013, "logits/chosen": -54044507.060402684, "rewards/rejected": 0.005870287878471508, "logps/rejected": -180.181914747807, "logits/rejected": -48291307.04093567, "rewards/margins": 0.0013820036665003416, "regularization/tv": 0.0005456676008179784, "regularization/advantage_range": 0.5456675887107849, "kl": 0.05188734456896782, "stability/response_length_mean": 55.375, "stability/response_length_std": 16.361431121826172, "stability/response_length_var": 267.6964416503906, "stability/token_entropy_mean": 3.810546875, "stability/repetition_rate_mean": 0.4206075370311737}
+{"current_steps": 55, "total_steps": 196, "loss": 0.5002, "lr": 5.4e-07, "epoch": 0.28150991682661547, "percentage": 28.06, "elapsed_time": "0:14:43", "remaining_time": "0:37:44", "rewards/chosen": 0.0005464234522410802, "logps/chosen": -344.50655085403724, "logits/chosen": -56957481.34161491, "rewards/rejected": 0.00290468803741647, "logps/rejected": -230.41897602201257, "logits/rejected": -49573807.49685535, "rewards/margins": -0.0023582645851753896, "regularization/tv": 0.0006177856703288853, "regularization/advantage_range": 0.617785632610321, "kl": 0.0480094812810421, "stability/response_length_mean": 22.375, "stability/response_length_std": 30.236862182617188, "stability/response_length_var": 914.2678833007812, "stability/token_entropy_mean": 4.3671875, "stability/repetition_rate_mean": 0.10418051481246948}
+{"current_steps": 60, "total_steps": 196, "loss": 0.5004, "lr": 5.9e-07, "epoch": 0.30710172744721687, "percentage": 30.61, "elapsed_time": "0:16:04", "remaining_time": "0:36:26", "rewards/chosen": 0.009183568665475556, "logps/chosen": -359.6289772727273, "logits/chosen": -52675124.75151515, "rewards/rejected": 0.00757992805973176, "logps/rejected": -178.1511214717742, "logits/rejected": -47822603.561290324, "rewards/margins": 0.0016036406057437962, "regularization/tv": 0.0005291791749186814, "regularization/advantage_range": 0.5291790962219238, "kl": 0.04956696555018425, "stability/response_length_mean": 51.125, "stability/response_length_std": 21.90523338317871, "stability/response_length_var": 479.83929443359375, "stability/token_entropy_mean": 5.322265625, "stability/repetition_rate_mean": 0.21226298809051514}
+{"current_steps": 65, "total_steps": 196, "loss": 0.5002, "lr": 6.4e-07, "epoch": 0.3326935380678183, "percentage": 33.16, "elapsed_time": "0:17:25", "remaining_time": "0:35:06", "rewards/chosen": 0.011598679327195691, "logps/chosen": -381.83079637096773, "logits/chosen": -57571228.90322581, "rewards/rejected": 0.00793868267174923, "logps/rejected": -183.50297111742424, "logits/rejected": -47274263.27272727, "rewards/margins": 0.0036599966554464614, "regularization/tv": 0.0005749912816099823, "regularization/advantage_range": 0.5749912261962891, "kl": 0.07576815038919449, "stability/response_length_mean": 17.875, "stability/response_length_std": 23.485177993774414, "stability/response_length_var": 551.5535888671875, "stability/token_entropy_mean": 5.66796875, "stability/repetition_rate_mean": 0.18097305297851562}
+{"current_steps": 70, "total_steps": 196, "loss": 0.4998, "lr": 6.9e-07, "epoch": 0.3582853486884197, "percentage": 35.71, "elapsed_time": "0:18:45", "remaining_time": "0:33:45", "rewards/chosen": 0.009184343584122196, "logps/chosen": -379.0366179435484, "logits/chosen": -52759284.438709676, "rewards/rejected": 0.006874283877286044, "logps/rejected": -220.35340909090908, "logits/rejected": -49026904.43636364, "rewards/margins": 0.002310059706836152, "regularization/tv": 0.0006015223334543407, "regularization/advantage_range": 0.6015222668647766, "kl": 0.07580217719078064, "stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.68359375, "stability/repetition_rate_mean": 0.1796875}
+{"current_steps": 75, "total_steps": 196, "loss": 0.5, "lr": 7.4e-07, "epoch": 0.3838771593090211, "percentage": 38.27, "elapsed_time": "0:20:06", "remaining_time": "0:32:26", "rewards/chosen": 0.011890417820698506, "logps/chosen": -314.1518950591216, "logits/chosen": -49505473.72972973, "rewards/rejected": 0.007232982058857762, "logps/rejected": -228.81252271075581, "logits/rejected": -47898463.255813956, "rewards/margins": 0.004657435761840744, "regularization/tv": 0.0006583484937436879, "regularization/advantage_range": 0.6583485007286072, "kl": 0.09119434654712677, "stability/response_length_mean": 24.25, "stability/response_length_std": 32.92306900024414, "stability/response_length_var": 1083.9285888671875, "stability/token_entropy_mean": 1.91650390625, "stability/repetition_rate_mean": 0.068359375}
+{"current_steps": 80, "total_steps": 196, "loss": 0.4998, "lr": 7.9e-07, "epoch": 0.4094689699296225, "percentage": 40.82, "elapsed_time": "0:21:27", "remaining_time": "0:31:06", "rewards/chosen": 0.014854513256755096, "logps/chosen": -403.8815190397351, "logits/chosen": -58620202.38410596, "rewards/rejected": 0.006409630267577764, "logps/rejected": -200.99170210798817, "logits/rejected": -53447098.31952663, "rewards/margins": 0.008444882989177331, "regularization/tv": 0.0006350657204166055, "regularization/advantage_range": 0.6350656747817993, "kl": 0.10768461227416992, "stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.203125, "stability/repetition_rate_mean": 0.19479167461395264}
+{"current_steps": 85, "total_steps": 196, "loss": 0.4995, "lr": 8.399999999999999e-07, "epoch": 0.4350607805502239, "percentage": 43.37, "elapsed_time": "0:22:47", "remaining_time": "0:29:46", "rewards/chosen": 0.014681977610434256, "logps/chosen": -358.18568548387094, "logits/chosen": -51473758.14193548, "rewards/rejected": 0.0063012014735828745, "logps/rejected": -230.36178977272726, "logits/rejected": -47141112.24242424, "rewards/margins": 0.00838077613685138, "regularization/tv": 0.0006356575759127736, "regularization/advantage_range": 0.6356574892997742, "kl": 0.10215399414300919, "stability/response_length_mean": 32.5, "stability/response_length_std": 33.67915725708008, "stability/response_length_var": 1134.2857666015625, "stability/token_entropy_mean": 4.91015625, "stability/repetition_rate_mean": 0.095703125}
+{"current_steps": 90, "total_steps": 196, "loss": 0.4994, "lr": 8.9e-07, "epoch": 0.46065259117082535, "percentage": 45.92, "elapsed_time": "0:24:08", "remaining_time": "0:28:26", "rewards/chosen": 0.016429907236343775, "logps/chosen": -380.6251001602564, "logits/chosen": -58412242.05128205, "rewards/rejected": 0.005432141990196414, "logps/rejected": -187.1724823742378, "logits/rejected": -50579849.36585366, "rewards/margins": 0.01099776524614736, "regularization/tv": 0.0007223256980068982, "regularization/advantage_range": 0.7223256826400757, "kl": 0.14038605988025665, "stability/response_length_mean": 33.875, "stability/response_length_std": 29.132394790649414, "stability/response_length_var": 848.6964111328125, "stability/token_entropy_mean": 3.72021484375, "stability/repetition_rate_mean": 0.2647964060306549}
+{"current_steps": 95, "total_steps": 196, "loss": 0.4994, "lr": 9.399999999999999e-07, "epoch": 0.48624440179142675, "percentage": 48.47, "elapsed_time": "0:25:29", "remaining_time": "0:27:05", "rewards/chosen": 0.01595103287998634, "logps/chosen": -314.8338607594937, "logits/chosen": -50392932.4556962, "rewards/rejected": 0.0039456310095610445, "logps/rejected": -207.10426311728395, "logits/rejected": -48968514.37037037, "rewards/margins": 0.012005401870425297, "regularization/tv": 0.0007024412043392658, "regularization/advantage_range": 0.7024412155151367, "kl": 0.12943434715270996, "stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.685546875, "stability/repetition_rate_mean": 0.20155483484268188}
+{"current_steps": 100, "total_steps": 196, "loss": 0.4991, "lr": 9.9e-07, "epoch": 0.5118362124120281, "percentage": 51.02, "elapsed_time": "0:26:49", "remaining_time": "0:25:45", "rewards/chosen": 0.023018472903483623, "logps/chosen": -359.63999155405406, "logits/chosen": -54706445.83783784, "rewards/rejected": 0.01134931625321854, "logps/rejected": -216.46538880813952, "logits/rejected": -50905885.76744186, "rewards/margins": 0.011669156650265083, "regularization/tv": 0.0007743232999928296, "regularization/advantage_range": 0.7743232846260071, "kl": 0.15796113014221191, "stability/response_length_mean": 37.5, "stability/response_length_std": 29.640705108642578, "stability/response_length_var": 878.5714111328125, "stability/token_entropy_mean": 4.150390625, "stability/repetition_rate_mean": 0.23383361101150513}
+{"current_steps": 105, "total_steps": 196, "loss": 0.4989, "lr": 9.999815554351224e-07, "epoch": 0.5374280230326296, "percentage": 53.57, "elapsed_time": "0:28:10", "remaining_time": "0:24:24", "rewards/chosen": 0.0319610503782709, "logps/chosen": -399.5828313253012, "logits/chosen": -59367004.530120485, "rewards/rejected": 0.011829294167555772, "logps/rejected": -209.8855265827922, "logits/rejected": -53783964.25974026, "rewards/margins": 0.020131756210715127, "regularization/tv": 0.001071037957444787, "regularization/advantage_range": 1.0710378885269165, "kl": 0.1611875742673874, "stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.41796875, "stability/repetition_rate_mean": 0.2053571343421936}
+{"current_steps": 110, "total_steps": 196, "loss": 0.4988, "lr": 9.999066267225444e-07, "epoch": 0.5630198336532309, "percentage": 56.12, "elapsed_time": "0:29:31", "remaining_time": "0:23:05", "rewards/chosen": 0.02724501885563494, "logps/chosen": -378.29080384036143, "logits/chosen": -57380000.38554217, "rewards/rejected": 0.01249428925576148, "logps/rejected": -205.96319500811688, "logits/rejected": -53610043.84415584, "rewards/margins": 0.01475072959987346, "regularization/tv": 0.0009510820964351296, "regularization/advantage_range": 0.9510820508003235, "kl": 0.19908857345581055, "stability/response_length_mean": 32.25, "stability/response_length_std": 30.541540145874023, "stability/response_length_var": 932.7857055664062, "stability/token_entropy_mean": 4.4296875, "stability/repetition_rate_mean": 0.3234747052192688}
+{"current_steps": 115, "total_steps": 196, "loss": 0.4986, "lr": 9.997740697079592e-07, "epoch": 0.5886116442738324, "percentage": 58.67, "elapsed_time": "0:30:53", "remaining_time": "0:21:45", "rewards/chosen": 0.03414275072797944, "logps/chosen": -364.25111253955697, "logits/chosen": -60607008.405063294, "rewards/rejected": 0.012134410716869213, "logps/rejected": -201.07520736882717, "logits/rejected": -53416820.938271604, "rewards/margins": 0.022008340011110225, "regularization/tv": 0.0010460062185302377, "regularization/advantage_range": 1.046006202697754, "kl": 0.2150437831878662, "stability/response_length_mean": 47.625, "stability/response_length_std": 27.21311378479004, "stability/response_length_var": 740.5535888671875, "stability/token_entropy_mean": 4.78125, "stability/repetition_rate_mean": 0.3148462176322937}
+{"current_steps": 120, "total_steps": 196, "loss": 0.4981, "lr": 9.995838996722914e-07, "epoch": 0.6142034548944337, "percentage": 61.22, "elapsed_time": "0:32:12", "remaining_time": "0:20:24", "rewards/chosen": 0.03744382172628166, "logps/chosen": -383.87048100490193, "logits/chosen": -57751137.04575163, "rewards/rejected": 0.01369288296042802, "logps/rejected": -212.07943488023952, "logits/rejected": -54370886.51497006, "rewards/margins": 0.023750938765853644, "regularization/tv": 0.0011908583110198379, "regularization/advantage_range": 1.190858244895935, "kl": 0.221540167927742, "stability/response_length_mean": 27.875, "stability/response_length_std": 30.31943130493164, "stability/response_length_var": 919.2678833007812, "stability/token_entropy_mean": 4.646484375, "stability/repetition_rate_mean": 0.1762429028749466}
+{"current_steps": 125, "total_steps": 196, "loss": 0.4983, "lr": 9.993361385379874e-07, "epoch": 0.6397952655150352, "percentage": 63.78, "elapsed_time": "0:33:33", "remaining_time": "0:19:03", "rewards/chosen": 0.035337840041069137, "logps/chosen": -417.02595248287673, "logits/chosen": -60704396.2739726, "rewards/rejected": 0.01157299129442237, "logps/rejected": -216.25145923132183, "logits/rejected": -55673891.31034483, "rewards/margins": 0.023764848746646767, "regularization/tv": 0.0014626210322603583, "regularization/advantage_range": 1.4626210927963257, "kl": 0.2525067925453186, "stability/response_length_mean": 15.875, "stability/response_length_std": 24.584766387939453, "stability/response_length_var": 604.4107055664062, "stability/token_entropy_mean": 3.5, "stability/repetition_rate_mean": 0.15937499701976776}
+{"current_steps": 130, "total_steps": 196, "loss": 0.4979, "lr": 9.99030814866488e-07, "epoch": 0.6653870761356366, "percentage": 66.33, "elapsed_time": "0:34:53", "remaining_time": "0:17:42", "rewards/chosen": 0.039759131400815904, "logps/chosen": -392.2603326612903, "logits/chosen": -57406563.09677419, "rewards/rejected": 0.010868869405804258, "logps/rejected": -183.81022727272727, "logits/rejected": -53861922.13333333, "rewards/margins": 0.028890261995011646, "regularization/tv": 0.0016048549441620708, "regularization/advantage_range": 1.604854941368103, "kl": 0.2731947600841522, "stability/response_length_mean": 43.25, "stability/response_length_std": 29.620214462280273, "stability/response_length_var": 877.3571166992188, "stability/token_entropy_mean": 3.6943359375, "stability/repetition_rate_mean": 0.23874999582767487}
+{"current_steps": 135, "total_steps": 196, "loss": 0.4972, "lr": 9.986679638549358e-07, "epoch": 0.690978886756238, "percentage": 68.88, "elapsed_time": "0:36:13", "remaining_time": "0:16:22", "rewards/chosen": 0.051993937580132044, "logps/chosen": -401.296227952454, "logits/chosen": -59866928.68711656, "rewards/rejected": 0.017273878595631592, "logps/rejected": -247.69799960191082, "logits/rejected": -55031821.04458599, "rewards/margins": 0.03472005898450045, "regularization/tv": 0.0018137602601200342, "regularization/advantage_range": 1.8137600421905518, "kl": 0.3012661039829254, "stability/response_length_mean": 29.25, "stability/response_length_std": 29.936838150024414, "stability/response_length_var": 896.2142944335938, "stability/token_entropy_mean": 4.078125, "stability/repetition_rate_mean": 0.2230902761220932}
+{"current_steps": 140, "total_steps": 196, "loss": 0.497, "lr": 9.982476273321175e-07, "epoch": 0.7165706973768394, "percentage": 71.43, "elapsed_time": "0:37:34", "remaining_time": "0:15:01", "rewards/chosen": 0.05096650557084517, "logps/chosen": -366.6800899621212, "logits/chosen": -57111142.4, "rewards/rejected": 0.00036814520435948523, "logps/rejected": -190.03079637096775, "logits/rejected": -50093432.56774194, "rewards/margins": 0.050598360366485684, "regularization/tv": 0.0021453395020216703, "regularization/advantage_range": 2.1453397274017334, "kl": 0.2736550271511078, "stability/response_length_mean": 52.75, "stability/response_length_std": 22.72663688659668, "stability/response_length_var": 516.5, "stability/token_entropy_mean": 5.111328125, "stability/repetition_rate_mean": 0.3053385317325592}
+{"current_steps": 145, "total_steps": 196, "loss": 0.497, "lr": 9.977698537536417e-07, "epoch": 0.7421625079974408, "percentage": 73.98, "elapsed_time": "0:38:54", "remaining_time": "0:13:41", "rewards/chosen": 0.04440879821777344, "logps/chosen": -329.8277249623494, "logits/chosen": -51165295.03614458, "rewards/rejected": 0.011527130356082669, "logps/rejected": -199.37265371347402, "logits/rejected": -51708233.14285714, "rewards/margins": 0.03288166786169077, "regularization/tv": 0.001656163134612143, "regularization/advantage_range": 1.656163215637207, "kl": 0.3225500285625458, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.73423957824707, "stability/response_length_var": 884.125, "stability/token_entropy_mean": 5.26953125, "stability/repetition_rate_mean": 0.24838952720165253}
+{"current_steps": 150, "total_steps": 196, "loss": 0.4967, "lr": 9.972346981963546e-07, "epoch": 0.7677543186180422, "percentage": 76.53, "elapsed_time": "0:40:16", "remaining_time": "0:12:20", "rewards/chosen": 0.06905886706183939, "logps/chosen": -397.2438419117647, "logits/chosen": -55434553.22352941, "rewards/rejected": 0.013413952191670736, "logps/rejected": -207.591796875, "logits/rejected": -50214345.38666666, "rewards/margins": 0.055644914870168656, "regularization/tv": 0.002014032332226634, "regularization/advantage_range": 2.0140323638916016, "kl": 0.3542978763580322, "stability/response_length_mean": 29.375, "stability/response_length_std": 29.193626403808594, "stability/response_length_var": 852.2678833007812, "stability/token_entropy_mean": 4.59375, "stability/repetition_rate_mean": 0.23066024482250214}
+{"current_steps": 155, "total_steps": 196, "loss": 0.4968, "lr": 9.966422223519885e-07, "epoch": 0.7933461292386437, "percentage": 79.08, "elapsed_time": "0:41:37", "remaining_time": "0:11:00", "rewards/chosen": 0.051560289719525504, "logps/chosen": -360.47366727941176, "logits/chosen": -54049448.65882353, "rewards/rejected": 0.006060927708943685, "logps/rejected": -191.245, "logits/rejected": -49388888.74666667, "rewards/margins": 0.04549936201058182, "regularization/tv": 0.002201819559559226, "regularization/advantage_range": 2.201819658279419, "kl": 0.351535439491272, "stability/response_length_mean": 61.375, "stability/response_length_std": 7.424621105194092, "stability/response_length_var": 55.125, "stability/token_entropy_mean": 4.544921875, "stability/repetition_rate_mean": 0.3171784281730652}
+{"current_steps": 160, "total_steps": 196, "loss": 0.4963, "lr": 9.959924945200523e-07, "epoch": 0.818937939859245, "percentage": 81.63, "elapsed_time": "0:42:58", "remaining_time": "0:09:40", "rewards/chosen": 0.07432699771154494, "logps/chosen": -416.15206473214283, "logits/chosen": -55283943.61904762, "rewards/rejected": 0.01710961053245946, "logps/rejected": -186.7262541118421, "logits/rejected": -49011105.684210524, "rewards/margins": 0.05721738717908548, "regularization/tv": 0.0024417235981673002, "regularization/advantage_range": 2.441723585128784, "kl": 0.38903898000717163, "stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.3837890625, "stability/repetition_rate_mean": 0.31046873331069946}
+{"current_steps": 165, "total_steps": 196, "loss": 0.4954, "lr": 9.952855895999568e-07, "epoch": 0.8445297504798465, "percentage": 84.18, "elapsed_time": "0:44:18", "remaining_time": "0:08:19", "rewards/chosen": 0.06546377861636808, "logps/chosen": -370.26210039511494, "logits/chosen": -54472480.36781609, "rewards/rejected": 0.016622740928440877, "logps/rejected": -212.42588827054794, "logits/rejected": -49860436.16438356, "rewards/margins": 0.048841037687927205, "regularization/tv": 0.002269331831485033, "regularization/advantage_range": 2.269331455230713, "kl": 0.4146839678287506, "stability/response_length_mean": 22.5, "stability/response_length_std": 28.33977508544922, "stability/response_length_var": 803.1428833007812, "stability/token_entropy_mean": 4.87109375, "stability/repetition_rate_mean": 0.2411106526851654}
+{"current_steps": 170, "total_steps": 196, "loss": 0.4948, "lr": 9.9452158908238e-07, "epoch": 0.8701215611004478, "percentage": 86.73, "elapsed_time": "0:45:40", "remaining_time": "0:06:59", "rewards/chosen": 0.0933977894529083, "logps/chosen": -440.4078679733728, "logits/chosen": -59181152.94674556, "rewards/rejected": 0.02611484748638229, "logps/rejected": -190.99668874172184, "logits/rejected": -52670104.58278146, "rewards/margins": 0.06728294196652601, "regularization/tv": 0.0025218818336725235, "regularization/advantage_range": 2.5218818187713623, "kl": 0.5150820016860962, "stability/response_length_mean": 37.25, "stability/response_length_std": 30.92733383178711, "stability/response_length_var": 956.5, "stability/token_entropy_mean": 3.80126953125, "stability/repetition_rate_mean": 0.2678554654121399}
+{"current_steps": 175, "total_steps": 196, "loss": 0.4945, "lr": 9.937005810398745e-07, "epoch": 0.8957133717210493, "percentage": 89.29, "elapsed_time": "0:47:00", "remaining_time": "0:05:38", "rewards/chosen": 0.08484269911984363, "logps/chosen": -355.58876129518075, "logits/chosen": -56722012.530120485, "rewards/rejected": 0.020287346530270266, "logps/rejected": -194.6046570616883, "logits/rejected": -48136594.28571428, "rewards/margins": 0.06455535258957336, "regularization/tv": 0.0030055036768317223, "regularization/advantage_range": 3.0055034160614014, "kl": 0.5376589894294739, "stability/response_length_mean": 39.75, "stability/response_length_std": 27.473363876342773, "stability/response_length_var": 754.7857055664062, "stability/token_entropy_mean": 4.73046875, "stability/repetition_rate_mean": 0.2454427182674408}
+{"current_steps": 180, "total_steps": 196, "loss": 0.4938, "lr": 9.928226601167138e-07, "epoch": 0.9213051823416507, "percentage": 91.84, "elapsed_time": "0:48:20", "remaining_time": "0:04:17", "rewards/chosen": 0.0989208455466054, "logps/chosen": -375.1566334355828, "logits/chosen": -52728517.88957055, "rewards/rejected": 0.03126500974035567, "logps/rejected": -210.98556926751593, "logits/rejected": -48089241.27388535, "rewards/margins": 0.06765583580624973, "regularization/tv": 0.0032784626819193363, "regularization/advantage_range": 3.2784626483917236, "kl": 0.5820572972297668, "stability/response_length_mean": 40.375, "stability/response_length_std": 32.61874008178711, "stability/response_length_var": 1063.982177734375, "stability/token_entropy_mean": 3.091796875, "stability/repetition_rate_mean": 0.26171875}
+{"current_steps": 185, "total_steps": 196, "loss": 0.4943, "lr": 9.918879275179817e-07, "epoch": 0.946896992962252, "percentage": 94.39, "elapsed_time": "0:49:40", "remaining_time": "0:02:57", "rewards/chosen": 0.09057903289794922, "logps/chosen": -311.0955825617284, "logits/chosen": -54704355.55555555, "rewards/rejected": 0.015578183946730216, "logps/rejected": -196.88338113132912, "logits/rejected": -49614323.037974685, "rewards/margins": 0.075000848951219, "regularization/tv": 0.003431336721405387, "regularization/advantage_range": 3.4313366413116455, "kl": 0.6085312962532043, "stability/response_length_mean": 26.25, "stability/response_length_std": 31.45404815673828, "stability/response_length_var": 989.3571166992188, "stability/token_entropy_mean": 3.7890625, "stability/repetition_rate_mean": 0.14541903138160706}
+{"current_steps": 190, "total_steps": 196, "loss": 0.4933, "lr": 9.90896490997906e-07, "epoch": 0.9724888035828535, "percentage": 96.94, "elapsed_time": "0:51:01", "remaining_time": "0:01:36", "rewards/chosen": 0.11779353433026525, "logps/chosen": -391.2273811751497, "logits/chosen": -53173057.91616766, "rewards/rejected": 0.03322306177974527, "logps/rejected": -210.86637050653596, "logits/rejected": -49808035.97385621, "rewards/margins": 0.08457047255051998, "regularization/tv": 0.003950292710214853, "regularization/advantage_range": 3.9502923488616943, "kl": 0.6512568593025208, "stability/response_length_mean": 17.625, "stability/response_length_std": 28.680194854736328, "stability/response_length_var": 822.5535888671875, "stability/token_entropy_mean": 4.8427734375, "stability/repetition_rate_mean": 0.1901041716337204}
+{"current_steps": 195, "total_steps": 196, "loss": 0.4937, "lr": 9.898484648474362e-07, "epoch": 0.9980806142034548, "percentage": 99.49, "elapsed_time": "0:52:21", "remaining_time": "0:00:16", "rewards/chosen": 0.11162065075289819, "logps/chosen": -391.6687247983871, "logits/chosen": -60185910.5032258, "rewards/rejected": 0.0352558251583215, "logps/rejected": -197.56677320075758, "logits/rejected": -53942718.836363636, "rewards/margins": 0.07636482559457669, "regularization/tv": 0.003479656297713518, "regularization/advantage_range": 3.479656219482422, "kl": 0.6606561541557312, "stability/response_length_mean": 52.875, "stability/response_length_std": 22.850053787231445, "stability/response_length_var": 522.125, "stability/token_entropy_mean": 5.18359375, "stability/repetition_rate_mean": 0.42074424028396606}
+{"current_steps": 196, "total_steps": 196, "epoch": 1.0, "percentage": 100.0, "elapsed_time": "0:52:35", "remaining_time": "0:00:00"}

mistral7b_kto_tv_rho0.001/trainer_state.json

@@ -0,0 +1,974 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 196,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.025591810620601407,
+      "grad_norm": 0.5810207724571228,
+      "kl": 0.006213414948433638,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59713886.315789476,
+      "logits/rejected": -50416222.47619048,
+      "logps/chosen": -439.0241056743421,
+      "logps/rejected": -198.41524832589286,
+      "loss": 0.5003,
+      "loss/base_kto": 0.4996565878391266,
+      "metrics/advantage_var": 0.04205384477972984,
+      "regularization/advantage_range": 0.39461883902549744,
+      "regularization/tv": 0.0003946188371628523,
+      "rewards/chosen": 0.0021508518410356422,
+      "rewards/margins": 0.0027365540889532944,
+      "rewards/rejected": -0.0005857022479176521,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.6166413426399231,
+      "kl": 0.019698908552527428,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59302034.28571428,
+      "logits/rejected": -50026792.09638554,
+      "logps/chosen": -429.94201501623377,
+      "logps/rejected": -192.86311652861446,
+      "loss": 0.5008,
+      "loss/base_kto": 0.5001104474067688,
+      "metrics/advantage_var": 0.06644176691770554,
+      "regularization/advantage_range": 0.6011344790458679,
+      "regularization/tv": 0.0006011345540173352,
+      "rewards/chosen": -0.0003495565676069879,
+      "rewards/margins": -0.0008308161839905749,
+      "rewards/rejected": 0.000481259616383587,
+      "stability/repetition_rate_mean": 0.3668472170829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.2646484375,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.6122223734855652,
+      "kl": 0.013860845938324928,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56237422.15757576,
+      "logits/rejected": -51927885.62580645,
+      "logps/chosen": -404.140625,
+      "logps/rejected": -200.65655241935485,
+      "loss": 0.5005,
+      "loss/base_kto": 0.5005344748497009,
+      "metrics/advantage_var": 0.06271388381719589,
+      "regularization/advantage_range": 0.579309344291687,
+      "regularization/tv": 0.0005793093587271869,
+      "rewards/chosen": -0.0009836391066059922,
+      "rewards/margins": -0.00473424715089192,
+      "rewards/rejected": 0.003750608044285928,
+      "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": 3.76171875,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.6354861259460449,
+      "kl": 0.016369277611374855,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51669683.2,
+      "logits/rejected": -48070534.4,
+      "logps/chosen": -343.7012939453125,
+      "logps/rejected": -225.653271484375,
+      "loss": 0.5008,
+      "loss/base_kto": 0.5001991391181946,
+      "metrics/advantage_var": 0.06498073786497116,
+      "regularization/advantage_range": 0.5812579989433289,
+      "regularization/tv": 0.0005812579765915871,
+      "rewards/chosen": -0.00045777419582009314,
+      "rewards/margins": -0.0014659445732831955,
+      "rewards/rejected": 0.0010081703774631023,
+      "stability/repetition_rate_mean": 0.2040860652923584,
+      "stability/response_length_mean": 42.125,
+      "stability/response_length_std": 27.01553726196289,
+      "stability/response_length_var": 729.8392944335938,
+      "stability/token_entropy_mean": 4.44921875,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.5954239368438721,
+      "kl": 0.019973892718553543,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56468356.4137931,
+      "logits/rejected": -52007339.154285714,
+      "logps/chosen": -378.46373922413795,
+      "logps/rejected": -202.78017857142856,
+      "loss": 0.5008,
+      "loss/base_kto": 0.4998336732387543,
+      "metrics/advantage_var": 0.060091253370046616,
+      "regularization/advantage_range": 0.5770506262779236,
+      "regularization/tv": 0.0005770506686531007,
+      "rewards/chosen": 0.002771485262903674,
+      "rewards/margins": 0.0015031191985595401,
+      "rewards/rejected": 0.0012683660643441337,
+      "stability/repetition_rate_mean": 0.18011869490146637,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 27.66121482849121,
+      "stability/response_length_var": 765.1428833007812,
+      "stability/token_entropy_mean": 4.8046875,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.5939051508903503,
+      "kl": 0.018949225544929504,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57724185.77181208,
+      "logits/rejected": -51514062.596491225,
+      "logps/chosen": -468.5343435402685,
+      "logps/rejected": -198.2335754751462,
+      "loss": 0.5006,
+      "loss/base_kto": 0.4996163845062256,
+      "metrics/advantage_var": 0.06395137310028076,
+      "regularization/advantage_range": 0.5929219126701355,
+      "regularization/tv": 0.0005929219187237322,
+      "rewards/chosen": 0.004691862419947681,
+      "rewards/margins": 0.0032610034258597036,
+      "rewards/rejected": 0.0014308589940879776,
+      "stability/repetition_rate_mean": 0.185546875,
+      "stability/response_length_mean": 48.375,
+      "stability/response_length_std": 28.93311309814453,
+      "stability/response_length_var": 837.125,
+      "stability/token_entropy_mean": 4.5625,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6523953676223755,
+      "kl": 0.023824213072657585,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57645127.29113924,
+      "logits/rejected": -53893480.2962963,
+      "logps/chosen": -385.89912974683546,
+      "logps/rejected": -195.09599247685185,
+      "loss": 0.5004,
+      "loss/base_kto": 0.49997439980506897,
+      "metrics/advantage_var": 0.06881736218929291,
+      "regularization/advantage_range": 0.5582152605056763,
+      "regularization/tv": 0.0005582152516581118,
+      "rewards/chosen": 0.003739585604848741,
+      "rewards/margins": 0.0006903532632236689,
+      "rewards/rejected": 0.0030492323416250722,
+      "stability/repetition_rate_mean": 0.3566158413887024,
+      "stability/response_length_mean": 53.375,
+      "stability/response_length_std": 21.14533042907715,
+      "stability/response_length_var": 447.125,
+      "stability/token_entropy_mean": 5.337890625,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.57895427942276,
+      "kl": 0.03129922226071358,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56629888.0,
+      "logits/rejected": -50223590.4,
+      "logps/chosen": -421.85048828125,
+      "logps/rejected": -219.295849609375,
+      "loss": 0.5004,
+      "loss/base_kto": 0.49950963258743286,
+      "metrics/advantage_var": 0.0760921984910965,
+      "regularization/advantage_range": 0.6682614088058472,
+      "regularization/tv": 0.0006682613748125732,
+      "rewards/chosen": 0.005213038995862007,
+      "rewards/margins": 0.0038720893673598765,
+      "rewards/rejected": 0.0013409496285021305,
+      "stability/repetition_rate_mean": 0.142578125,
+      "stability/response_length_mean": 33.125,
+      "stability/response_length_std": 33.01704025268555,
+      "stability/response_length_var": 1090.125,
+      "stability/token_entropy_mean": 4.97265625,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.6361792683601379,
+      "kl": 0.041447144001722336,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52700613.58389262,
+      "logits/rejected": -50780142.03508772,
+      "logps/chosen": -352.2842648909396,
+      "logps/rejected": -243.8423793859649,
+      "loss": 0.5004,
+      "loss/base_kto": 0.49984151124954224,
+      "metrics/advantage_var": 0.07464335113763809,
+      "regularization/advantage_range": 0.6375295519828796,
+      "regularization/tv": 0.0006375295924954116,
+      "rewards/chosen": 0.004419918028300241,
+      "rewards/margins": 0.001242195384609741,
+      "rewards/rejected": 0.0031777226436905,
+      "stability/repetition_rate_mean": 0.0935424193739891,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.87324333190918,
+      "stability/response_length_var": 892.4107055664062,
+      "stability/token_entropy_mean": 4.755859375,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.5714625120162964,
+      "kl": 0.05188734456896782,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54044507.060402684,
+      "logits/rejected": -48291307.04093567,
+      "logps/chosen": -375.12659920302013,
+      "logps/rejected": -180.181914747807,
+      "loss": 0.5004,
+      "loss/base_kto": 0.49978047609329224,
+      "metrics/advantage_var": 0.05654274299740791,
+      "regularization/advantage_range": 0.5456675887107849,
+      "regularization/tv": 0.0005456676008179784,
+      "rewards/chosen": 0.00725229154497185,
+      "rewards/margins": 0.0013820036665003416,
+      "rewards/rejected": 0.005870287878471508,
+      "stability/repetition_rate_mean": 0.4206075370311737,
+      "stability/response_length_mean": 55.375,
+      "stability/response_length_std": 16.361431121826172,
+      "stability/response_length_var": 267.6964416503906,
+      "stability/token_entropy_mean": 3.810546875,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.6483849883079529,
+      "kl": 0.0480094812810421,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56957481.34161491,
+      "logits/rejected": -49573807.49685535,
+      "logps/chosen": -344.50655085403724,
+      "logps/rejected": -230.41897602201257,
+      "loss": 0.5002,
+      "loss/base_kto": 0.5002776980400085,
+      "metrics/advantage_var": 0.06638753414154053,
+      "regularization/advantage_range": 0.617785632610321,
+      "regularization/tv": 0.0006177856703288853,
+      "rewards/chosen": 0.0005464234522410802,
+      "rewards/margins": -0.0023582645851753896,
+      "rewards/rejected": 0.00290468803741647,
+      "stability/repetition_rate_mean": 0.10418051481246948,
+      "stability/response_length_mean": 22.375,
+      "stability/response_length_std": 30.236862182617188,
+      "stability/response_length_var": 914.2678833007812,
+      "stability/token_entropy_mean": 4.3671875,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.5852304697036743,
+      "kl": 0.04956696555018425,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52675124.75151515,
+      "logits/rejected": -47822603.561290324,
+      "logps/chosen": -359.6289772727273,
+      "logps/rejected": -178.1511214717742,
+      "loss": 0.5004,
+      "loss/base_kto": 0.49980688095092773,
+      "metrics/advantage_var": 0.05555450916290283,
+      "regularization/advantage_range": 0.5291790962219238,
+      "regularization/tv": 0.0005291791749186814,
+      "rewards/chosen": 0.009183568665475556,
+      "rewards/margins": 0.0016036406057437962,
+      "rewards/rejected": 0.00757992805973176,
+      "stability/repetition_rate_mean": 0.21226298809051514,
+      "stability/response_length_mean": 51.125,
+      "stability/response_length_std": 21.90523338317871,
+      "stability/response_length_var": 479.83929443359375,
+      "stability/token_entropy_mean": 5.322265625,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.5865442752838135,
+      "kl": 0.07576815038919449,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57571228.90322581,
+      "logits/rejected": -47274263.27272727,
+      "logps/chosen": -381.83079637096773,
+      "logps/rejected": -183.50297111742424,
+      "loss": 0.5002,
+      "loss/base_kto": 0.49959665536880493,
+      "metrics/advantage_var": 0.06296278536319733,
+      "regularization/advantage_range": 0.5749912261962891,
+      "regularization/tv": 0.0005749912816099823,
+      "rewards/chosen": 0.011598679327195691,
+      "rewards/margins": 0.0036599966554464614,
+      "rewards/rejected": 0.00793868267174923,
+      "stability/repetition_rate_mean": 0.18097305297851562,
+      "stability/response_length_mean": 17.875,
+      "stability/response_length_std": 23.485177993774414,
+      "stability/response_length_var": 551.5535888671875,
+      "stability/token_entropy_mean": 5.66796875,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5472328662872314,
+      "kl": 0.07580217719078064,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52759284.438709676,
+      "logits/rejected": -49026904.43636364,
+      "logps/chosen": -379.0366179435484,
+      "logps/rejected": -220.35340909090908,
+      "loss": 0.4998,
+      "loss/base_kto": 0.4997020363807678,
+      "metrics/advantage_var": 0.07085214555263519,
+      "regularization/advantage_range": 0.6015222668647766,
+      "regularization/tv": 0.0006015223334543407,
+      "rewards/chosen": 0.009184343584122196,
+      "rewards/margins": 0.002310059706836152,
+      "rewards/rejected": 0.006874283877286044,
+      "stability/repetition_rate_mean": 0.1796875,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.68359375,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.5962035655975342,
+      "kl": 0.09119434654712677,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49505473.72972973,
+      "logits/rejected": -47898463.255813956,
+      "logps/chosen": -314.1518950591216,
+      "logps/rejected": -228.81252271075581,
+      "loss": 0.5,
+      "loss/base_kto": 0.4993656575679779,
+      "metrics/advantage_var": 0.07352752983570099,
+      "regularization/advantage_range": 0.6583485007286072,
+      "regularization/tv": 0.0006583484937436879,
+      "rewards/chosen": 0.011890417820698506,
+      "rewards/margins": 0.004657435761840744,
+      "rewards/rejected": 0.007232982058857762,
+      "stability/repetition_rate_mean": 0.068359375,
+      "stability/response_length_mean": 24.25,
+      "stability/response_length_std": 32.92306900024414,
+      "stability/response_length_var": 1083.9285888671875,
+      "stability/token_entropy_mean": 1.91650390625,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.5844407081604004,
+      "kl": 0.10768461227416992,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58620202.38410596,
+      "logits/rejected": -53447098.31952663,
+      "logps/chosen": -403.8815190397351,
+      "logps/rejected": -200.99170210798817,
+      "loss": 0.4998,
+      "loss/base_kto": 0.49889513850212097,
+      "metrics/advantage_var": 0.07821118086576462,
+      "regularization/advantage_range": 0.6350656747817993,
+      "regularization/tv": 0.0006350657204166055,
+      "rewards/chosen": 0.014854513256755096,
+      "rewards/margins": 0.008444882989177331,
+      "rewards/rejected": 0.006409630267577764,
+      "stability/repetition_rate_mean": 0.19479167461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.203125,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.5908569097518921,
+      "kl": 0.10215399414300919,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51473758.14193548,
+      "logits/rejected": -47141112.24242424,
+      "logps/chosen": -358.18568548387094,
+      "logps/rejected": -230.36178977272726,
+      "loss": 0.4995,
+      "loss/base_kto": 0.49893245100975037,
+      "metrics/advantage_var": 0.07610619813203812,
+      "regularization/advantage_range": 0.6356574892997742,
+      "regularization/tv": 0.0006356575759127736,
+      "rewards/chosen": 0.014681977610434256,
+      "rewards/margins": 0.00838077613685138,
+      "rewards/rejected": 0.0063012014735828745,
+      "stability/repetition_rate_mean": 0.095703125,
+      "stability/response_length_mean": 32.5,
+      "stability/response_length_std": 33.67915725708008,
+      "stability/response_length_var": 1134.2857666015625,
+      "stability/token_entropy_mean": 4.91015625,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.5829842686653137,
+      "kl": 0.14038605988025665,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58412242.05128205,
+      "logits/rejected": -50579849.36585366,
+      "logps/chosen": -380.6251001602564,
+      "logps/rejected": -187.1724823742378,
+      "loss": 0.4994,
+      "loss/base_kto": 0.49847766757011414,
+      "metrics/advantage_var": 0.09148811548948288,
+      "regularization/advantage_range": 0.7223256826400757,
+      "regularization/tv": 0.0007223256980068982,
+      "rewards/chosen": 0.016429907236343775,
+      "rewards/margins": 0.01099776524614736,
+      "rewards/rejected": 0.005432141990196414,
+      "stability/repetition_rate_mean": 0.2647964060306549,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 29.132394790649414,
+      "stability/response_length_var": 848.6964111328125,
+      "stability/token_entropy_mean": 3.72021484375,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.5847001671791077,
+      "kl": 0.12943434715270996,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50392932.4556962,
+      "logits/rejected": -48968514.37037037,
+      "logps/chosen": -314.8338607594937,
+      "logps/rejected": -207.10426311728395,
+      "loss": 0.4994,
+      "loss/base_kto": 0.49842140078544617,
+      "metrics/advantage_var": 0.09590301662683487,
+      "regularization/advantage_range": 0.7024412155151367,
+      "regularization/tv": 0.0007024412043392658,
+      "rewards/chosen": 0.01595103287998634,
+      "rewards/margins": 0.012005401870425297,
+      "rewards/rejected": 0.0039456310095610445,
+      "stability/repetition_rate_mean": 0.20155483484268188,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 31.1264705657959,
+      "stability/response_length_var": 968.8571166992188,
+      "stability/token_entropy_mean": 3.685546875,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.5611183643341064,
+      "kl": 0.15796113014221191,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54706445.83783784,
+      "logits/rejected": -50905885.76744186,
+      "logps/chosen": -359.63999155405406,
+      "logps/rejected": -216.46538880813952,
+      "loss": 0.4991,
+      "loss/base_kto": 0.49846044182777405,
+      "metrics/advantage_var": 0.09960363060235977,
+      "regularization/advantage_range": 0.7743232846260071,
+      "regularization/tv": 0.0007743232999928296,
+      "rewards/chosen": 0.023018472903483623,
+      "rewards/margins": 0.011669156650265083,
+      "rewards/rejected": 0.01134931625321854,
+      "stability/repetition_rate_mean": 0.23383361101150513,
+      "stability/response_length_mean": 37.5,
+      "stability/response_length_std": 29.640705108642578,
+      "stability/response_length_var": 878.5714111328125,
+      "stability/token_entropy_mean": 4.150390625,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6100413799285889,
+      "kl": 0.1611875742673874,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59367004.530120485,
+      "logits/rejected": -53783964.25974026,
+      "logps/chosen": -399.5828313253012,
+      "logps/rejected": -209.8855265827922,
+      "loss": 0.4989,
+      "loss/base_kto": 0.49745798110961914,
+      "metrics/advantage_var": 0.18634693324565887,
+      "regularization/advantage_range": 1.0710378885269165,
+      "regularization/tv": 0.001071037957444787,
+      "rewards/chosen": 0.0319610503782709,
+      "rewards/margins": 0.020131756210715127,
+      "rewards/rejected": 0.011829294167555772,
+      "stability/repetition_rate_mean": 0.2053571343421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.41796875,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.5926874876022339,
+      "kl": 0.19908857345581055,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57380000.38554217,
+      "logits/rejected": -53610043.84415584,
+      "logps/chosen": -378.29080384036143,
+      "logps/rejected": -205.96319500811688,
+      "loss": 0.4988,
+      "loss/base_kto": 0.49805688858032227,
+      "metrics/advantage_var": 0.14875562489032745,
+      "regularization/advantage_range": 0.9510820508003235,
+      "regularization/tv": 0.0009510820964351296,
+      "rewards/chosen": 0.02724501885563494,
+      "rewards/margins": 0.01475072959987346,
+      "rewards/rejected": 0.01249428925576148,
+      "stability/repetition_rate_mean": 0.3234747052192688,
+      "stability/response_length_mean": 32.25,
+      "stability/response_length_std": 30.541540145874023,
+      "stability/response_length_var": 932.7857055664062,
+      "stability/token_entropy_mean": 4.4296875,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.5929023623466492,
+      "kl": 0.2150437831878662,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60607008.405063294,
+      "logits/rejected": -53416820.938271604,
+      "logps/chosen": -364.25111253955697,
+      "logps/rejected": -201.07520736882717,
+      "loss": 0.4986,
+      "loss/base_kto": 0.4972267150878906,
+      "metrics/advantage_var": 0.17768144607543945,
+      "regularization/advantage_range": 1.046006202697754,
+      "regularization/tv": 0.0010460062185302377,
+      "rewards/chosen": 0.03414275072797944,
+      "rewards/margins": 0.022008340011110225,
+      "rewards/rejected": 0.012134410716869213,
+      "stability/repetition_rate_mean": 0.3148462176322937,
+      "stability/response_length_mean": 47.625,
+      "stability/response_length_std": 27.21311378479004,
+      "stability/response_length_var": 740.5535888671875,
+      "stability/token_entropy_mean": 4.78125,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.5756809711456299,
+      "kl": 0.221540167927742,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57751137.04575163,
+      "logits/rejected": -54370886.51497006,
+      "logps/chosen": -383.87048100490193,
+      "logps/rejected": -212.07943488023952,
+      "loss": 0.4981,
+      "loss/base_kto": 0.496986448764801,
+      "metrics/advantage_var": 0.20863568782806396,
+      "regularization/advantage_range": 1.190858244895935,
+      "regularization/tv": 0.0011908583110198379,
+      "rewards/chosen": 0.03744382172628166,
+      "rewards/margins": 0.023750938765853644,
+      "rewards/rejected": 0.01369288296042802,
+      "stability/repetition_rate_mean": 0.1762429028749466,
+      "stability/response_length_mean": 27.875,
+      "stability/response_length_std": 30.31943130493164,
+      "stability/response_length_var": 919.2678833007812,
+      "stability/token_entropy_mean": 4.646484375,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.5690529942512512,
+      "kl": 0.2525067925453186,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60704396.2739726,
+      "logits/rejected": -55673891.31034483,
+      "logps/chosen": -417.02595248287673,
+      "logps/rejected": -216.25145923132183,
+      "loss": 0.4983,
+      "loss/base_kto": 0.4970117211341858,
+      "metrics/advantage_var": 0.32070493698120117,
+      "regularization/advantage_range": 1.4626210927963257,
+      "regularization/tv": 0.0014626210322603583,
+      "rewards/chosen": 0.035337840041069137,
+      "rewards/margins": 0.023764848746646767,
+      "rewards/rejected": 0.01157299129442237,
+      "stability/repetition_rate_mean": 0.15937499701976776,
+      "stability/response_length_mean": 15.875,
+      "stability/response_length_std": 24.584766387939453,
+      "stability/response_length_var": 604.4107055664062,
+      "stability/token_entropy_mean": 3.5,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.5786094069480896,
+      "kl": 0.2731947600841522,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57406563.09677419,
+      "logits/rejected": -53861922.13333333,
+      "logps/chosen": -392.2603326612903,
+      "logps/rejected": -183.81022727272727,
+      "loss": 0.4979,
+      "loss/base_kto": 0.4962916374206543,
+      "metrics/advantage_var": 0.4037338197231293,
+      "regularization/advantage_range": 1.604854941368103,
+      "regularization/tv": 0.0016048549441620708,
+      "rewards/chosen": 0.039759131400815904,
+      "rewards/margins": 0.028890261995011646,
+      "rewards/rejected": 0.010868869405804258,
+      "stability/repetition_rate_mean": 0.23874999582767487,
+      "stability/response_length_mean": 43.25,
+      "stability/response_length_std": 29.620214462280273,
+      "stability/response_length_var": 877.3571166992188,
+      "stability/token_entropy_mean": 3.6943359375,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5664066672325134,
+      "kl": 0.3012661039829254,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59866928.68711656,
+      "logits/rejected": -55031821.04458599,
+      "logps/chosen": -401.296227952454,
+      "logps/rejected": -247.69799960191082,
+      "loss": 0.4972,
+      "loss/base_kto": 0.49553966522216797,
+      "metrics/advantage_var": 0.4732483923435211,
+      "regularization/advantage_range": 1.8137600421905518,
+      "regularization/tv": 0.0018137602601200342,
+      "rewards/chosen": 0.051993937580132044,
+      "rewards/margins": 0.03472005898450045,
+      "rewards/rejected": 0.017273878595631592,
+      "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": 4.078125,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.6262583136558533,
+      "kl": 0.2736550271511078,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57111142.4,
+      "logits/rejected": -50093432.56774194,
+      "logps/chosen": -366.6800899621212,
+      "logps/rejected": -190.03079637096775,
+      "loss": 0.497,
+      "loss/base_kto": 0.49387773871421814,
+      "metrics/advantage_var": 0.812842845916748,
+      "regularization/advantage_range": 2.1453397274017334,
+      "regularization/tv": 0.0021453395020216703,
+      "rewards/chosen": 0.05096650557084517,
+      "rewards/margins": 0.050598360366485684,
+      "rewards/rejected": 0.00036814520435948523,
+      "stability/repetition_rate_mean": 0.3053385317325592,
+      "stability/response_length_mean": 52.75,
+      "stability/response_length_std": 22.72663688659668,
+      "stability/response_length_var": 516.5,
+      "stability/token_entropy_mean": 5.111328125,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.6019709706306458,
+      "kl": 0.3225500285625458,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51165295.03614458,
+      "logits/rejected": -51708233.14285714,
+      "logps/chosen": -329.8277249623494,
+      "logps/rejected": -199.37265371347402,
+      "loss": 0.497,
+      "loss/base_kto": 0.49590760469436646,
+      "metrics/advantage_var": 0.36499539017677307,
+      "regularization/advantage_range": 1.656163215637207,
+      "regularization/tv": 0.001656163134612143,
+      "rewards/chosen": 0.04440879821777344,
+      "rewards/margins": 0.03288166786169077,
+      "rewards/rejected": 0.011527130356082669,
+      "stability/repetition_rate_mean": 0.24838952720165253,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.73423957824707,
+      "stability/response_length_var": 884.125,
+      "stability/token_entropy_mean": 5.26953125,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.6047873497009277,
+      "kl": 0.3542978763580322,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55434553.22352941,
+      "logits/rejected": -50214345.38666666,
+      "logps/chosen": -397.2438419117647,
+      "logps/rejected": -207.591796875,
+      "loss": 0.4967,
+      "loss/base_kto": 0.49276694655418396,
+      "metrics/advantage_var": 0.5820550322532654,
+      "regularization/advantage_range": 2.0140323638916016,
+      "regularization/tv": 0.002014032332226634,
+      "rewards/chosen": 0.06905886706183939,
+      "rewards/margins": 0.055644914870168656,
+      "rewards/rejected": 0.013413952191670736,
+      "stability/repetition_rate_mean": 0.23066024482250214,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.193626403808594,
+      "stability/response_length_var": 852.2678833007812,
+      "stability/token_entropy_mean": 4.59375,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.561375617980957,
+      "kl": 0.351535439491272,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54049448.65882353,
+      "logits/rejected": -49388888.74666667,
+      "logps/chosen": -360.47366727941176,
+      "logps/rejected": -191.245,
+      "loss": 0.4968,
+      "loss/base_kto": 0.4943511188030243,
+      "metrics/advantage_var": 0.7565718293190002,
+      "regularization/advantage_range": 2.201819658279419,
+      "regularization/tv": 0.002201819559559226,
+      "rewards/chosen": 0.051560289719525504,
+      "rewards/margins": 0.04549936201058182,
+      "rewards/rejected": 0.006060927708943685,
+      "stability/repetition_rate_mean": 0.3171784281730652,
+      "stability/response_length_mean": 61.375,
+      "stability/response_length_std": 7.424621105194092,
+      "stability/response_length_var": 55.125,
+      "stability/token_entropy_mean": 4.544921875,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.5659539103507996,
+      "kl": 0.38903898000717163,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55283943.61904762,
+      "logits/rejected": -49011105.684210524,
+      "logps/chosen": -416.15206473214283,
+      "logps/rejected": -186.7262541118421,
+      "loss": 0.4963,
+      "loss/base_kto": 0.4927336871623993,
+      "metrics/advantage_var": 0.7581729888916016,
+      "regularization/advantage_range": 2.441723585128784,
+      "regularization/tv": 0.0024417235981673002,
+      "rewards/chosen": 0.07432699771154494,
+      "rewards/margins": 0.05721738717908548,
+      "rewards/rejected": 0.01710961053245946,
+      "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": 4.3837890625,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.6592298150062561,
+      "kl": 0.4146839678287506,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54472480.36781609,
+      "logits/rejected": -49860436.16438356,
+      "logps/chosen": -370.26210039511494,
+      "logps/rejected": -212.42588827054794,
+      "loss": 0.4954,
+      "loss/base_kto": 0.4939287304878235,
+      "metrics/advantage_var": 0.6743358969688416,
+      "regularization/advantage_range": 2.269331455230713,
+      "regularization/tv": 0.002269331831485033,
+      "rewards/chosen": 0.06546377861636808,
+      "rewards/margins": 0.048841037687927205,
+      "rewards/rejected": 0.016622740928440877,
+      "stability/repetition_rate_mean": 0.2411106526851654,
+      "stability/response_length_mean": 22.5,
+      "stability/response_length_std": 28.33977508544922,
+      "stability/response_length_var": 803.1428833007812,
+      "stability/token_entropy_mean": 4.87109375,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.564885139465332,
+      "kl": 0.5150820016860962,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59181152.94674556,
+      "logits/rejected": -52670104.58278146,
+      "logps/chosen": -440.4078679733728,
+      "logps/rejected": -190.99668874172184,
+      "loss": 0.4948,
+      "loss/base_kto": 0.49169474840164185,
+      "metrics/advantage_var": 0.832883358001709,
+      "regularization/advantage_range": 2.5218818187713623,
+      "regularization/tv": 0.0025218818336725235,
+      "rewards/chosen": 0.0933977894529083,
+      "rewards/margins": 0.06728294196652601,
+      "rewards/rejected": 0.02611484748638229,
+      "stability/repetition_rate_mean": 0.2678554654121399,
+      "stability/response_length_mean": 37.25,
+      "stability/response_length_std": 30.92733383178711,
+      "stability/response_length_var": 956.5,
+      "stability/token_entropy_mean": 3.80126953125,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.6160000562667847,
+      "kl": 0.5376589894294739,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56722012.530120485,
+      "logits/rejected": -48136594.28571428,
+      "logps/chosen": -355.58876129518075,
+      "logps/rejected": -194.6046570616883,
+      "loss": 0.4945,
+      "loss/base_kto": 0.4918275475502014,
+      "metrics/advantage_var": 1.3826684951782227,
+      "regularization/advantage_range": 3.0055034160614014,
+      "regularization/tv": 0.0030055036768317223,
+      "rewards/chosen": 0.08484269911984363,
+      "rewards/margins": 0.06455535258957336,
+      "rewards/rejected": 0.020287346530270266,
+      "stability/repetition_rate_mean": 0.2454427182674408,
+      "stability/response_length_mean": 39.75,
+      "stability/response_length_std": 27.473363876342773,
+      "stability/response_length_var": 754.7857055664062,
+      "stability/token_entropy_mean": 4.73046875,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.578485906124115,
+      "kl": 0.5820572972297668,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52728517.88957055,
+      "logits/rejected": -48089241.27388535,
+      "logps/chosen": -375.1566334355828,
+      "logps/rejected": -210.98556926751593,
+      "loss": 0.4938,
+      "loss/base_kto": 0.49145957827568054,
+      "metrics/advantage_var": 1.5378645658493042,
+      "regularization/advantage_range": 3.2784626483917236,
+      "regularization/tv": 0.0032784626819193363,
+      "rewards/chosen": 0.0989208455466054,
+      "rewards/margins": 0.06765583580624973,
+      "rewards/rejected": 0.03126500974035567,
+      "stability/repetition_rate_mean": 0.26171875,
+      "stability/response_length_mean": 40.375,
+      "stability/response_length_std": 32.61874008178711,
+      "stability/response_length_var": 1063.982177734375,
+      "stability/token_entropy_mean": 3.091796875,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.5877865552902222,
+      "kl": 0.6085312962532043,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54704355.55555555,
+      "logits/rejected": -49614323.037974685,
+      "logps/chosen": -311.0955825617284,
+      "logps/rejected": -196.88338113132912,
+      "loss": 0.4943,
+      "loss/base_kto": 0.49103212356567383,
+      "metrics/advantage_var": 2.168581485748291,
+      "regularization/advantage_range": 3.4313366413116455,
+      "regularization/tv": 0.003431336721405387,
+      "rewards/chosen": 0.09057903289794922,
+      "rewards/margins": 0.075000848951219,
+      "rewards/rejected": 0.015578183946730216,
+      "stability/repetition_rate_mean": 0.14541903138160706,
+      "stability/response_length_mean": 26.25,
+      "stability/response_length_std": 31.45404815673828,
+      "stability/response_length_var": 989.3571166992188,
+      "stability/token_entropy_mean": 3.7890625,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.6331533193588257,
+      "kl": 0.6512568593025208,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53173057.91616766,
+      "logits/rejected": -49808035.97385621,
+      "logps/chosen": -391.2273811751497,
+      "logps/rejected": -210.86637050653596,
+      "loss": 0.4933,
+      "loss/base_kto": 0.4893794655799866,
+      "metrics/advantage_var": 2.1304852962493896,
+      "regularization/advantage_range": 3.9502923488616943,
+      "regularization/tv": 0.003950292710214853,
+      "rewards/chosen": 0.11779353433026525,
+      "rewards/margins": 0.08457047255051998,
+      "rewards/rejected": 0.03322306177974527,
+      "stability/repetition_rate_mean": 0.1901041716337204,
+      "stability/response_length_mean": 17.625,
+      "stability/response_length_std": 28.680194854736328,
+      "stability/response_length_var": 822.5535888671875,
+      "stability/token_entropy_mean": 4.8427734375,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.5744519829750061,
+      "kl": 0.6606561541557312,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60185910.5032258,
+      "logits/rejected": -53942718.836363636,
+      "logps/chosen": -391.6687247983871,
+      "logps/rejected": -197.56677320075758,
+      "loss": 0.4937,
+      "loss/base_kto": 0.4902775287628174,
+      "metrics/advantage_var": 1.463286280632019,
+      "regularization/advantage_range": 3.479656219482422,
+      "regularization/tv": 0.003479656297713518,
+      "rewards/chosen": 0.11162065075289819,
+      "rewards/margins": 0.07636482559457669,
+      "rewards/rejected": 0.0352558251583215,
+      "stability/repetition_rate_mean": 0.42074424028396606,
+      "stability/response_length_mean": 52.875,
+      "stability/response_length_std": 22.850053787231445,
+      "stability/response_length_var": 522.125,
+      "stability/token_entropy_mean": 5.18359375,
+      "step": 195
+    },
+    {
+      "epoch": 1.0,
+      "step": 196,
+      "total_flos": 3.1861866257511875e+18,
+      "train_loss": 0.49831690745694296,
+      "train_runtime": 3155.7241,
+      "train_samples_per_second": 31.688,
+      "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_rho0.001/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c494ab7cafd29dea773d452695f132b657423e6fff5ae7d83d91a8b8b8e7981
+size 8017