Upload folder using huggingface_hub

mistral7b_kto_kl-beta0.1/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_kl
+  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_kl
+
+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_kl-beta0.1/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": [
+    "up_proj",
+    "k_proj",
+    "o_proj",
+    "v_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

mistral7b_kto_kl-beta0.1/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f05c1f1864ae5beccb095c536c167bd5ed57edef1462c4931ec4b590181233d7
+size 42002584

mistral7b_kto_kl-beta0.1/all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.5030260469232287,
+    "train_runtime": 3154.7919,
+    "train_samples_per_second": 31.698,
+    "train_steps_per_second": 0.062
+}

mistral7b_kto_kl-beta0.1/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_kl-beta0.1/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_kl-beta0.1/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": [
+    "up_proj",
+    "k_proj",
+    "o_proj",
+    "v_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

mistral7b_kto_kl-beta0.1/checkpoint-196/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f05c1f1864ae5beccb095c536c167bd5ed57edef1462c4931ec4b590181233d7
+size 42002584

mistral7b_kto_kl-beta0.1/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_kl-beta0.1/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 sha256:0cee56507b02ea73d4594c5f611123bd61f0c3fcaca0c4e2093edf3e23c03e5c
+size 31467333

mistral7b_kto_kl-beta0.1/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:be676fb9b8a3cf818cb94ceb2dd4367aa6c5d1be1cf4e9baa02a8a06ddd66d90
+size 31467333

mistral7b_kto_kl-beta0.1/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:e4f9a3a18e54318e434590740ea2235aec04c9d30ebc37b4845286bd79fcd4e5
+size 31467333

mistral7b_kto_kl-beta0.1/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 sha256:333904c9fb151632189826c833aa5e01f67352fe7a736688292ca92439159e95
+size 31467397

mistral7b_kto_kl-beta0.1/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 sha256:32ed34a600c18ef5cdd7d5f0ac4e7926bb9484b15397011943dd8043a0148bd6
+size 31467397

mistral7b_kto_kl-beta0.1/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 sha256:5be94e3a3a14a7b5a847e8ebe2b9f9c96bba2047ffad29b8ef854dcfc626a0e1
+size 31467397

mistral7b_kto_kl-beta0.1/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:e40f9f1741479ae4949742ca12614ffc3def7b8d2573973ed9174bb6b33a9afa
+size 31467397

mistral7b_kto_kl-beta0.1/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 sha256:8db11c4e4bdf92c2f247b44df7e8e6ebd25d17d7eff17d35ee9226b54e751930
+size 31467397

mistral7b_kto_kl-beta0.1/checkpoint-196/global_step196/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:165bcc2cfac8552454a3596a93c1b65f0e68e9cc192b8f50a98321d7a6d2ded6
+size 42288409

mistral7b_kto_kl-beta0.1/checkpoint-196/latest

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

mistral7b_kto_kl-beta0.1/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_kl-beta0.1/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_kl-beta0.1/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_kl-beta0.1/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_kl-beta0.1/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_kl-beta0.1/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_kl-beta0.1/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_kl-beta0.1/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_kl-beta0.1/checkpoint-196/scheduler.pt

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

mistral7b_kto_kl-beta0.1/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_kl-beta0.1/checkpoint-196/tokenizer.model

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

mistral7b_kto_kl-beta0.1/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_kl-beta0.1/checkpoint-196/trainer_state.json

@@ -0,0 +1,1004 @@
+{
+  "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": 1.1286565065383911,
+      "kl": 0.009834487922489643,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59720272.84210526,
+      "logits/rejected": -50419004.952380955,
+      "logps/chosen": -439.009765625,
+      "logps/rejected": -198.3979957217262,
+      "loss": 0.502,
+      "loss/base_kto": 0.4997194707393646,
+      "loss/total_with_kl": 0.5024515390396118,
+      "metrics/advantage_var": 0.0546402707695961,
+      "regularization/advantage_var": 0.0546402707695961,
+      "regularization/kl": 0.0027320135850459337,
+      "rewards/chosen": 0.0035851931101397463,
+      "rewards/margins": 0.002444372626773098,
+      "rewards/rejected": 0.0011408204833666484,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 1.0271365642547607,
+      "kl": 0.02555137872695923,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59294946.077922076,
+      "logits/rejected": -50014485.590361446,
+      "logps/chosen": -429.9174107142857,
+      "logps/rejected": -192.83664344879517,
+      "loss": 0.5032,
+      "loss/base_kto": 0.5001684427261353,
+      "loss/total_with_kl": 0.5038444995880127,
+      "metrics/advantage_var": 0.07352278381586075,
+      "regularization/advantage_var": 0.07352278381586075,
+      "regularization/kl": 0.0036761388182640076,
+      "rewards/chosen": 0.0021098085812159945,
+      "rewards/margins": -0.0010197663266154045,
+      "rewards/rejected": 0.003129574907831399,
+      "stability/repetition_rate_mean": 0.3766128420829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.369140625,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 1.1246036291122437,
+      "kl": 0.01402486115694046,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56229081.21212121,
+      "logits/rejected": -51912063.174193546,
+      "logps/chosen": -404.1037878787879,
+      "logps/rejected": -200.66993447580646,
+      "loss": 0.5033,
+      "loss/base_kto": 0.4999401271343231,
+      "loss/total_with_kl": 0.5033776164054871,
+      "metrics/advantage_var": 0.0687500610947609,
+      "regularization/advantage_var": 0.0687500610947609,
+      "regularization/kl": 0.0034375027753412724,
+      "rewards/chosen": 0.0027040712761156487,
+      "rewards/margins": 0.0002901616101157749,
+      "rewards/rejected": 0.002413909665999874,
+      "stability/repetition_rate_mean": 0.275390625,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.92341423034668,
+      "stability/response_length_var": 895.4107055664062,
+      "stability/token_entropy_mean": 3.6796875,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 1.0175650119781494,
+      "kl": 0.012321279384195805,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51667027.2,
+      "logits/rejected": -48065526.4,
+      "logps/chosen": -343.695458984375,
+      "logps/rejected": -225.64677734375,
+      "loss": 0.5033,
+      "loss/base_kto": 0.5002518892288208,
+      "loss/total_with_kl": 0.5026945471763611,
+      "metrics/advantage_var": 0.04885442927479744,
+      "regularization/advantage_var": 0.04885442927479744,
+      "regularization/kl": 0.0024427215103060007,
+      "rewards/chosen": 0.0001240897923707962,
+      "rewards/margins": -0.0015346232801675798,
+      "rewards/rejected": 0.001658713072538376,
+      "stability/repetition_rate_mean": 0.2021329402923584,
+      "stability/response_length_mean": 42.125,
+      "stability/response_length_std": 27.01553726196289,
+      "stability/response_length_var": 729.8392944335938,
+      "stability/token_entropy_mean": 4.43359375,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 1.0005894899368286,
+      "kl": 0.01893586292862892,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56474902.951724134,
+      "logits/rejected": -52020124.525714286,
+      "logps/chosen": -378.4993534482759,
+      "logps/rejected": -202.78513392857144,
+      "loss": 0.5032,
+      "loss/base_kto": 0.5001356601715088,
+      "loss/total_with_kl": 0.5026346445083618,
+      "metrics/advantage_var": 0.049978986382484436,
+      "regularization/advantage_var": 0.049978986382484436,
+      "regularization/kl": 0.0024989494122564793,
+      "rewards/chosen": -0.000791168212890625,
+      "rewards/margins": -0.0015635289464678084,
+      "rewards/rejected": 0.0007723607335771833,
+      "stability/repetition_rate_mean": 0.16825254261493683,
+      "stability/response_length_mean": 43.375,
+      "stability/response_length_std": 28.54538917541504,
+      "stability/response_length_var": 814.8392944335938,
+      "stability/token_entropy_mean": 4.763671875,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 1.047668695449829,
+      "kl": 0.019050994887948036,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57737765.798657715,
+      "logits/rejected": -51525733.80116959,
+      "logps/chosen": -468.53869546979865,
+      "logps/rejected": -198.22558022660817,
+      "loss": 0.5033,
+      "loss/base_kto": 0.4998214840888977,
+      "loss/total_with_kl": 0.5030900239944458,
+      "metrics/advantage_var": 0.0653710812330246,
+      "regularization/advantage_var": 0.0653710812330246,
+      "regularization/kl": 0.0032685536425560713,
+      "rewards/chosen": 0.004255342803545446,
+      "rewards/margins": 0.002025002526303064,
+      "rewards/rejected": 0.0022303402772423818,
+      "stability/repetition_rate_mean": 0.1662016361951828,
+      "stability/response_length_mean": 43.0,
+      "stability/response_length_std": 29.602123260498047,
+      "stability/response_length_var": 876.2857055664062,
+      "stability/token_entropy_mean": 4.587890625,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 2.3248555660247803,
+      "kl": 0.021593520417809486,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57636539.94936709,
+      "logits/rejected": -53914326.913580246,
+      "logps/chosen": -385.92696795886076,
+      "logps/rejected": -195.1241801697531,
+      "loss": 0.5036,
+      "loss/base_kto": 0.4999656677246094,
+      "loss/total_with_kl": 0.5032640695571899,
+      "metrics/advantage_var": 0.06596841663122177,
+      "regularization/advantage_var": 0.06596841663122177,
+      "regularization/kl": 0.0032984204590320587,
+      "rewards/chosen": 0.0009543428881258904,
+      "rewards/margins": 0.0007249228627183583,
+      "rewards/rejected": 0.00022942002540753213,
+      "stability/repetition_rate_mean": 0.3244977593421936,
+      "stability/response_length_mean": 56.875,
+      "stability/response_length_std": 20.152544021606445,
+      "stability/response_length_var": 406.125,
+      "stability/token_entropy_mean": 5.46484375,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.9593210816383362,
+      "kl": 0.01618729531764984,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56643635.2,
+      "logits/rejected": -50236915.2,
+      "logps/chosen": -421.9146484375,
+      "logps/rejected": -219.2841796875,
+      "loss": 0.5033,
+      "loss/base_kto": 0.5004779100418091,
+      "loss/total_with_kl": 0.5045997500419617,
+      "metrics/advantage_var": 0.08243563026189804,
+      "regularization/advantage_var": 0.08243563026189804,
+      "regularization/kl": 0.004121781792491674,
+      "rewards/chosen": -0.00120118148624897,
+      "rewards/margins": -0.003708549775183201,
+      "rewards/rejected": 0.002507368288934231,
+      "stability/repetition_rate_mean": 0.13671875,
+      "stability/response_length_mean": 33.125,
+      "stability/response_length_std": 33.01704025268555,
+      "stability/response_length_var": 1090.125,
+      "stability/token_entropy_mean": 4.8876953125,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 1.1373188495635986,
+      "kl": 0.0272687915712595,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52697338.84563758,
+      "logits/rejected": -50778291.649122804,
+      "logps/chosen": -352.29383389261744,
+      "logps/rejected": -243.8361659356725,
+      "loss": 0.5033,
+      "loss/base_kto": 0.5000495910644531,
+      "loss/total_with_kl": 0.5032973289489746,
+      "metrics/advantage_var": 0.06495420634746552,
+      "regularization/advantage_var": 0.06495420634746552,
+      "regularization/kl": 0.0032477104105055332,
+      "rewards/chosen": 0.003462229799104217,
+      "rewards/margins": -0.0003369061228871424,
+      "rewards/rejected": 0.0037991359219913595,
+      "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": 4.759765625,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.9557572603225708,
+      "kl": 0.027951324358582497,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54038854.44295302,
+      "logits/rejected": -48308792.88888889,
+      "logps/chosen": -375.13472630033556,
+      "logps/rejected": -180.2223935489766,
+      "loss": 0.5029,
+      "loss/base_kto": 0.49937987327575684,
+      "loss/total_with_kl": 0.5020853877067566,
+      "metrics/advantage_var": 0.05411091074347496,
+      "regularization/advantage_var": 0.05411091074347496,
+      "regularization/kl": 0.002705545863136649,
+      "rewards/chosen": 0.006440740703736376,
+      "rewards/margins": 0.004618997412015367,
+      "rewards/rejected": 0.0018217432917210093,
+      "stability/repetition_rate_mean": 0.4069356620311737,
+      "stability/response_length_mean": 55.375,
+      "stability/response_length_std": 16.361431121826172,
+      "stability/response_length_var": 267.6964416503906,
+      "stability/token_entropy_mean": 3.8125,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 1.5015521049499512,
+      "kl": 0.04579087719321251,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56948112.69565217,
+      "logits/rejected": -49574068.327044025,
+      "logps/chosen": -344.4958753881988,
+      "logps/rejected": -230.41398879716982,
+      "loss": 0.5034,
+      "loss/base_kto": 0.5002328753471375,
+      "loss/total_with_kl": 0.5030208826065063,
+      "metrics/advantage_var": 0.05576041340827942,
+      "regularization/advantage_var": 0.05576041340827942,
+      "regularization/kl": 0.0027880207635462284,
+      "rewards/chosen": 0.001614014924683186,
+      "rewards/margins": -0.0017921153681972743,
+      "rewards/rejected": 0.00340613029288046,
+      "stability/repetition_rate_mean": 0.103515625,
+      "stability/response_length_mean": 24.5,
+      "stability/response_length_std": 32.710853576660156,
+      "stability/response_length_var": 1070.0,
+      "stability/token_entropy_mean": 4.61328125,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 1.0588984489440918,
+      "kl": 0.03464470058679581,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52673585.64848485,
+      "logits/rejected": -47836140.18064516,
+      "logps/chosen": -359.6580492424242,
+      "logps/rejected": -178.18500504032258,
+      "loss": 0.5031,
+      "loss/base_kto": 0.49969473481178284,
+      "loss/total_with_kl": 0.5024840831756592,
+      "metrics/advantage_var": 0.0557870976626873,
+      "regularization/advantage_var": 0.0557870976626873,
+      "regularization/kl": 0.0027893546503037214,
+      "rewards/chosen": 0.006273941560225053,
+      "rewards/margins": 0.0020832758844772734,
+      "rewards/rejected": 0.0041906656757477794,
+      "stability/repetition_rate_mean": 0.1456473171710968,
+      "stability/response_length_mean": 39.25,
+      "stability/response_length_std": 32.07023620605469,
+      "stability/response_length_var": 1028.5,
+      "stability/token_entropy_mean": 3.931640625,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 1.038607120513916,
+      "kl": 0.03578026220202446,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57568037.98709677,
+      "logits/rejected": -47274167.07878788,
+      "logps/chosen": -381.87852822580646,
+      "logps/rejected": -183.5539891098485,
+      "loss": 0.503,
+      "loss/base_kto": 0.4994705319404602,
+      "loss/total_with_kl": 0.5023882985115051,
+      "metrics/advantage_var": 0.058355946093797684,
+      "regularization/advantage_var": 0.058355946093797684,
+      "regularization/kl": 0.002917797304689884,
+      "rewards/chosen": 0.00682763361161755,
+      "rewards/margins": 0.003990938027233671,
+      "rewards/rejected": 0.002836695584383878,
+      "stability/repetition_rate_mean": 0.1857219785451889,
+      "stability/response_length_mean": 20.625,
+      "stability/response_length_std": 27.176342010498047,
+      "stability/response_length_var": 738.5535888671875,
+      "stability/token_entropy_mean": 5.537109375,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.974204957485199,
+      "kl": 0.04303906485438347,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52785099.1483871,
+      "logits/rejected": -49040083.00606061,
+      "logps/chosen": -379.097404233871,
+      "logps/rejected": -220.38963068181818,
+      "loss": 0.5033,
+      "loss/base_kto": 0.500048816204071,
+      "loss/total_with_kl": 0.5032418966293335,
+      "metrics/advantage_var": 0.06386099010705948,
+      "regularization/advantage_var": 0.06386099010705948,
+      "regularization/kl": 0.0031930492259562016,
+      "rewards/chosen": 0.0031055967653951335,
+      "rewards/margins": -0.00014737946430142933,
+      "rewards/rejected": 0.003252976229696563,
+      "stability/repetition_rate_mean": 0.177734375,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.7109375,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.9141969084739685,
+      "kl": 0.03373977914452553,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49510382.7027027,
+      "logits/rejected": -47913043.34883721,
+      "logps/chosen": -314.18810705236484,
+      "logps/rejected": -228.88971656976744,
+      "loss": 0.5033,
+      "loss/base_kto": 0.4988802373409271,
+      "loss/total_with_kl": 0.5023767948150635,
+      "metrics/advantage_var": 0.06993144005537033,
+      "regularization/advantage_var": 0.06993144005537033,
+      "regularization/kl": 0.0034965721424669027,
+      "rewards/chosen": 0.008269893961983759,
+      "rewards/margins": 0.008756556226810444,
+      "rewards/rejected": -0.0004866622648266859,
+      "stability/repetition_rate_mean": 0.08308292925357819,
+      "stability/response_length_mean": 22.75,
+      "stability/response_length_std": 31.07479476928711,
+      "stability/response_length_var": 965.6428833007812,
+      "stability/token_entropy_mean": 2.158203125,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.998468279838562,
+      "kl": 0.04045908525586128,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58651403.867549665,
+      "logits/rejected": -53481308.40236686,
+      "logps/chosen": -403.97845095198676,
+      "logps/rejected": -201.0639330621302,
+      "loss": 0.503,
+      "loss/base_kto": 0.4992879033088684,
+      "loss/total_with_kl": 0.5029799342155457,
+      "metrics/advantage_var": 0.07383989542722702,
+      "regularization/advantage_var": 0.07383989542722702,
+      "regularization/kl": 0.0036919948179274797,
+      "rewards/chosen": 0.005166642713230966,
+      "rewards/margins": 0.005980765803433253,
+      "rewards/rejected": -0.0008141230902022864,
+      "stability/repetition_rate_mean": 0.18307292461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.25390625,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.8892827033996582,
+      "kl": 0.03160804137587547,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51504227.09677419,
+      "logits/rejected": -47165461.72121212,
+      "logps/chosen": -358.3259576612903,
+      "logps/rejected": -230.4444128787879,
+      "loss": 0.5029,
+      "loss/base_kto": 0.49965283274650574,
+      "loss/total_with_kl": 0.5036584734916687,
+      "metrics/advantage_var": 0.08011424541473389,
+      "regularization/advantage_var": 0.08011424541473389,
+      "regularization/kl": 0.004005712457001209,
+      "rewards/chosen": 0.0006552665464339718,
+      "rewards/margins": 0.0026155838402368687,
+      "rewards/rejected": -0.001960317293802897,
+      "stability/repetition_rate_mean": 0.1637834906578064,
+      "stability/response_length_mean": 25.875,
+      "stability/response_length_std": 28.792049407958984,
+      "stability/response_length_var": 828.9821166992188,
+      "stability/token_entropy_mean": 4.9140625,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 1.1928011178970337,
+      "kl": 0.04686165973544121,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58435885.94871795,
+      "logits/rejected": -50597822.43902439,
+      "logps/chosen": -380.74829727564105,
+      "logps/rejected": -187.2274080602134,
+      "loss": 0.5028,
+      "loss/base_kto": 0.49942442774772644,
+      "loss/total_with_kl": 0.5026821494102478,
+      "metrics/advantage_var": 0.06515531986951828,
+      "regularization/advantage_var": 0.06515531986951828,
+      "regularization/kl": 0.003257766366004944,
+      "rewards/chosen": 0.004109331048451937,
+      "rewards/margins": 0.00416946904576518,
+      "rewards/rejected": -6.013799731324359e-05,
+      "stability/repetition_rate_mean": 0.3110203742980957,
+      "stability/response_length_mean": 33.625,
+      "stability/response_length_std": 29.442134857177734,
+      "stability/response_length_var": 866.8392944335938,
+      "stability/token_entropy_mean": 3.7568359375,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.8256028294563293,
+      "kl": 0.043993134051561356,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50422958.98734177,
+      "logits/rejected": -49016667.654320985,
+      "logps/chosen": -314.97245846518985,
+      "logps/rejected": -207.1459297839506,
+      "loss": 0.5029,
+      "loss/base_kto": 0.4996494948863983,
+      "loss/total_with_kl": 0.5025039315223694,
+      "metrics/advantage_var": 0.057088762521743774,
+      "regularization/advantage_var": 0.057088762521743774,
+      "regularization/kl": 0.002854438265785575,
+      "rewards/chosen": 0.002092463117611559,
+      "rewards/margins": 0.0023125887506850007,
+      "rewards/rejected": -0.00022012563307344178,
+      "stability/repetition_rate_mean": 0.19374233484268188,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 31.1264705657959,
+      "stability/response_length_var": 968.8571166992188,
+      "stability/token_entropy_mean": 3.720703125,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.9563474655151367,
+      "kl": 0.04170067235827446,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54757957.18918919,
+      "logits/rejected": -50958068.093023255,
+      "logps/chosen": -359.7959512246622,
+      "logps/rejected": -216.56615643168604,
+      "loss": 0.5027,
+      "loss/base_kto": 0.4991760849952698,
+      "loss/total_with_kl": 0.5025018453598022,
+      "metrics/advantage_var": 0.06651490181684494,
+      "regularization/advantage_var": 0.06651490181684494,
+      "regularization/kl": 0.003325745230540633,
+      "rewards/chosen": 0.007421926872150318,
+      "rewards/margins": 0.0061471956020477505,
+      "rewards/rejected": 0.0012747312701025674,
+      "stability/repetition_rate_mean": 0.22173713147640228,
+      "stability/response_length_mean": 37.5,
+      "stability/response_length_std": 30.137779235839844,
+      "stability/response_length_var": 908.2857055664062,
+      "stability/token_entropy_mean": 4.244140625,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 1.0580958127975464,
+      "kl": 0.0368778295814991,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59432700.915662654,
+      "logits/rejected": -53874594.90909091,
+      "logps/chosen": -399.81428840361446,
+      "logps/rejected": -209.9729099025974,
+      "loss": 0.503,
+      "loss/base_kto": 0.49930858612060547,
+      "loss/total_with_kl": 0.5037649273872375,
+      "metrics/advantage_var": 0.08912745863199234,
+      "regularization/advantage_var": 0.08912745863199234,
+      "regularization/kl": 0.004456373397260904,
+      "rewards/chosen": 0.008813725896628506,
+      "rewards/margins": 0.005722130700523837,
+      "rewards/rejected": 0.003091595196104669,
+      "stability/repetition_rate_mean": 0.1916852593421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.35546875,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.9517597556114197,
+      "kl": 0.04355233162641525,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57453074.5060241,
+      "logits/rejected": -53690860.051948056,
+      "logps/chosen": -378.5363798945783,
+      "logps/rejected": -206.05907568993507,
+      "loss": 0.5029,
+      "loss/base_kto": 0.5001128911972046,
+      "loss/total_with_kl": 0.5033425688743591,
+      "metrics/advantage_var": 0.06459267437458038,
+      "regularization/advantage_var": 0.06459267437458038,
+      "regularization/kl": 0.0032296336721628904,
+      "rewards/chosen": 0.002688568040549037,
+      "rewards/margins": -0.00021768283571664982,
+      "rewards/rejected": 0.0029062508762656867,
+      "stability/repetition_rate_mean": 0.3362334370613098,
+      "stability/response_length_mean": 31.125,
+      "stability/response_length_std": 28.412460327148438,
+      "stability/response_length_var": 807.2678833007812,
+      "stability/token_entropy_mean": 4.0439453125,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.8893980979919434,
+      "kl": 0.04345473647117615,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60690730.12658228,
+      "logits/rejected": -53506616.88888889,
+      "logps/chosen": -364.50076641613924,
+      "logps/rejected": -201.16813753858025,
+      "loss": 0.5031,
+      "loss/base_kto": 0.4992709159851074,
+      "loss/total_with_kl": 0.5025972723960876,
+      "metrics/advantage_var": 0.06652548164129257,
+      "regularization/advantage_var": 0.06652548164129257,
+      "regularization/kl": 0.0033262737561017275,
+      "rewards/chosen": 0.009177133252349081,
+      "rewards/margins": 0.006336246738956951,
+      "rewards/rejected": 0.0028408865133921304,
+      "stability/repetition_rate_mean": 0.19457165896892548,
+      "stability/response_length_mean": 37.0,
+      "stability/response_length_std": 30.06184196472168,
+      "stability/response_length_var": 903.7142944335938,
+      "stability/token_entropy_mean": 4.916015625,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.9201934933662415,
+      "kl": 0.029955614358186722,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57835580.23529412,
+      "logits/rejected": -54506011.59281437,
+      "logps/chosen": -384.1818831699346,
+      "logps/rejected": -212.20988491766468,
+      "loss": 0.503,
+      "loss/base_kto": 0.4993014335632324,
+      "loss/total_with_kl": 0.5029810070991516,
+      "metrics/advantage_var": 0.07359055429697037,
+      "regularization/advantage_var": 0.07359055429697037,
+      "regularization/kl": 0.003679528133943677,
+      "rewards/chosen": 0.006303458431966944,
+      "rewards/margins": 0.005656958526209382,
+      "rewards/rejected": 0.0006464999057575614,
+      "stability/repetition_rate_mean": 0.18938209116458893,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.687660217285156,
+      "stability/response_length_var": 881.3571166992188,
+      "stability/token_entropy_mean": 4.56640625,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 1.0461677312850952,
+      "kl": 0.03671300411224365,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60829527.67123288,
+      "logits/rejected": -55803668.59770115,
+      "logps/chosen": -417.33845248287673,
+      "logps/rejected": -216.37194683908046,
+      "loss": 0.5031,
+      "loss/base_kto": 0.4993533194065094,
+      "loss/total_with_kl": 0.5025196671485901,
+      "metrics/advantage_var": 0.06332679837942123,
+      "regularization/advantage_var": 0.06332679837942123,
+      "regularization/kl": 0.003166340058669448,
+      "rewards/chosen": 0.0040864528041996365,
+      "rewards/margins": 0.004561904496967164,
+      "rewards/rejected": -0.00047545169276752694,
+      "stability/repetition_rate_mean": 0.16215278208255768,
+      "stability/response_length_mean": 15.25,
+      "stability/response_length_std": 24.93276596069336,
+      "stability/response_length_var": 621.6428833007812,
+      "stability/token_entropy_mean": 3.6328125,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.9537745714187622,
+      "kl": 0.03913918882608414,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57505567.38064516,
+      "logits/rejected": -53976231.56363636,
+      "logps/chosen": -392.59354838709675,
+      "logps/rejected": -183.94876893939394,
+      "loss": 0.503,
+      "loss/base_kto": 0.4988265633583069,
+      "loss/total_with_kl": 0.5021371841430664,
+      "metrics/advantage_var": 0.06621303409337997,
+      "regularization/advantage_var": 0.06621303409337997,
+      "regularization/kl": 0.0033106517512351274,
+      "rewards/chosen": 0.00643845142856721,
+      "rewards/margins": 0.009423405077683494,
+      "rewards/rejected": -0.002984953649116285,
+      "stability/repetition_rate_mean": 0.19009943306446075,
+      "stability/response_length_mean": 38.0,
+      "stability/response_length_std": 29.13515281677246,
+      "stability/response_length_var": 848.8571166992188,
+      "stability/token_entropy_mean": 3.91796875,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.9990471601486206,
+      "kl": 0.04277459532022476,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59995487.80368098,
+      "logits/rejected": -55185649.32484076,
+      "logps/chosen": -401.7632524923313,
+      "logps/rejected": -247.86673964968153,
+      "loss": 0.5032,
+      "loss/base_kto": 0.4993653893470764,
+      "loss/total_with_kl": 0.5040350556373596,
+      "metrics/advantage_var": 0.09339304268360138,
+      "regularization/advantage_var": 0.09339304268360138,
+      "regularization/kl": 0.004669652320444584,
+      "rewards/chosen": 0.005291132473506811,
+      "rewards/margins": 0.0048918331279284525,
+      "rewards/rejected": 0.0003992993455783577,
+      "stability/repetition_rate_mean": 0.2191840261220932,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 30.20406723022461,
+      "stability/response_length_var": 912.2857055664062,
+      "stability/token_entropy_mean": 3.416015625,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.8735517263412476,
+      "kl": 0.03989999368786812,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57242692.266666666,
+      "logits/rejected": -50260034.06451613,
+      "logps/chosen": -367.13778409090907,
+      "logps/rejected": -190.0336693548387,
+      "loss": 0.5028,
+      "loss/base_kto": 0.4993356168270111,
+      "loss/total_with_kl": 0.5030570030212402,
+      "metrics/advantage_var": 0.07442746311426163,
+      "regularization/advantage_var": 0.07442746311426163,
+      "regularization/kl": 0.0037213729228824377,
+      "rewards/chosen": 0.005194586696046772,
+      "rewards/margins": 0.0051130529737076335,
+      "rewards/rejected": 8.15337223391379e-05,
+      "stability/repetition_rate_mean": 0.3268229067325592,
+      "stability/response_length_mean": 52.75,
+      "stability/response_length_std": 22.72663688659668,
+      "stability/response_length_var": 516.5,
+      "stability/token_entropy_mean": 4.783203125,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.9999533891677856,
+      "kl": 0.047912172973155975,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51305524.43373494,
+      "logits/rejected": -51872003.32467532,
+      "logps/chosen": -330.2233151355422,
+      "logps/rejected": -199.49925172483765,
+      "loss": 0.5032,
+      "loss/base_kto": 0.4992775022983551,
+      "loss/total_with_kl": 0.502557635307312,
+      "metrics/advantage_var": 0.06560272723436356,
+      "regularization/advantage_var": 0.06560272723436356,
+      "regularization/kl": 0.003280136501416564,
+      "rewards/chosen": 0.004852092768772539,
+      "rewards/margins": 0.005986405444143338,
+      "rewards/rejected": -0.0011343126753707985,
+      "stability/repetition_rate_mean": 0.24838952720165253,
+      "stability/response_length_mean": 28.75,
+      "stability/response_length_std": 29.86517333984375,
+      "stability/response_length_var": 891.9285888671875,
+      "stability/token_entropy_mean": 5.314453125,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 1.1828984022140503,
+      "kl": 0.04824293777346611,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55574064.1882353,
+      "logits/rejected": -50405259.946666665,
+      "logps/chosen": -397.8425551470588,
+      "logps/rejected": -207.70329427083334,
+      "loss": 0.5029,
+      "loss/base_kto": 0.49907246232032776,
+      "loss/total_with_kl": 0.5030288100242615,
+      "metrics/advantage_var": 0.07912679761648178,
+      "regularization/advantage_var": 0.07912679761648178,
+      "regularization/kl": 0.003956340253353119,
+      "rewards/chosen": 0.009192322282230153,
+      "rewards/margins": 0.006929297634199555,
+      "rewards/rejected": 0.002263024648030599,
+      "stability/repetition_rate_mean": 0.20100826025009155,
+      "stability/response_length_mean": 24.125,
+      "stability/response_length_std": 25.637235641479492,
+      "stability/response_length_var": 657.2678833007812,
+      "stability/token_entropy_mean": 4.59765625,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 1.0469462871551514,
+      "kl": 0.04922879859805107,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54217438.870588236,
+      "logits/rejected": -49562466.986666664,
+      "logps/chosen": -360.94535845588234,
+      "logps/rejected": -191.27700520833332,
+      "loss": 0.503,
+      "loss/base_kto": 0.4997846782207489,
+      "loss/total_with_kl": 0.5039902925491333,
+      "metrics/advantage_var": 0.08411123603582382,
+      "regularization/advantage_var": 0.08411123603582382,
+      "regularization/kl": 0.004205561708658934,
+      "rewards/chosen": 0.004395649012397317,
+      "rewards/margins": 0.0015353484714732452,
+      "rewards/rejected": 0.002860300540924072,
+      "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": 4.6044921875,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.9236876964569092,
+      "kl": 0.04486330226063728,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55477516.190476194,
+      "logits/rejected": -49231777.684210524,
+      "logps/chosen": -416.8061755952381,
+      "logps/rejected": -186.89340049342104,
+      "loss": 0.5031,
+      "loss/base_kto": 0.49898120760917664,
+      "loss/total_with_kl": 0.5027669072151184,
+      "metrics/advantage_var": 0.07571418583393097,
+      "regularization/advantage_var": 0.07571418583393097,
+      "regularization/kl": 0.003785709384828806,
+      "rewards/chosen": 0.008917692161741712,
+      "rewards/margins": 0.008523785749585707,
+      "rewards/rejected": 0.00039390641215600466,
+      "stability/repetition_rate_mean": 0.291015625,
+      "stability/response_length_mean": 26.25,
+      "stability/response_length_std": 31.535694122314453,
+      "stability/response_length_var": 994.5,
+      "stability/token_entropy_mean": 3.771484375,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.9776092171669006,
+      "kl": 0.04533420130610466,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54658201.01149426,
+      "logits/rejected": -50095780.82191781,
+      "logps/chosen": -370.85578304597703,
+      "logps/rejected": -212.5767604880137,
+      "loss": 0.5029,
+      "loss/base_kto": 0.4993962347507477,
+      "loss/total_with_kl": 0.5035357475280762,
+      "metrics/advantage_var": 0.0827893316745758,
+      "regularization/advantage_var": 0.0827893316745758,
+      "regularization/kl": 0.004139466676861048,
+      "rewards/chosen": 0.00609781139198391,
+      "rewards/margins": 0.004562121499966231,
+      "rewards/rejected": 0.001535689892017678,
+      "stability/repetition_rate_mean": 0.20468750596046448,
+      "stability/response_length_mean": 12.625,
+      "stability/response_length_std": 21.460512161254883,
+      "stability/response_length_var": 460.5535583496094,
+      "stability/token_entropy_mean": 4.81640625,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.823735773563385,
+      "kl": 0.0640627071261406,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59383468.686390534,
+      "logits/rejected": -52950897.58940397,
+      "logps/chosen": -441.2536982248521,
+      "logps/rejected": -191.25607926324503,
+      "loss": 0.5029,
+      "loss/base_kto": 0.4989679455757141,
+      "loss/total_with_kl": 0.5037684440612793,
+      "metrics/advantage_var": 0.09601064026355743,
+      "regularization/advantage_var": 0.09601064026355743,
+      "regularization/kl": 0.004800532013177872,
+      "rewards/chosen": 0.008814329931721885,
+      "rewards/margins": 0.00863874767439423,
+      "rewards/rejected": 0.0001755822573276545,
+      "stability/repetition_rate_mean": 0.21061541140079498,
+      "stability/response_length_mean": 28.25,
+      "stability/response_length_std": 30.844078063964844,
+      "stability/response_length_var": 951.3571166992188,
+      "stability/token_entropy_mean": 3.33251953125,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 1.019155502319336,
+      "kl": 0.058815985918045044,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56930439.71084338,
+      "logits/rejected": -48423304.31168831,
+      "logps/chosen": -356.34346762048193,
+      "logps/rejected": -194.7735643262987,
+      "loss": 0.5026,
+      "loss/base_kto": 0.49923601746559143,
+      "loss/total_with_kl": 0.5028472542762756,
+      "metrics/advantage_var": 0.07222381234169006,
+      "regularization/advantage_var": 0.07222381234169006,
+      "regularization/kl": 0.0036111907102167606,
+      "rewards/chosen": 0.00937136253678655,
+      "rewards/margins": 0.005975195153221019,
+      "rewards/rejected": 0.003396167383565531,
+      "stability/repetition_rate_mean": 0.2878348231315613,
+      "stability/response_length_mean": 32.25,
+      "stability/response_length_std": 27.932315826416016,
+      "stability/response_length_var": 780.2142944335938,
+      "stability/token_entropy_mean": 4.72265625,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.9375177025794983,
+      "kl": 0.06844082474708557,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52935007.80368098,
+      "logits/rejected": -48337998.26751592,
+      "logps/chosen": -376.0254265720859,
+      "logps/rejected": -211.25706608280254,
+      "loss": 0.5028,
+      "loss/base_kto": 0.49898138642311096,
+      "loss/total_with_kl": 0.5028926134109497,
+      "metrics/advantage_var": 0.07822488993406296,
+      "regularization/advantage_var": 0.07822488993406296,
+      "regularization/kl": 0.003911244682967663,
+      "rewards/chosen": 0.012041092650290647,
+      "rewards/margins": 0.007926249172421917,
+      "rewards/rejected": 0.00411484347786873,
+      "stability/repetition_rate_mean": 0.162109375,
+      "stability/response_length_mean": 40.375,
+      "stability/response_length_std": 32.61874008178711,
+      "stability/response_length_var": 1063.982177734375,
+      "stability/token_entropy_mean": 3.357421875,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.935251772403717,
+      "kl": 0.0741070955991745,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54936120.88888889,
+      "logits/rejected": -49878696.50632911,
+      "logps/chosen": -311.8755787037037,
+      "logps/rejected": -196.9727675039557,
+      "loss": 0.5028,
+      "loss/base_kto": 0.4992446005344391,
+      "loss/total_with_kl": 0.503296971321106,
+      "metrics/advantage_var": 0.08104676008224487,
+      "regularization/advantage_var": 0.08104676008224487,
+      "regularization/kl": 0.004052338190376759,
+      "rewards/chosen": 0.012579193821659795,
+      "rewards/margins": 0.005940466211035505,
+      "rewards/rejected": 0.0066387276106242895,
+      "stability/repetition_rate_mean": 0.18797221779823303,
+      "stability/response_length_mean": 24.125,
+      "stability/response_length_std": 29.019390106201172,
+      "stability/response_length_var": 842.125,
+      "stability/token_entropy_mean": 3.87109375,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.986784815788269,
+      "kl": 0.06636197119951248,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53416365.22155689,
+      "logits/rejected": -50075306.666666664,
+      "logps/chosen": -392.29271145209583,
+      "logps/rejected": -211.18162785947712,
+      "loss": 0.5028,
+      "loss/base_kto": 0.4987066388130188,
+      "loss/total_with_kl": 0.5025004744529724,
+      "metrics/advantage_var": 0.07587640732526779,
+      "regularization/advantage_var": 0.07587640732526779,
+      "regularization/kl": 0.0037938205059617758,
+      "rewards/chosen": 0.011261776535810824,
+      "rewards/margins": 0.009564409612522242,
+      "rewards/rejected": 0.0016973669232885822,
+      "stability/repetition_rate_mean": 0.1822916716337204,
+      "stability/response_length_mean": 17.625,
+      "stability/response_length_std": 28.680194854736328,
+      "stability/response_length_var": 822.5535888671875,
+      "stability/token_entropy_mean": 4.8212890625,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 1.358068823814392,
+      "kl": 0.05749697610735893,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60522562.06451613,
+      "logits/rejected": -54295080.339393936,
+      "logps/chosen": -392.64884072580645,
+      "logps/rejected": -197.9084872159091,
+      "loss": 0.5029,
+      "loss/base_kto": 0.49840664863586426,
+      "loss/total_with_kl": 0.5019518733024597,
+      "metrics/advantage_var": 0.07090426236391068,
+      "regularization/advantage_var": 0.07090426236391068,
+      "regularization/kl": 0.0035452134907245636,
+      "rewards/chosen": 0.013610498366817351,
+      "rewards/margins": 0.012526261972542033,
+      "rewards/rejected": 0.0010842363942753185,
+      "stability/repetition_rate_mean": 0.4140625,
+      "stability/response_length_mean": 48.125,
+      "stability/response_length_std": 29.39600372314453,
+      "stability/response_length_var": 864.125,
+      "stability/token_entropy_mean": 4.4609375,
+      "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_kl-beta0.1/checkpoint-196/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3dc25d257c329c7b76aee6e6676887abb30f25f66f6b540e02a495bf57fdef9
+size 8017

mistral7b_kto_kl-beta0.1/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_kl-beta0.1/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_kl-beta0.1/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": 3.369140625, "stability/repetition_rate_mean": 0.3766128420829773, "step": 10}
+{"stability/response_length_mean": 29.375, "stability/response_length_std": 29.92341423034668, "stability/response_length_var": 895.4107055664062, "stability/token_entropy_mean": 3.6796875, "stability/repetition_rate_mean": 0.275390625, "step": 15}
+{"stability/response_length_mean": 42.125, "stability/response_length_std": 27.01553726196289, "stability/response_length_var": 729.8392944335938, "stability/token_entropy_mean": 4.43359375, "stability/repetition_rate_mean": 0.2021329402923584, "step": 20}
+{"stability/response_length_mean": 43.375, "stability/response_length_std": 28.54538917541504, "stability/response_length_var": 814.8392944335938, "stability/token_entropy_mean": 4.763671875, "stability/repetition_rate_mean": 0.16825254261493683, "step": 25}
+{"stability/response_length_mean": 43.0, "stability/response_length_std": 29.602123260498047, "stability/response_length_var": 876.2857055664062, "stability/token_entropy_mean": 4.587890625, "stability/repetition_rate_mean": 0.1662016361951828, "step": 30}
+{"stability/response_length_mean": 56.875, "stability/response_length_std": 20.152544021606445, "stability/response_length_var": 406.125, "stability/token_entropy_mean": 5.46484375, "stability/repetition_rate_mean": 0.3244977593421936, "step": 35}
+{"stability/response_length_mean": 33.125, "stability/response_length_std": 33.01704025268555, "stability/response_length_var": 1090.125, "stability/token_entropy_mean": 4.8876953125, "stability/repetition_rate_mean": 0.13671875, "step": 40}
+{"stability/response_length_mean": 28.875, "stability/response_length_std": 29.87324333190918, "stability/response_length_var": 892.4107055664062, "stability/token_entropy_mean": 4.759765625, "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": 3.8125, "stability/repetition_rate_mean": 0.4069356620311737, "step": 50}
+{"stability/response_length_mean": 24.5, "stability/response_length_std": 32.710853576660156, "stability/response_length_var": 1070.0, "stability/token_entropy_mean": 4.61328125, "stability/repetition_rate_mean": 0.103515625, "step": 55}
+{"stability/response_length_mean": 39.25, "stability/response_length_std": 32.07023620605469, "stability/response_length_var": 1028.5, "stability/token_entropy_mean": 3.931640625, "stability/repetition_rate_mean": 0.1456473171710968, "step": 60}
+{"stability/response_length_mean": 20.625, "stability/response_length_std": 27.176342010498047, "stability/response_length_var": 738.5535888671875, "stability/token_entropy_mean": 5.537109375, "stability/repetition_rate_mean": 0.1857219785451889, "step": 65}
+{"stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.7109375, "stability/repetition_rate_mean": 0.177734375, "step": 70}
+{"stability/response_length_mean": 22.75, "stability/response_length_std": 31.07479476928711, "stability/response_length_var": 965.6428833007812, "stability/token_entropy_mean": 2.158203125, "stability/repetition_rate_mean": 0.08308292925357819, "step": 75}
+{"stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.25390625, "stability/repetition_rate_mean": 0.18307292461395264, "step": 80}
+{"stability/response_length_mean": 25.875, "stability/response_length_std": 28.792049407958984, "stability/response_length_var": 828.9821166992188, "stability/token_entropy_mean": 4.9140625, "stability/repetition_rate_mean": 0.1637834906578064, "step": 85}
+{"stability/response_length_mean": 33.625, "stability/response_length_std": 29.442134857177734, "stability/response_length_var": 866.8392944335938, "stability/token_entropy_mean": 3.7568359375, "stability/repetition_rate_mean": 0.3110203742980957, "step": 90}
+{"stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.720703125, "stability/repetition_rate_mean": 0.19374233484268188, "step": 95}
+{"stability/response_length_mean": 37.5, "stability/response_length_std": 30.137779235839844, "stability/response_length_var": 908.2857055664062, "stability/token_entropy_mean": 4.244140625, "stability/repetition_rate_mean": 0.22173713147640228, "step": 100}
+{"stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.35546875, "stability/repetition_rate_mean": 0.1916852593421936, "step": 105}
+{"stability/response_length_mean": 31.125, "stability/response_length_std": 28.412460327148438, "stability/response_length_var": 807.2678833007812, "stability/token_entropy_mean": 4.0439453125, "stability/repetition_rate_mean": 0.3362334370613098, "step": 110}
+{"stability/response_length_mean": 37.0, "stability/response_length_std": 30.06184196472168, "stability/response_length_var": 903.7142944335938, "stability/token_entropy_mean": 4.916015625, "stability/repetition_rate_mean": 0.19457165896892548, "step": 115}
+{"stability/response_length_mean": 29.25, "stability/response_length_std": 29.687660217285156, "stability/response_length_var": 881.3571166992188, "stability/token_entropy_mean": 4.56640625, "stability/repetition_rate_mean": 0.18938209116458893, "step": 120}
+{"stability/response_length_mean": 15.25, "stability/response_length_std": 24.93276596069336, "stability/response_length_var": 621.6428833007812, "stability/token_entropy_mean": 3.6328125, "stability/repetition_rate_mean": 0.16215278208255768, "step": 125}
+{"stability/response_length_mean": 38.0, "stability/response_length_std": 29.13515281677246, "stability/response_length_var": 848.8571166992188, "stability/token_entropy_mean": 3.91796875, "stability/repetition_rate_mean": 0.19009943306446075, "step": 130}
+{"stability/response_length_mean": 29.0, "stability/response_length_std": 30.20406723022461, "stability/response_length_var": 912.2857055664062, "stability/token_entropy_mean": 3.416015625, "stability/repetition_rate_mean": 0.2191840261220932, "step": 135}
+{"stability/response_length_mean": 52.75, "stability/response_length_std": 22.72663688659668, "stability/response_length_var": 516.5, "stability/token_entropy_mean": 4.783203125, "stability/repetition_rate_mean": 0.3268229067325592, "step": 140}
+{"stability/response_length_mean": 28.75, "stability/response_length_std": 29.86517333984375, "stability/response_length_var": 891.9285888671875, "stability/token_entropy_mean": 5.314453125, "stability/repetition_rate_mean": 0.24838952720165253, "step": 145}
+{"stability/response_length_mean": 24.125, "stability/response_length_std": 25.637235641479492, "stability/response_length_var": 657.2678833007812, "stability/token_entropy_mean": 4.59765625, "stability/repetition_rate_mean": 0.20100826025009155, "step": 150}
+{"stability/response_length_mean": 61.375, "stability/response_length_std": 7.424621105194092, "stability/response_length_var": 55.125, "stability/token_entropy_mean": 4.6044921875, "stability/repetition_rate_mean": 0.3171784281730652, "step": 155}
+{"stability/response_length_mean": 26.25, "stability/response_length_std": 31.535694122314453, "stability/response_length_var": 994.5, "stability/token_entropy_mean": 3.771484375, "stability/repetition_rate_mean": 0.291015625, "step": 160}
+{"stability/response_length_mean": 12.625, "stability/response_length_std": 21.460512161254883, "stability/response_length_var": 460.5535583496094, "stability/token_entropy_mean": 4.81640625, "stability/repetition_rate_mean": 0.20468750596046448, "step": 165}
+{"stability/response_length_mean": 28.25, "stability/response_length_std": 30.844078063964844, "stability/response_length_var": 951.3571166992188, "stability/token_entropy_mean": 3.33251953125, "stability/repetition_rate_mean": 0.21061541140079498, "step": 170}
+{"stability/response_length_mean": 32.25, "stability/response_length_std": 27.932315826416016, "stability/response_length_var": 780.2142944335938, "stability/token_entropy_mean": 4.72265625, "stability/repetition_rate_mean": 0.2878348231315613, "step": 175}
+{"stability/response_length_mean": 40.375, "stability/response_length_std": 32.61874008178711, "stability/response_length_var": 1063.982177734375, "stability/token_entropy_mean": 3.357421875, "stability/repetition_rate_mean": 0.162109375, "step": 180}
+{"stability/response_length_mean": 24.125, "stability/response_length_std": 29.019390106201172, "stability/response_length_var": 842.125, "stability/token_entropy_mean": 3.87109375, "stability/repetition_rate_mean": 0.18797221779823303, "step": 185}
+{"stability/response_length_mean": 17.625, "stability/response_length_std": 28.680194854736328, "stability/response_length_var": 822.5535888671875, "stability/token_entropy_mean": 4.8212890625, "stability/repetition_rate_mean": 0.1822916716337204, "step": 190}
+{"stability/response_length_mean": 48.125, "stability/response_length_std": 29.39600372314453, "stability/response_length_var": 864.125, "stability/token_entropy_mean": 4.4609375, "stability/repetition_rate_mean": 0.4140625, "step": 195}

mistral7b_kto_kl-beta0.1/tokenizer.model

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

mistral7b_kto_kl-beta0.1/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_kl-beta0.1/train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.5030260469232287,
+    "train_runtime": 3154.7919,
+    "train_samples_per_second": 31.698,
+    "train_steps_per_second": 0.062
+}

mistral7b_kto_kl-beta0.1/trainer_log.jsonl

@@ -0,0 +1,40 @@
+{"current_steps": 5, "total_steps": 196, "loss": 0.502, "lr": 4e-08, "epoch": 0.025591810620601407, "percentage": 2.55, "elapsed_time": "0:01:18", "remaining_time": "0:50:04", "rewards/chosen": 0.0035851931101397463, "logps/chosen": -439.009765625, "logits/chosen": -59720272.84210526, "rewards/rejected": 0.0011408204833666484, "logps/rejected": -198.3979957217262, "logits/rejected": -50419004.952380955, "rewards/margins": 0.002444372626773098, "regularization/kl": 0.0027320135850459337, "regularization/advantage_var": 0.0546402707695961, "kl": 0.009834487922489643}
+{"current_steps": 10, "total_steps": 196, "loss": 0.5032, "lr": 9e-08, "epoch": 0.05118362124120281, "percentage": 5.1, "elapsed_time": "0:02:39", "remaining_time": "0:49:25", "rewards/chosen": 0.0021098085812159945, "logps/chosen": -429.9174107142857, "logits/chosen": -59294946.077922076, "rewards/rejected": 0.003129574907831399, "logps/rejected": -192.83664344879517, "logits/rejected": -50014485.590361446, "rewards/margins": -0.0010197663266154045, "regularization/kl": 0.0036761388182640076, "regularization/advantage_var": 0.07352278381586075, "kl": 0.02555137872695923, "stability/response_length_mean": 46.125, "stability/response_length_std": 27.231481552124023, "stability/response_length_var": 741.5535888671875, "stability/token_entropy_mean": 3.369140625, "stability/repetition_rate_mean": 0.3766128420829773}
+{"current_steps": 15, "total_steps": 196, "loss": 0.5033, "lr": 1.4e-07, "epoch": 0.07677543186180422, "percentage": 7.65, "elapsed_time": "0:03:59", "remaining_time": "0:48:07", "rewards/chosen": 0.0027040712761156487, "logps/chosen": -404.1037878787879, "logits/chosen": -56229081.21212121, "rewards/rejected": 0.002413909665999874, "logps/rejected": -200.66993447580646, "logits/rejected": -51912063.174193546, "rewards/margins": 0.0002901616101157749, "regularization/kl": 0.0034375027753412724, "regularization/advantage_var": 0.0687500610947609, "kl": 0.01402486115694046, "stability/response_length_mean": 29.375, "stability/response_length_std": 29.92341423034668, "stability/response_length_var": 895.4107055664062, "stability/token_entropy_mean": 3.6796875, "stability/repetition_rate_mean": 0.275390625}
+{"current_steps": 20, "total_steps": 196, "loss": 0.5033, "lr": 1.8999999999999998e-07, "epoch": 0.10236724248240563, "percentage": 10.2, "elapsed_time": "0:05:20", "remaining_time": "0:47:02", "rewards/chosen": 0.0001240897923707962, "logps/chosen": -343.695458984375, "logits/chosen": -51667027.2, "rewards/rejected": 0.001658713072538376, "logps/rejected": -225.64677734375, "logits/rejected": -48065526.4, "rewards/margins": -0.0015346232801675798, "regularization/kl": 0.0024427215103060007, "regularization/advantage_var": 0.04885442927479744, "kl": 0.012321279384195805, "stability/response_length_mean": 42.125, "stability/response_length_std": 27.01553726196289, "stability/response_length_var": 729.8392944335938, "stability/token_entropy_mean": 4.43359375, "stability/repetition_rate_mean": 0.2021329402923584}
+{"current_steps": 25, "total_steps": 196, "loss": 0.5032, "lr": 2.4e-07, "epoch": 0.12795905310300704, "percentage": 12.76, "elapsed_time": "0:06:40", "remaining_time": "0:45:40", "rewards/chosen": -0.000791168212890625, "logps/chosen": -378.4993534482759, "logits/chosen": -56474902.951724134, "rewards/rejected": 0.0007723607335771833, "logps/rejected": -202.78513392857144, "logits/rejected": -52020124.525714286, "rewards/margins": -0.0015635289464678084, "regularization/kl": 0.0024989494122564793, "regularization/advantage_var": 0.049978986382484436, "kl": 0.01893586292862892, "stability/response_length_mean": 43.375, "stability/response_length_std": 28.54538917541504, "stability/response_length_var": 814.8392944335938, "stability/token_entropy_mean": 4.763671875, "stability/repetition_rate_mean": 0.16825254261493683}
+{"current_steps": 30, "total_steps": 196, "loss": 0.5033, "lr": 2.9e-07, "epoch": 0.15355086372360843, "percentage": 15.31, "elapsed_time": "0:08:00", "remaining_time": "0:44:19", "rewards/chosen": 0.004255342803545446, "logps/chosen": -468.53869546979865, "logits/chosen": -57737765.798657715, "rewards/rejected": 0.0022303402772423818, "logps/rejected": -198.22558022660817, "logits/rejected": -51525733.80116959, "rewards/margins": 0.002025002526303064, "regularization/kl": 0.0032685536425560713, "regularization/advantage_var": 0.0653710812330246, "kl": 0.019050994887948036, "stability/response_length_mean": 43.0, "stability/response_length_std": 29.602123260498047, "stability/response_length_var": 876.2857055664062, "stability/token_entropy_mean": 4.587890625, "stability/repetition_rate_mean": 0.1662016361951828}
+{"current_steps": 35, "total_steps": 196, "loss": 0.5036, "lr": 3.4000000000000003e-07, "epoch": 0.17914267434420986, "percentage": 17.86, "elapsed_time": "0:09:21", "remaining_time": "0:43:05", "rewards/chosen": 0.0009543428881258904, "logps/chosen": -385.92696795886076, "logits/chosen": -57636539.94936709, "rewards/rejected": 0.00022942002540753213, "logps/rejected": -195.1241801697531, "logits/rejected": -53914326.913580246, "rewards/margins": 0.0007249228627183583, "regularization/kl": 0.0032984204590320587, "regularization/advantage_var": 0.06596841663122177, "kl": 0.021593520417809486, "stability/response_length_mean": 56.875, "stability/response_length_std": 20.152544021606445, "stability/response_length_var": 406.125, "stability/token_entropy_mean": 5.46484375, "stability/repetition_rate_mean": 0.3244977593421936}
+{"current_steps": 40, "total_steps": 196, "loss": 0.5033, "lr": 3.8999999999999997e-07, "epoch": 0.20473448496481125, "percentage": 20.41, "elapsed_time": "0:10:43", "remaining_time": "0:41:47", "rewards/chosen": -0.00120118148624897, "logps/chosen": -421.9146484375, "logits/chosen": -56643635.2, "rewards/rejected": 0.002507368288934231, "logps/rejected": -219.2841796875, "logits/rejected": -50236915.2, "rewards/margins": -0.003708549775183201, "regularization/kl": 0.004121781792491674, "regularization/advantage_var": 0.08243563026189804, "kl": 0.01618729531764984, "stability/response_length_mean": 33.125, "stability/response_length_std": 33.01704025268555, "stability/response_length_var": 1090.125, "stability/token_entropy_mean": 4.8876953125, "stability/repetition_rate_mean": 0.13671875}
+{"current_steps": 45, "total_steps": 196, "loss": 0.5033, "lr": 4.3999999999999997e-07, "epoch": 0.23032629558541268, "percentage": 22.96, "elapsed_time": "0:12:03", "remaining_time": "0:40:28", "rewards/chosen": 0.003462229799104217, "logps/chosen": -352.29383389261744, "logits/chosen": -52697338.84563758, "rewards/rejected": 0.0037991359219913595, "logps/rejected": -243.8361659356725, "logits/rejected": -50778291.649122804, "rewards/margins": -0.0003369061228871424, "regularization/kl": 0.0032477104105055332, "regularization/advantage_var": 0.06495420634746552, "kl": 0.0272687915712595, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.87324333190918, "stability/response_length_var": 892.4107055664062, "stability/token_entropy_mean": 4.759765625, "stability/repetition_rate_mean": 0.0935424193739891}
+{"current_steps": 50, "total_steps": 196, "loss": 0.5029, "lr": 4.9e-07, "epoch": 0.2559181062060141, "percentage": 25.51, "elapsed_time": "0:13:23", "remaining_time": "0:39:06", "rewards/chosen": 0.006440740703736376, "logps/chosen": -375.13472630033556, "logits/chosen": -54038854.44295302, "rewards/rejected": 0.0018217432917210093, "logps/rejected": -180.2223935489766, "logits/rejected": -48308792.88888889, "rewards/margins": 0.004618997412015367, "regularization/kl": 0.002705545863136649, "regularization/advantage_var": 0.05411091074347496, "kl": 0.027951324358582497, "stability/response_length_mean": 55.375, "stability/response_length_std": 16.361431121826172, "stability/response_length_var": 267.6964416503906, "stability/token_entropy_mean": 3.8125, "stability/repetition_rate_mean": 0.4069356620311737}
+{"current_steps": 55, "total_steps": 196, "loss": 0.5034, "lr": 5.4e-07, "epoch": 0.28150991682661547, "percentage": 28.06, "elapsed_time": "0:14:43", "remaining_time": "0:37:44", "rewards/chosen": 0.001614014924683186, "logps/chosen": -344.4958753881988, "logits/chosen": -56948112.69565217, "rewards/rejected": 0.00340613029288046, "logps/rejected": -230.41398879716982, "logits/rejected": -49574068.327044025, "rewards/margins": -0.0017921153681972743, "regularization/kl": 0.0027880207635462284, "regularization/advantage_var": 0.05576041340827942, "kl": 0.04579087719321251, "stability/response_length_mean": 24.5, "stability/response_length_std": 32.710853576660156, "stability/response_length_var": 1070.0, "stability/token_entropy_mean": 4.61328125, "stability/repetition_rate_mean": 0.103515625}
+{"current_steps": 60, "total_steps": 196, "loss": 0.5031, "lr": 5.9e-07, "epoch": 0.30710172744721687, "percentage": 30.61, "elapsed_time": "0:16:04", "remaining_time": "0:36:27", "rewards/chosen": 0.006273941560225053, "logps/chosen": -359.6580492424242, "logits/chosen": -52673585.64848485, "rewards/rejected": 0.0041906656757477794, "logps/rejected": -178.18500504032258, "logits/rejected": -47836140.18064516, "rewards/margins": 0.0020832758844772734, "regularization/kl": 0.0027893546503037214, "regularization/advantage_var": 0.0557870976626873, "kl": 0.03464470058679581, "stability/response_length_mean": 39.25, "stability/response_length_std": 32.07023620605469, "stability/response_length_var": 1028.5, "stability/token_entropy_mean": 3.931640625, "stability/repetition_rate_mean": 0.1456473171710968}
+{"current_steps": 65, "total_steps": 196, "loss": 0.503, "lr": 6.4e-07, "epoch": 0.3326935380678183, "percentage": 33.16, "elapsed_time": "0:17:25", "remaining_time": "0:35:07", "rewards/chosen": 0.00682763361161755, "logps/chosen": -381.87852822580646, "logits/chosen": -57568037.98709677, "rewards/rejected": 0.002836695584383878, "logps/rejected": -183.5539891098485, "logits/rejected": -47274167.07878788, "rewards/margins": 0.003990938027233671, "regularization/kl": 0.002917797304689884, "regularization/advantage_var": 0.058355946093797684, "kl": 0.03578026220202446, "stability/response_length_mean": 20.625, "stability/response_length_std": 27.176342010498047, "stability/response_length_var": 738.5535888671875, "stability/token_entropy_mean": 5.537109375, "stability/repetition_rate_mean": 0.1857219785451889}
+{"current_steps": 70, "total_steps": 196, "loss": 0.5033, "lr": 6.9e-07, "epoch": 0.3582853486884197, "percentage": 35.71, "elapsed_time": "0:18:45", "remaining_time": "0:33:46", "rewards/chosen": 0.0031055967653951335, "logps/chosen": -379.097404233871, "logits/chosen": -52785099.1483871, "rewards/rejected": 0.003252976229696563, "logps/rejected": -220.38963068181818, "logits/rejected": -49040083.00606061, "rewards/margins": -0.00014737946430142933, "regularization/kl": 0.0031930492259562016, "regularization/advantage_var": 0.06386099010705948, "kl": 0.04303906485438347, "stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.7109375, "stability/repetition_rate_mean": 0.177734375}
+{"current_steps": 75, "total_steps": 196, "loss": 0.5033, "lr": 7.4e-07, "epoch": 0.3838771593090211, "percentage": 38.27, "elapsed_time": "0:20:06", "remaining_time": "0:32:26", "rewards/chosen": 0.008269893961983759, "logps/chosen": -314.18810705236484, "logits/chosen": -49510382.7027027, "rewards/rejected": -0.0004866622648266859, "logps/rejected": -228.88971656976744, "logits/rejected": -47913043.34883721, "rewards/margins": 0.008756556226810444, "regularization/kl": 0.0034965721424669027, "regularization/advantage_var": 0.06993144005537033, "kl": 0.03373977914452553, "stability/response_length_mean": 22.75, "stability/response_length_std": 31.07479476928711, "stability/response_length_var": 965.6428833007812, "stability/token_entropy_mean": 2.158203125, "stability/repetition_rate_mean": 0.08308292925357819}
+{"current_steps": 80, "total_steps": 196, "loss": 0.503, "lr": 7.9e-07, "epoch": 0.4094689699296225, "percentage": 40.82, "elapsed_time": "0:21:27", "remaining_time": "0:31:07", "rewards/chosen": 0.005166642713230966, "logps/chosen": -403.97845095198676, "logits/chosen": -58651403.867549665, "rewards/rejected": -0.0008141230902022864, "logps/rejected": -201.0639330621302, "logits/rejected": -53481308.40236686, "rewards/margins": 0.005980765803433253, "regularization/kl": 0.0036919948179274797, "regularization/advantage_var": 0.07383989542722702, "kl": 0.04045908525586128, "stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.25390625, "stability/repetition_rate_mean": 0.18307292461395264}
+{"current_steps": 85, "total_steps": 196, "loss": 0.5029, "lr": 8.399999999999999e-07, "epoch": 0.4350607805502239, "percentage": 43.37, "elapsed_time": "0:22:47", "remaining_time": "0:29:46", "rewards/chosen": 0.0006552665464339718, "logps/chosen": -358.3259576612903, "logits/chosen": -51504227.09677419, "rewards/rejected": -0.001960317293802897, "logps/rejected": -230.4444128787879, "logits/rejected": -47165461.72121212, "rewards/margins": 0.0026155838402368687, "regularization/kl": 0.004005712457001209, "regularization/advantage_var": 0.08011424541473389, "kl": 0.03160804137587547, "stability/response_length_mean": 25.875, "stability/response_length_std": 28.792049407958984, "stability/response_length_var": 828.9821166992188, "stability/token_entropy_mean": 4.9140625, "stability/repetition_rate_mean": 0.1637834906578064}
+{"current_steps": 90, "total_steps": 196, "loss": 0.5028, "lr": 8.9e-07, "epoch": 0.46065259117082535, "percentage": 45.92, "elapsed_time": "0:24:09", "remaining_time": "0:28:26", "rewards/chosen": 0.004109331048451937, "logps/chosen": -380.74829727564105, "logits/chosen": -58435885.94871795, "rewards/rejected": -6.013799731324359e-05, "logps/rejected": -187.2274080602134, "logits/rejected": -50597822.43902439, "rewards/margins": 0.00416946904576518, "regularization/kl": 0.003257766366004944, "regularization/advantage_var": 0.06515531986951828, "kl": 0.04686165973544121, "stability/response_length_mean": 33.625, "stability/response_length_std": 29.442134857177734, "stability/response_length_var": 866.8392944335938, "stability/token_entropy_mean": 3.7568359375, "stability/repetition_rate_mean": 0.3110203742980957}
+{"current_steps": 95, "total_steps": 196, "loss": 0.5029, "lr": 9.399999999999999e-07, "epoch": 0.48624440179142675, "percentage": 48.47, "elapsed_time": "0:25:29", "remaining_time": "0:27:05", "rewards/chosen": 0.002092463117611559, "logps/chosen": -314.97245846518985, "logits/chosen": -50422958.98734177, "rewards/rejected": -0.00022012563307344178, "logps/rejected": -207.1459297839506, "logits/rejected": -49016667.654320985, "rewards/margins": 0.0023125887506850007, "regularization/kl": 0.002854438265785575, "regularization/advantage_var": 0.057088762521743774, "kl": 0.043993134051561356, "stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.720703125, "stability/repetition_rate_mean": 0.19374233484268188}
+{"current_steps": 100, "total_steps": 196, "loss": 0.5027, "lr": 9.9e-07, "epoch": 0.5118362124120281, "percentage": 51.02, "elapsed_time": "0:26:49", "remaining_time": "0:25:45", "rewards/chosen": 0.007421926872150318, "logps/chosen": -359.7959512246622, "logits/chosen": -54757957.18918919, "rewards/rejected": 0.0012747312701025674, "logps/rejected": -216.56615643168604, "logits/rejected": -50958068.093023255, "rewards/margins": 0.0061471956020477505, "regularization/kl": 0.003325745230540633, "regularization/advantage_var": 0.06651490181684494, "kl": 0.04170067235827446, "stability/response_length_mean": 37.5, "stability/response_length_std": 30.137779235839844, "stability/response_length_var": 908.2857055664062, "stability/token_entropy_mean": 4.244140625, "stability/repetition_rate_mean": 0.22173713147640228}
+{"current_steps": 105, "total_steps": 196, "loss": 0.503, "lr": 9.999815554351224e-07, "epoch": 0.5374280230326296, "percentage": 53.57, "elapsed_time": "0:28:10", "remaining_time": "0:24:25", "rewards/chosen": 0.008813725896628506, "logps/chosen": -399.81428840361446, "logits/chosen": -59432700.915662654, "rewards/rejected": 0.003091595196104669, "logps/rejected": -209.9729099025974, "logits/rejected": -53874594.90909091, "rewards/margins": 0.005722130700523837, "regularization/kl": 0.004456373397260904, "regularization/advantage_var": 0.08912745863199234, "kl": 0.0368778295814991, "stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.35546875, "stability/repetition_rate_mean": 0.1916852593421936}
+{"current_steps": 110, "total_steps": 196, "loss": 0.5029, "lr": 9.999066267225444e-07, "epoch": 0.5630198336532309, "percentage": 56.12, "elapsed_time": "0:29:31", "remaining_time": "0:23:05", "rewards/chosen": 0.002688568040549037, "logps/chosen": -378.5363798945783, "logits/chosen": -57453074.5060241, "rewards/rejected": 0.0029062508762656867, "logps/rejected": -206.05907568993507, "logits/rejected": -53690860.051948056, "rewards/margins": -0.00021768283571664982, "regularization/kl": 0.0032296336721628904, "regularization/advantage_var": 0.06459267437458038, "kl": 0.04355233162641525, "stability/response_length_mean": 31.125, "stability/response_length_std": 28.412460327148438, "stability/response_length_var": 807.2678833007812, "stability/token_entropy_mean": 4.0439453125, "stability/repetition_rate_mean": 0.3362334370613098}
+{"current_steps": 115, "total_steps": 196, "loss": 0.5031, "lr": 9.997740697079592e-07, "epoch": 0.5886116442738324, "percentage": 58.67, "elapsed_time": "0:30:53", "remaining_time": "0:21:45", "rewards/chosen": 0.009177133252349081, "logps/chosen": -364.50076641613924, "logits/chosen": -60690730.12658228, "rewards/rejected": 0.0028408865133921304, "logps/rejected": -201.16813753858025, "logits/rejected": -53506616.88888889, "rewards/margins": 0.006336246738956951, "regularization/kl": 0.0033262737561017275, "regularization/advantage_var": 0.06652548164129257, "kl": 0.04345473647117615, "stability/response_length_mean": 37.0, "stability/response_length_std": 30.06184196472168, "stability/response_length_var": 903.7142944335938, "stability/token_entropy_mean": 4.916015625, "stability/repetition_rate_mean": 0.19457165896892548}
+{"current_steps": 120, "total_steps": 196, "loss": 0.503, "lr": 9.995838996722914e-07, "epoch": 0.6142034548944337, "percentage": 61.22, "elapsed_time": "0:32:12", "remaining_time": "0:20:24", "rewards/chosen": 0.006303458431966944, "logps/chosen": -384.1818831699346, "logits/chosen": -57835580.23529412, "rewards/rejected": 0.0006464999057575614, "logps/rejected": -212.20988491766468, "logits/rejected": -54506011.59281437, "rewards/margins": 0.005656958526209382, "regularization/kl": 0.003679528133943677, "regularization/advantage_var": 0.07359055429697037, "kl": 0.029955614358186722, "stability/response_length_mean": 29.25, "stability/response_length_std": 29.687660217285156, "stability/response_length_var": 881.3571166992188, "stability/token_entropy_mean": 4.56640625, "stability/repetition_rate_mean": 0.18938209116458893}
+{"current_steps": 125, "total_steps": 196, "loss": 0.5031, "lr": 9.993361385379874e-07, "epoch": 0.6397952655150352, "percentage": 63.78, "elapsed_time": "0:33:33", "remaining_time": "0:19:03", "rewards/chosen": 0.0040864528041996365, "logps/chosen": -417.33845248287673, "logits/chosen": -60829527.67123288, "rewards/rejected": -0.00047545169276752694, "logps/rejected": -216.37194683908046, "logits/rejected": -55803668.59770115, "rewards/margins": 0.004561904496967164, "regularization/kl": 0.003166340058669448, "regularization/advantage_var": 0.06332679837942123, "kl": 0.03671300411224365, "stability/response_length_mean": 15.25, "stability/response_length_std": 24.93276596069336, "stability/response_length_var": 621.6428833007812, "stability/token_entropy_mean": 3.6328125, "stability/repetition_rate_mean": 0.16215278208255768}
+{"current_steps": 130, "total_steps": 196, "loss": 0.503, "lr": 9.99030814866488e-07, "epoch": 0.6653870761356366, "percentage": 66.33, "elapsed_time": "0:34:53", "remaining_time": "0:17:42", "rewards/chosen": 0.00643845142856721, "logps/chosen": -392.59354838709675, "logits/chosen": -57505567.38064516, "rewards/rejected": -0.002984953649116285, "logps/rejected": -183.94876893939394, "logits/rejected": -53976231.56363636, "rewards/margins": 0.009423405077683494, "regularization/kl": 0.0033106517512351274, "regularization/advantage_var": 0.06621303409337997, "kl": 0.03913918882608414, "stability/response_length_mean": 38.0, "stability/response_length_std": 29.13515281677246, "stability/response_length_var": 848.8571166992188, "stability/token_entropy_mean": 3.91796875, "stability/repetition_rate_mean": 0.19009943306446075}
+{"current_steps": 135, "total_steps": 196, "loss": 0.5032, "lr": 9.986679638549358e-07, "epoch": 0.690978886756238, "percentage": 68.88, "elapsed_time": "0:36:13", "remaining_time": "0:16:22", "rewards/chosen": 0.005291132473506811, "logps/chosen": -401.7632524923313, "logits/chosen": -59995487.80368098, "rewards/rejected": 0.0003992993455783577, "logps/rejected": -247.86673964968153, "logits/rejected": -55185649.32484076, "rewards/margins": 0.0048918331279284525, "regularization/kl": 0.004669652320444584, "regularization/advantage_var": 0.09339304268360138, "kl": 0.04277459532022476, "stability/response_length_mean": 29.0, "stability/response_length_std": 30.20406723022461, "stability/response_length_var": 912.2857055664062, "stability/token_entropy_mean": 3.416015625, "stability/repetition_rate_mean": 0.2191840261220932}
+{"current_steps": 140, "total_steps": 196, "loss": 0.5028, "lr": 9.982476273321175e-07, "epoch": 0.7165706973768394, "percentage": 71.43, "elapsed_time": "0:37:34", "remaining_time": "0:15:01", "rewards/chosen": 0.005194586696046772, "logps/chosen": -367.13778409090907, "logits/chosen": -57242692.266666666, "rewards/rejected": 8.15337223391379e-05, "logps/rejected": -190.0336693548387, "logits/rejected": -50260034.06451613, "rewards/margins": 0.0051130529737076335, "regularization/kl": 0.0037213729228824377, "regularization/advantage_var": 0.07442746311426163, "kl": 0.03989999368786812, "stability/response_length_mean": 52.75, "stability/response_length_std": 22.72663688659668, "stability/response_length_var": 516.5, "stability/token_entropy_mean": 4.783203125, "stability/repetition_rate_mean": 0.3268229067325592}
+{"current_steps": 145, "total_steps": 196, "loss": 0.5032, "lr": 9.977698537536417e-07, "epoch": 0.7421625079974408, "percentage": 73.98, "elapsed_time": "0:38:54", "remaining_time": "0:13:41", "rewards/chosen": 0.004852092768772539, "logps/chosen": -330.2233151355422, "logits/chosen": -51305524.43373494, "rewards/rejected": -0.0011343126753707985, "logps/rejected": -199.49925172483765, "logits/rejected": -51872003.32467532, "rewards/margins": 0.005986405444143338, "regularization/kl": 0.003280136501416564, "regularization/advantage_var": 0.06560272723436356, "kl": 0.047912172973155975, "stability/response_length_mean": 28.75, "stability/response_length_std": 29.86517333984375, "stability/response_length_var": 891.9285888671875, "stability/token_entropy_mean": 5.314453125, "stability/repetition_rate_mean": 0.24838952720165253}
+{"current_steps": 150, "total_steps": 196, "loss": 0.5029, "lr": 9.972346981963546e-07, "epoch": 0.7677543186180422, "percentage": 76.53, "elapsed_time": "0:40:15", "remaining_time": "0:12:20", "rewards/chosen": 0.009192322282230153, "logps/chosen": -397.8425551470588, "logits/chosen": -55574064.1882353, "rewards/rejected": 0.002263024648030599, "logps/rejected": -207.70329427083334, "logits/rejected": -50405259.946666665, "rewards/margins": 0.006929297634199555, "regularization/kl": 0.003956340253353119, "regularization/advantage_var": 0.07912679761648178, "kl": 0.04824293777346611, "stability/response_length_mean": 24.125, "stability/response_length_std": 25.637235641479492, "stability/response_length_var": 657.2678833007812, "stability/token_entropy_mean": 4.59765625, "stability/repetition_rate_mean": 0.20100826025009155}
+{"current_steps": 155, "total_steps": 196, "loss": 0.503, "lr": 9.966422223519885e-07, "epoch": 0.7933461292386437, "percentage": 79.08, "elapsed_time": "0:41:37", "remaining_time": "0:11:00", "rewards/chosen": 0.004395649012397317, "logps/chosen": -360.94535845588234, "logits/chosen": -54217438.870588236, "rewards/rejected": 0.002860300540924072, "logps/rejected": -191.27700520833332, "logits/rejected": -49562466.986666664, "rewards/margins": 0.0015353484714732452, "regularization/kl": 0.004205561708658934, "regularization/advantage_var": 0.08411123603582382, "kl": 0.04922879859805107, "stability/response_length_mean": 61.375, "stability/response_length_std": 7.424621105194092, "stability/response_length_var": 55.125, "stability/token_entropy_mean": 4.6044921875, "stability/repetition_rate_mean": 0.3171784281730652}
+{"current_steps": 160, "total_steps": 196, "loss": 0.5031, "lr": 9.959924945200523e-07, "epoch": 0.818937939859245, "percentage": 81.63, "elapsed_time": "0:42:57", "remaining_time": "0:09:40", "rewards/chosen": 0.008917692161741712, "logps/chosen": -416.8061755952381, "logits/chosen": -55477516.190476194, "rewards/rejected": 0.00039390641215600466, "logps/rejected": -186.89340049342104, "logits/rejected": -49231777.684210524, "rewards/margins": 0.008523785749585707, "regularization/kl": 0.003785709384828806, "regularization/advantage_var": 0.07571418583393097, "kl": 0.04486330226063728, "stability/response_length_mean": 26.25, "stability/response_length_std": 31.535694122314453, "stability/response_length_var": 994.5, "stability/token_entropy_mean": 3.771484375, "stability/repetition_rate_mean": 0.291015625}
+{"current_steps": 165, "total_steps": 196, "loss": 0.5029, "lr": 9.952855895999568e-07, "epoch": 0.8445297504798465, "percentage": 84.18, "elapsed_time": "0:44:18", "remaining_time": "0:08:19", "rewards/chosen": 0.00609781139198391, "logps/chosen": -370.85578304597703, "logits/chosen": -54658201.01149426, "rewards/rejected": 0.001535689892017678, "logps/rejected": -212.5767604880137, "logits/rejected": -50095780.82191781, "rewards/margins": 0.004562121499966231, "regularization/kl": 0.004139466676861048, "regularization/advantage_var": 0.0827893316745758, "kl": 0.04533420130610466, "stability/response_length_mean": 12.625, "stability/response_length_std": 21.460512161254883, "stability/response_length_var": 460.5535583496094, "stability/token_entropy_mean": 4.81640625, "stability/repetition_rate_mean": 0.20468750596046448}
+{"current_steps": 170, "total_steps": 196, "loss": 0.5029, "lr": 9.9452158908238e-07, "epoch": 0.8701215611004478, "percentage": 86.73, "elapsed_time": "0:45:39", "remaining_time": "0:06:59", "rewards/chosen": 0.008814329931721885, "logps/chosen": -441.2536982248521, "logits/chosen": -59383468.686390534, "rewards/rejected": 0.0001755822573276545, "logps/rejected": -191.25607926324503, "logits/rejected": -52950897.58940397, "rewards/margins": 0.00863874767439423, "regularization/kl": 0.004800532013177872, "regularization/advantage_var": 0.09601064026355743, "kl": 0.0640627071261406, "stability/response_length_mean": 28.25, "stability/response_length_std": 30.844078063964844, "stability/response_length_var": 951.3571166992188, "stability/token_entropy_mean": 3.33251953125, "stability/repetition_rate_mean": 0.21061541140079498}
+{"current_steps": 175, "total_steps": 196, "loss": 0.5026, "lr": 9.937005810398745e-07, "epoch": 0.8957133717210493, "percentage": 89.29, "elapsed_time": "0:47:00", "remaining_time": "0:05:38", "rewards/chosen": 0.00937136253678655, "logps/chosen": -356.34346762048193, "logits/chosen": -56930439.71084338, "rewards/rejected": 0.003396167383565531, "logps/rejected": -194.7735643262987, "logits/rejected": -48423304.31168831, "rewards/margins": 0.005975195153221019, "regularization/kl": 0.0036111907102167606, "regularization/advantage_var": 0.07222381234169006, "kl": 0.058815985918045044, "stability/response_length_mean": 32.25, "stability/response_length_std": 27.932315826416016, "stability/response_length_var": 780.2142944335938, "stability/token_entropy_mean": 4.72265625, "stability/repetition_rate_mean": 0.2878348231315613}
+{"current_steps": 180, "total_steps": 196, "loss": 0.5028, "lr": 9.928226601167138e-07, "epoch": 0.9213051823416507, "percentage": 91.84, "elapsed_time": "0:48:20", "remaining_time": "0:04:17", "rewards/chosen": 0.012041092650290647, "logps/chosen": -376.0254265720859, "logits/chosen": -52935007.80368098, "rewards/rejected": 0.00411484347786873, "logps/rejected": -211.25706608280254, "logits/rejected": -48337998.26751592, "rewards/margins": 0.007926249172421917, "regularization/kl": 0.003911244682967663, "regularization/advantage_var": 0.07822488993406296, "kl": 0.06844082474708557, "stability/response_length_mean": 40.375, "stability/response_length_std": 32.61874008178711, "stability/response_length_var": 1063.982177734375, "stability/token_entropy_mean": 3.357421875, "stability/repetition_rate_mean": 0.162109375}
+{"current_steps": 185, "total_steps": 196, "loss": 0.5028, "lr": 9.918879275179817e-07, "epoch": 0.946896992962252, "percentage": 94.39, "elapsed_time": "0:49:40", "remaining_time": "0:02:57", "rewards/chosen": 0.012579193821659795, "logps/chosen": -311.8755787037037, "logits/chosen": -54936120.88888889, "rewards/rejected": 0.0066387276106242895, "logps/rejected": -196.9727675039557, "logits/rejected": -49878696.50632911, "rewards/margins": 0.005940466211035505, "regularization/kl": 0.004052338190376759, "regularization/advantage_var": 0.08104676008224487, "kl": 0.0741070955991745, "stability/response_length_mean": 24.125, "stability/response_length_std": 29.019390106201172, "stability/response_length_var": 842.125, "stability/token_entropy_mean": 3.87109375, "stability/repetition_rate_mean": 0.18797221779823303}
+{"current_steps": 190, "total_steps": 196, "loss": 0.5028, "lr": 9.90896490997906e-07, "epoch": 0.9724888035828535, "percentage": 96.94, "elapsed_time": "0:51:00", "remaining_time": "0:01:36", "rewards/chosen": 0.011261776535810824, "logps/chosen": -392.29271145209583, "logits/chosen": -53416365.22155689, "rewards/rejected": 0.0016973669232885822, "logps/rejected": -211.18162785947712, "logits/rejected": -50075306.666666664, "rewards/margins": 0.009564409612522242, "regularization/kl": 0.0037938205059617758, "regularization/advantage_var": 0.07587640732526779, "kl": 0.06636197119951248, "stability/response_length_mean": 17.625, "stability/response_length_std": 28.680194854736328, "stability/response_length_var": 822.5535888671875, "stability/token_entropy_mean": 4.8212890625, "stability/repetition_rate_mean": 0.1822916716337204}
+{"current_steps": 195, "total_steps": 196, "loss": 0.5029, "lr": 9.898484648474362e-07, "epoch": 0.9980806142034548, "percentage": 99.49, "elapsed_time": "0:52:21", "remaining_time": "0:00:16", "rewards/chosen": 0.013610498366817351, "logps/chosen": -392.64884072580645, "logits/chosen": -60522562.06451613, "rewards/rejected": 0.0010842363942753185, "logps/rejected": -197.9084872159091, "logits/rejected": -54295080.339393936, "rewards/margins": 0.012526261972542033, "regularization/kl": 0.0035452134907245636, "regularization/advantage_var": 0.07090426236391068, "kl": 0.05749697610735893, "stability/response_length_mean": 48.125, "stability/response_length_std": 29.39600372314453, "stability/response_length_var": 864.125, "stability/token_entropy_mean": 4.4609375, "stability/repetition_rate_mean": 0.4140625}
+{"current_steps": 196, "total_steps": 196, "epoch": 1.0, "percentage": 100.0, "elapsed_time": "0:52:34", "remaining_time": "0:00:00"}

mistral7b_kto_kl-beta0.1/trainer_state.json

@@ -0,0 +1,1013 @@
+{
+  "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": 1.1286565065383911,
+      "kl": 0.009834487922489643,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59720272.84210526,
+      "logits/rejected": -50419004.952380955,
+      "logps/chosen": -439.009765625,
+      "logps/rejected": -198.3979957217262,
+      "loss": 0.502,
+      "loss/base_kto": 0.4997194707393646,
+      "loss/total_with_kl": 0.5024515390396118,
+      "metrics/advantage_var": 0.0546402707695961,
+      "regularization/advantage_var": 0.0546402707695961,
+      "regularization/kl": 0.0027320135850459337,
+      "rewards/chosen": 0.0035851931101397463,
+      "rewards/margins": 0.002444372626773098,
+      "rewards/rejected": 0.0011408204833666484,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 1.0271365642547607,
+      "kl": 0.02555137872695923,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59294946.077922076,
+      "logits/rejected": -50014485.590361446,
+      "logps/chosen": -429.9174107142857,
+      "logps/rejected": -192.83664344879517,
+      "loss": 0.5032,
+      "loss/base_kto": 0.5001684427261353,
+      "loss/total_with_kl": 0.5038444995880127,
+      "metrics/advantage_var": 0.07352278381586075,
+      "regularization/advantage_var": 0.07352278381586075,
+      "regularization/kl": 0.0036761388182640076,
+      "rewards/chosen": 0.0021098085812159945,
+      "rewards/margins": -0.0010197663266154045,
+      "rewards/rejected": 0.003129574907831399,
+      "stability/repetition_rate_mean": 0.3766128420829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.369140625,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 1.1246036291122437,
+      "kl": 0.01402486115694046,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56229081.21212121,
+      "logits/rejected": -51912063.174193546,
+      "logps/chosen": -404.1037878787879,
+      "logps/rejected": -200.66993447580646,
+      "loss": 0.5033,
+      "loss/base_kto": 0.4999401271343231,
+      "loss/total_with_kl": 0.5033776164054871,
+      "metrics/advantage_var": 0.0687500610947609,
+      "regularization/advantage_var": 0.0687500610947609,
+      "regularization/kl": 0.0034375027753412724,
+      "rewards/chosen": 0.0027040712761156487,
+      "rewards/margins": 0.0002901616101157749,
+      "rewards/rejected": 0.002413909665999874,
+      "stability/repetition_rate_mean": 0.275390625,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.92341423034668,
+      "stability/response_length_var": 895.4107055664062,
+      "stability/token_entropy_mean": 3.6796875,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 1.0175650119781494,
+      "kl": 0.012321279384195805,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51667027.2,
+      "logits/rejected": -48065526.4,
+      "logps/chosen": -343.695458984375,
+      "logps/rejected": -225.64677734375,
+      "loss": 0.5033,
+      "loss/base_kto": 0.5002518892288208,
+      "loss/total_with_kl": 0.5026945471763611,
+      "metrics/advantage_var": 0.04885442927479744,
+      "regularization/advantage_var": 0.04885442927479744,
+      "regularization/kl": 0.0024427215103060007,
+      "rewards/chosen": 0.0001240897923707962,
+      "rewards/margins": -0.0015346232801675798,
+      "rewards/rejected": 0.001658713072538376,
+      "stability/repetition_rate_mean": 0.2021329402923584,
+      "stability/response_length_mean": 42.125,
+      "stability/response_length_std": 27.01553726196289,
+      "stability/response_length_var": 729.8392944335938,
+      "stability/token_entropy_mean": 4.43359375,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 1.0005894899368286,
+      "kl": 0.01893586292862892,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56474902.951724134,
+      "logits/rejected": -52020124.525714286,
+      "logps/chosen": -378.4993534482759,
+      "logps/rejected": -202.78513392857144,
+      "loss": 0.5032,
+      "loss/base_kto": 0.5001356601715088,
+      "loss/total_with_kl": 0.5026346445083618,
+      "metrics/advantage_var": 0.049978986382484436,
+      "regularization/advantage_var": 0.049978986382484436,
+      "regularization/kl": 0.0024989494122564793,
+      "rewards/chosen": -0.000791168212890625,
+      "rewards/margins": -0.0015635289464678084,
+      "rewards/rejected": 0.0007723607335771833,
+      "stability/repetition_rate_mean": 0.16825254261493683,
+      "stability/response_length_mean": 43.375,
+      "stability/response_length_std": 28.54538917541504,
+      "stability/response_length_var": 814.8392944335938,
+      "stability/token_entropy_mean": 4.763671875,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 1.047668695449829,
+      "kl": 0.019050994887948036,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57737765.798657715,
+      "logits/rejected": -51525733.80116959,
+      "logps/chosen": -468.53869546979865,
+      "logps/rejected": -198.22558022660817,
+      "loss": 0.5033,
+      "loss/base_kto": 0.4998214840888977,
+      "loss/total_with_kl": 0.5030900239944458,
+      "metrics/advantage_var": 0.0653710812330246,
+      "regularization/advantage_var": 0.0653710812330246,
+      "regularization/kl": 0.0032685536425560713,
+      "rewards/chosen": 0.004255342803545446,
+      "rewards/margins": 0.002025002526303064,
+      "rewards/rejected": 0.0022303402772423818,
+      "stability/repetition_rate_mean": 0.1662016361951828,
+      "stability/response_length_mean": 43.0,
+      "stability/response_length_std": 29.602123260498047,
+      "stability/response_length_var": 876.2857055664062,
+      "stability/token_entropy_mean": 4.587890625,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 2.3248555660247803,
+      "kl": 0.021593520417809486,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57636539.94936709,
+      "logits/rejected": -53914326.913580246,
+      "logps/chosen": -385.92696795886076,
+      "logps/rejected": -195.1241801697531,
+      "loss": 0.5036,
+      "loss/base_kto": 0.4999656677246094,
+      "loss/total_with_kl": 0.5032640695571899,
+      "metrics/advantage_var": 0.06596841663122177,
+      "regularization/advantage_var": 0.06596841663122177,
+      "regularization/kl": 0.0032984204590320587,
+      "rewards/chosen": 0.0009543428881258904,
+      "rewards/margins": 0.0007249228627183583,
+      "rewards/rejected": 0.00022942002540753213,
+      "stability/repetition_rate_mean": 0.3244977593421936,
+      "stability/response_length_mean": 56.875,
+      "stability/response_length_std": 20.152544021606445,
+      "stability/response_length_var": 406.125,
+      "stability/token_entropy_mean": 5.46484375,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.9593210816383362,
+      "kl": 0.01618729531764984,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56643635.2,
+      "logits/rejected": -50236915.2,
+      "logps/chosen": -421.9146484375,
+      "logps/rejected": -219.2841796875,
+      "loss": 0.5033,
+      "loss/base_kto": 0.5004779100418091,
+      "loss/total_with_kl": 0.5045997500419617,
+      "metrics/advantage_var": 0.08243563026189804,
+      "regularization/advantage_var": 0.08243563026189804,
+      "regularization/kl": 0.004121781792491674,
+      "rewards/chosen": -0.00120118148624897,
+      "rewards/margins": -0.003708549775183201,
+      "rewards/rejected": 0.002507368288934231,
+      "stability/repetition_rate_mean": 0.13671875,
+      "stability/response_length_mean": 33.125,
+      "stability/response_length_std": 33.01704025268555,
+      "stability/response_length_var": 1090.125,
+      "stability/token_entropy_mean": 4.8876953125,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 1.1373188495635986,
+      "kl": 0.0272687915712595,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52697338.84563758,
+      "logits/rejected": -50778291.649122804,
+      "logps/chosen": -352.29383389261744,
+      "logps/rejected": -243.8361659356725,
+      "loss": 0.5033,
+      "loss/base_kto": 0.5000495910644531,
+      "loss/total_with_kl": 0.5032973289489746,
+      "metrics/advantage_var": 0.06495420634746552,
+      "regularization/advantage_var": 0.06495420634746552,
+      "regularization/kl": 0.0032477104105055332,
+      "rewards/chosen": 0.003462229799104217,
+      "rewards/margins": -0.0003369061228871424,
+      "rewards/rejected": 0.0037991359219913595,
+      "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": 4.759765625,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.9557572603225708,
+      "kl": 0.027951324358582497,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54038854.44295302,
+      "logits/rejected": -48308792.88888889,
+      "logps/chosen": -375.13472630033556,
+      "logps/rejected": -180.2223935489766,
+      "loss": 0.5029,
+      "loss/base_kto": 0.49937987327575684,
+      "loss/total_with_kl": 0.5020853877067566,
+      "metrics/advantage_var": 0.05411091074347496,
+      "regularization/advantage_var": 0.05411091074347496,
+      "regularization/kl": 0.002705545863136649,
+      "rewards/chosen": 0.006440740703736376,
+      "rewards/margins": 0.004618997412015367,
+      "rewards/rejected": 0.0018217432917210093,
+      "stability/repetition_rate_mean": 0.4069356620311737,
+      "stability/response_length_mean": 55.375,
+      "stability/response_length_std": 16.361431121826172,
+      "stability/response_length_var": 267.6964416503906,
+      "stability/token_entropy_mean": 3.8125,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 1.5015521049499512,
+      "kl": 0.04579087719321251,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56948112.69565217,
+      "logits/rejected": -49574068.327044025,
+      "logps/chosen": -344.4958753881988,
+      "logps/rejected": -230.41398879716982,
+      "loss": 0.5034,
+      "loss/base_kto": 0.5002328753471375,
+      "loss/total_with_kl": 0.5030208826065063,
+      "metrics/advantage_var": 0.05576041340827942,
+      "regularization/advantage_var": 0.05576041340827942,
+      "regularization/kl": 0.0027880207635462284,
+      "rewards/chosen": 0.001614014924683186,
+      "rewards/margins": -0.0017921153681972743,
+      "rewards/rejected": 0.00340613029288046,
+      "stability/repetition_rate_mean": 0.103515625,
+      "stability/response_length_mean": 24.5,
+      "stability/response_length_std": 32.710853576660156,
+      "stability/response_length_var": 1070.0,
+      "stability/token_entropy_mean": 4.61328125,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 1.0588984489440918,
+      "kl": 0.03464470058679581,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52673585.64848485,
+      "logits/rejected": -47836140.18064516,
+      "logps/chosen": -359.6580492424242,
+      "logps/rejected": -178.18500504032258,
+      "loss": 0.5031,
+      "loss/base_kto": 0.49969473481178284,
+      "loss/total_with_kl": 0.5024840831756592,
+      "metrics/advantage_var": 0.0557870976626873,
+      "regularization/advantage_var": 0.0557870976626873,
+      "regularization/kl": 0.0027893546503037214,
+      "rewards/chosen": 0.006273941560225053,
+      "rewards/margins": 0.0020832758844772734,
+      "rewards/rejected": 0.0041906656757477794,
+      "stability/repetition_rate_mean": 0.1456473171710968,
+      "stability/response_length_mean": 39.25,
+      "stability/response_length_std": 32.07023620605469,
+      "stability/response_length_var": 1028.5,
+      "stability/token_entropy_mean": 3.931640625,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 1.038607120513916,
+      "kl": 0.03578026220202446,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57568037.98709677,
+      "logits/rejected": -47274167.07878788,
+      "logps/chosen": -381.87852822580646,
+      "logps/rejected": -183.5539891098485,
+      "loss": 0.503,
+      "loss/base_kto": 0.4994705319404602,
+      "loss/total_with_kl": 0.5023882985115051,
+      "metrics/advantage_var": 0.058355946093797684,
+      "regularization/advantage_var": 0.058355946093797684,
+      "regularization/kl": 0.002917797304689884,
+      "rewards/chosen": 0.00682763361161755,
+      "rewards/margins": 0.003990938027233671,
+      "rewards/rejected": 0.002836695584383878,
+      "stability/repetition_rate_mean": 0.1857219785451889,
+      "stability/response_length_mean": 20.625,
+      "stability/response_length_std": 27.176342010498047,
+      "stability/response_length_var": 738.5535888671875,
+      "stability/token_entropy_mean": 5.537109375,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.974204957485199,
+      "kl": 0.04303906485438347,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52785099.1483871,
+      "logits/rejected": -49040083.00606061,
+      "logps/chosen": -379.097404233871,
+      "logps/rejected": -220.38963068181818,
+      "loss": 0.5033,
+      "loss/base_kto": 0.500048816204071,
+      "loss/total_with_kl": 0.5032418966293335,
+      "metrics/advantage_var": 0.06386099010705948,
+      "regularization/advantage_var": 0.06386099010705948,
+      "regularization/kl": 0.0031930492259562016,
+      "rewards/chosen": 0.0031055967653951335,
+      "rewards/margins": -0.00014737946430142933,
+      "rewards/rejected": 0.003252976229696563,
+      "stability/repetition_rate_mean": 0.177734375,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.7109375,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.9141969084739685,
+      "kl": 0.03373977914452553,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49510382.7027027,
+      "logits/rejected": -47913043.34883721,
+      "logps/chosen": -314.18810705236484,
+      "logps/rejected": -228.88971656976744,
+      "loss": 0.5033,
+      "loss/base_kto": 0.4988802373409271,
+      "loss/total_with_kl": 0.5023767948150635,
+      "metrics/advantage_var": 0.06993144005537033,
+      "regularization/advantage_var": 0.06993144005537033,
+      "regularization/kl": 0.0034965721424669027,
+      "rewards/chosen": 0.008269893961983759,
+      "rewards/margins": 0.008756556226810444,
+      "rewards/rejected": -0.0004866622648266859,
+      "stability/repetition_rate_mean": 0.08308292925357819,
+      "stability/response_length_mean": 22.75,
+      "stability/response_length_std": 31.07479476928711,
+      "stability/response_length_var": 965.6428833007812,
+      "stability/token_entropy_mean": 2.158203125,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.998468279838562,
+      "kl": 0.04045908525586128,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58651403.867549665,
+      "logits/rejected": -53481308.40236686,
+      "logps/chosen": -403.97845095198676,
+      "logps/rejected": -201.0639330621302,
+      "loss": 0.503,
+      "loss/base_kto": 0.4992879033088684,
+      "loss/total_with_kl": 0.5029799342155457,
+      "metrics/advantage_var": 0.07383989542722702,
+      "regularization/advantage_var": 0.07383989542722702,
+      "regularization/kl": 0.0036919948179274797,
+      "rewards/chosen": 0.005166642713230966,
+      "rewards/margins": 0.005980765803433253,
+      "rewards/rejected": -0.0008141230902022864,
+      "stability/repetition_rate_mean": 0.18307292461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.25390625,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.8892827033996582,
+      "kl": 0.03160804137587547,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51504227.09677419,
+      "logits/rejected": -47165461.72121212,
+      "logps/chosen": -358.3259576612903,
+      "logps/rejected": -230.4444128787879,
+      "loss": 0.5029,
+      "loss/base_kto": 0.49965283274650574,
+      "loss/total_with_kl": 0.5036584734916687,
+      "metrics/advantage_var": 0.08011424541473389,
+      "regularization/advantage_var": 0.08011424541473389,
+      "regularization/kl": 0.004005712457001209,
+      "rewards/chosen": 0.0006552665464339718,
+      "rewards/margins": 0.0026155838402368687,
+      "rewards/rejected": -0.001960317293802897,
+      "stability/repetition_rate_mean": 0.1637834906578064,
+      "stability/response_length_mean": 25.875,
+      "stability/response_length_std": 28.792049407958984,
+      "stability/response_length_var": 828.9821166992188,
+      "stability/token_entropy_mean": 4.9140625,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 1.1928011178970337,
+      "kl": 0.04686165973544121,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58435885.94871795,
+      "logits/rejected": -50597822.43902439,
+      "logps/chosen": -380.74829727564105,
+      "logps/rejected": -187.2274080602134,
+      "loss": 0.5028,
+      "loss/base_kto": 0.49942442774772644,
+      "loss/total_with_kl": 0.5026821494102478,
+      "metrics/advantage_var": 0.06515531986951828,
+      "regularization/advantage_var": 0.06515531986951828,
+      "regularization/kl": 0.003257766366004944,
+      "rewards/chosen": 0.004109331048451937,
+      "rewards/margins": 0.00416946904576518,
+      "rewards/rejected": -6.013799731324359e-05,
+      "stability/repetition_rate_mean": 0.3110203742980957,
+      "stability/response_length_mean": 33.625,
+      "stability/response_length_std": 29.442134857177734,
+      "stability/response_length_var": 866.8392944335938,
+      "stability/token_entropy_mean": 3.7568359375,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.8256028294563293,
+      "kl": 0.043993134051561356,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50422958.98734177,
+      "logits/rejected": -49016667.654320985,
+      "logps/chosen": -314.97245846518985,
+      "logps/rejected": -207.1459297839506,
+      "loss": 0.5029,
+      "loss/base_kto": 0.4996494948863983,
+      "loss/total_with_kl": 0.5025039315223694,
+      "metrics/advantage_var": 0.057088762521743774,
+      "regularization/advantage_var": 0.057088762521743774,
+      "regularization/kl": 0.002854438265785575,
+      "rewards/chosen": 0.002092463117611559,
+      "rewards/margins": 0.0023125887506850007,
+      "rewards/rejected": -0.00022012563307344178,
+      "stability/repetition_rate_mean": 0.19374233484268188,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 31.1264705657959,
+      "stability/response_length_var": 968.8571166992188,
+      "stability/token_entropy_mean": 3.720703125,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.9563474655151367,
+      "kl": 0.04170067235827446,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54757957.18918919,
+      "logits/rejected": -50958068.093023255,
+      "logps/chosen": -359.7959512246622,
+      "logps/rejected": -216.56615643168604,
+      "loss": 0.5027,
+      "loss/base_kto": 0.4991760849952698,
+      "loss/total_with_kl": 0.5025018453598022,
+      "metrics/advantage_var": 0.06651490181684494,
+      "regularization/advantage_var": 0.06651490181684494,
+      "regularization/kl": 0.003325745230540633,
+      "rewards/chosen": 0.007421926872150318,
+      "rewards/margins": 0.0061471956020477505,
+      "rewards/rejected": 0.0012747312701025674,
+      "stability/repetition_rate_mean": 0.22173713147640228,
+      "stability/response_length_mean": 37.5,
+      "stability/response_length_std": 30.137779235839844,
+      "stability/response_length_var": 908.2857055664062,
+      "stability/token_entropy_mean": 4.244140625,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 1.0580958127975464,
+      "kl": 0.0368778295814991,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59432700.915662654,
+      "logits/rejected": -53874594.90909091,
+      "logps/chosen": -399.81428840361446,
+      "logps/rejected": -209.9729099025974,
+      "loss": 0.503,
+      "loss/base_kto": 0.49930858612060547,
+      "loss/total_with_kl": 0.5037649273872375,
+      "metrics/advantage_var": 0.08912745863199234,
+      "regularization/advantage_var": 0.08912745863199234,
+      "regularization/kl": 0.004456373397260904,
+      "rewards/chosen": 0.008813725896628506,
+      "rewards/margins": 0.005722130700523837,
+      "rewards/rejected": 0.003091595196104669,
+      "stability/repetition_rate_mean": 0.1916852593421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.35546875,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.9517597556114197,
+      "kl": 0.04355233162641525,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57453074.5060241,
+      "logits/rejected": -53690860.051948056,
+      "logps/chosen": -378.5363798945783,
+      "logps/rejected": -206.05907568993507,
+      "loss": 0.5029,
+      "loss/base_kto": 0.5001128911972046,
+      "loss/total_with_kl": 0.5033425688743591,
+      "metrics/advantage_var": 0.06459267437458038,
+      "regularization/advantage_var": 0.06459267437458038,
+      "regularization/kl": 0.0032296336721628904,
+      "rewards/chosen": 0.002688568040549037,
+      "rewards/margins": -0.00021768283571664982,
+      "rewards/rejected": 0.0029062508762656867,
+      "stability/repetition_rate_mean": 0.3362334370613098,
+      "stability/response_length_mean": 31.125,
+      "stability/response_length_std": 28.412460327148438,
+      "stability/response_length_var": 807.2678833007812,
+      "stability/token_entropy_mean": 4.0439453125,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.8893980979919434,
+      "kl": 0.04345473647117615,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60690730.12658228,
+      "logits/rejected": -53506616.88888889,
+      "logps/chosen": -364.50076641613924,
+      "logps/rejected": -201.16813753858025,
+      "loss": 0.5031,
+      "loss/base_kto": 0.4992709159851074,
+      "loss/total_with_kl": 0.5025972723960876,
+      "metrics/advantage_var": 0.06652548164129257,
+      "regularization/advantage_var": 0.06652548164129257,
+      "regularization/kl": 0.0033262737561017275,
+      "rewards/chosen": 0.009177133252349081,
+      "rewards/margins": 0.006336246738956951,
+      "rewards/rejected": 0.0028408865133921304,
+      "stability/repetition_rate_mean": 0.19457165896892548,
+      "stability/response_length_mean": 37.0,
+      "stability/response_length_std": 30.06184196472168,
+      "stability/response_length_var": 903.7142944335938,
+      "stability/token_entropy_mean": 4.916015625,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.9201934933662415,
+      "kl": 0.029955614358186722,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57835580.23529412,
+      "logits/rejected": -54506011.59281437,
+      "logps/chosen": -384.1818831699346,
+      "logps/rejected": -212.20988491766468,
+      "loss": 0.503,
+      "loss/base_kto": 0.4993014335632324,
+      "loss/total_with_kl": 0.5029810070991516,
+      "metrics/advantage_var": 0.07359055429697037,
+      "regularization/advantage_var": 0.07359055429697037,
+      "regularization/kl": 0.003679528133943677,
+      "rewards/chosen": 0.006303458431966944,
+      "rewards/margins": 0.005656958526209382,
+      "rewards/rejected": 0.0006464999057575614,
+      "stability/repetition_rate_mean": 0.18938209116458893,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.687660217285156,
+      "stability/response_length_var": 881.3571166992188,
+      "stability/token_entropy_mean": 4.56640625,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 1.0461677312850952,
+      "kl": 0.03671300411224365,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60829527.67123288,
+      "logits/rejected": -55803668.59770115,
+      "logps/chosen": -417.33845248287673,
+      "logps/rejected": -216.37194683908046,
+      "loss": 0.5031,
+      "loss/base_kto": 0.4993533194065094,
+      "loss/total_with_kl": 0.5025196671485901,
+      "metrics/advantage_var": 0.06332679837942123,
+      "regularization/advantage_var": 0.06332679837942123,
+      "regularization/kl": 0.003166340058669448,
+      "rewards/chosen": 0.0040864528041996365,
+      "rewards/margins": 0.004561904496967164,
+      "rewards/rejected": -0.00047545169276752694,
+      "stability/repetition_rate_mean": 0.16215278208255768,
+      "stability/response_length_mean": 15.25,
+      "stability/response_length_std": 24.93276596069336,
+      "stability/response_length_var": 621.6428833007812,
+      "stability/token_entropy_mean": 3.6328125,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.9537745714187622,
+      "kl": 0.03913918882608414,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57505567.38064516,
+      "logits/rejected": -53976231.56363636,
+      "logps/chosen": -392.59354838709675,
+      "logps/rejected": -183.94876893939394,
+      "loss": 0.503,
+      "loss/base_kto": 0.4988265633583069,
+      "loss/total_with_kl": 0.5021371841430664,
+      "metrics/advantage_var": 0.06621303409337997,
+      "regularization/advantage_var": 0.06621303409337997,
+      "regularization/kl": 0.0033106517512351274,
+      "rewards/chosen": 0.00643845142856721,
+      "rewards/margins": 0.009423405077683494,
+      "rewards/rejected": -0.002984953649116285,
+      "stability/repetition_rate_mean": 0.19009943306446075,
+      "stability/response_length_mean": 38.0,
+      "stability/response_length_std": 29.13515281677246,
+      "stability/response_length_var": 848.8571166992188,
+      "stability/token_entropy_mean": 3.91796875,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.9990471601486206,
+      "kl": 0.04277459532022476,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59995487.80368098,
+      "logits/rejected": -55185649.32484076,
+      "logps/chosen": -401.7632524923313,
+      "logps/rejected": -247.86673964968153,
+      "loss": 0.5032,
+      "loss/base_kto": 0.4993653893470764,
+      "loss/total_with_kl": 0.5040350556373596,
+      "metrics/advantage_var": 0.09339304268360138,
+      "regularization/advantage_var": 0.09339304268360138,
+      "regularization/kl": 0.004669652320444584,
+      "rewards/chosen": 0.005291132473506811,
+      "rewards/margins": 0.0048918331279284525,
+      "rewards/rejected": 0.0003992993455783577,
+      "stability/repetition_rate_mean": 0.2191840261220932,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 30.20406723022461,
+      "stability/response_length_var": 912.2857055664062,
+      "stability/token_entropy_mean": 3.416015625,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.8735517263412476,
+      "kl": 0.03989999368786812,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57242692.266666666,
+      "logits/rejected": -50260034.06451613,
+      "logps/chosen": -367.13778409090907,
+      "logps/rejected": -190.0336693548387,
+      "loss": 0.5028,
+      "loss/base_kto": 0.4993356168270111,
+      "loss/total_with_kl": 0.5030570030212402,
+      "metrics/advantage_var": 0.07442746311426163,
+      "regularization/advantage_var": 0.07442746311426163,
+      "regularization/kl": 0.0037213729228824377,
+      "rewards/chosen": 0.005194586696046772,
+      "rewards/margins": 0.0051130529737076335,
+      "rewards/rejected": 8.15337223391379e-05,
+      "stability/repetition_rate_mean": 0.3268229067325592,
+      "stability/response_length_mean": 52.75,
+      "stability/response_length_std": 22.72663688659668,
+      "stability/response_length_var": 516.5,
+      "stability/token_entropy_mean": 4.783203125,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.9999533891677856,
+      "kl": 0.047912172973155975,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51305524.43373494,
+      "logits/rejected": -51872003.32467532,
+      "logps/chosen": -330.2233151355422,
+      "logps/rejected": -199.49925172483765,
+      "loss": 0.5032,
+      "loss/base_kto": 0.4992775022983551,
+      "loss/total_with_kl": 0.502557635307312,
+      "metrics/advantage_var": 0.06560272723436356,
+      "regularization/advantage_var": 0.06560272723436356,
+      "regularization/kl": 0.003280136501416564,
+      "rewards/chosen": 0.004852092768772539,
+      "rewards/margins": 0.005986405444143338,
+      "rewards/rejected": -0.0011343126753707985,
+      "stability/repetition_rate_mean": 0.24838952720165253,
+      "stability/response_length_mean": 28.75,
+      "stability/response_length_std": 29.86517333984375,
+      "stability/response_length_var": 891.9285888671875,
+      "stability/token_entropy_mean": 5.314453125,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 1.1828984022140503,
+      "kl": 0.04824293777346611,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55574064.1882353,
+      "logits/rejected": -50405259.946666665,
+      "logps/chosen": -397.8425551470588,
+      "logps/rejected": -207.70329427083334,
+      "loss": 0.5029,
+      "loss/base_kto": 0.49907246232032776,
+      "loss/total_with_kl": 0.5030288100242615,
+      "metrics/advantage_var": 0.07912679761648178,
+      "regularization/advantage_var": 0.07912679761648178,
+      "regularization/kl": 0.003956340253353119,
+      "rewards/chosen": 0.009192322282230153,
+      "rewards/margins": 0.006929297634199555,
+      "rewards/rejected": 0.002263024648030599,
+      "stability/repetition_rate_mean": 0.20100826025009155,
+      "stability/response_length_mean": 24.125,
+      "stability/response_length_std": 25.637235641479492,
+      "stability/response_length_var": 657.2678833007812,
+      "stability/token_entropy_mean": 4.59765625,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 1.0469462871551514,
+      "kl": 0.04922879859805107,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54217438.870588236,
+      "logits/rejected": -49562466.986666664,
+      "logps/chosen": -360.94535845588234,
+      "logps/rejected": -191.27700520833332,
+      "loss": 0.503,
+      "loss/base_kto": 0.4997846782207489,
+      "loss/total_with_kl": 0.5039902925491333,
+      "metrics/advantage_var": 0.08411123603582382,
+      "regularization/advantage_var": 0.08411123603582382,
+      "regularization/kl": 0.004205561708658934,
+      "rewards/chosen": 0.004395649012397317,
+      "rewards/margins": 0.0015353484714732452,
+      "rewards/rejected": 0.002860300540924072,
+      "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": 4.6044921875,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.9236876964569092,
+      "kl": 0.04486330226063728,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55477516.190476194,
+      "logits/rejected": -49231777.684210524,
+      "logps/chosen": -416.8061755952381,
+      "logps/rejected": -186.89340049342104,
+      "loss": 0.5031,
+      "loss/base_kto": 0.49898120760917664,
+      "loss/total_with_kl": 0.5027669072151184,
+      "metrics/advantage_var": 0.07571418583393097,
+      "regularization/advantage_var": 0.07571418583393097,
+      "regularization/kl": 0.003785709384828806,
+      "rewards/chosen": 0.008917692161741712,
+      "rewards/margins": 0.008523785749585707,
+      "rewards/rejected": 0.00039390641215600466,
+      "stability/repetition_rate_mean": 0.291015625,
+      "stability/response_length_mean": 26.25,
+      "stability/response_length_std": 31.535694122314453,
+      "stability/response_length_var": 994.5,
+      "stability/token_entropy_mean": 3.771484375,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.9776092171669006,
+      "kl": 0.04533420130610466,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54658201.01149426,
+      "logits/rejected": -50095780.82191781,
+      "logps/chosen": -370.85578304597703,
+      "logps/rejected": -212.5767604880137,
+      "loss": 0.5029,
+      "loss/base_kto": 0.4993962347507477,
+      "loss/total_with_kl": 0.5035357475280762,
+      "metrics/advantage_var": 0.0827893316745758,
+      "regularization/advantage_var": 0.0827893316745758,
+      "regularization/kl": 0.004139466676861048,
+      "rewards/chosen": 0.00609781139198391,
+      "rewards/margins": 0.004562121499966231,
+      "rewards/rejected": 0.001535689892017678,
+      "stability/repetition_rate_mean": 0.20468750596046448,
+      "stability/response_length_mean": 12.625,
+      "stability/response_length_std": 21.460512161254883,
+      "stability/response_length_var": 460.5535583496094,
+      "stability/token_entropy_mean": 4.81640625,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.823735773563385,
+      "kl": 0.0640627071261406,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59383468.686390534,
+      "logits/rejected": -52950897.58940397,
+      "logps/chosen": -441.2536982248521,
+      "logps/rejected": -191.25607926324503,
+      "loss": 0.5029,
+      "loss/base_kto": 0.4989679455757141,
+      "loss/total_with_kl": 0.5037684440612793,
+      "metrics/advantage_var": 0.09601064026355743,
+      "regularization/advantage_var": 0.09601064026355743,
+      "regularization/kl": 0.004800532013177872,
+      "rewards/chosen": 0.008814329931721885,
+      "rewards/margins": 0.00863874767439423,
+      "rewards/rejected": 0.0001755822573276545,
+      "stability/repetition_rate_mean": 0.21061541140079498,
+      "stability/response_length_mean": 28.25,
+      "stability/response_length_std": 30.844078063964844,
+      "stability/response_length_var": 951.3571166992188,
+      "stability/token_entropy_mean": 3.33251953125,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 1.019155502319336,
+      "kl": 0.058815985918045044,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56930439.71084338,
+      "logits/rejected": -48423304.31168831,
+      "logps/chosen": -356.34346762048193,
+      "logps/rejected": -194.7735643262987,
+      "loss": 0.5026,
+      "loss/base_kto": 0.49923601746559143,
+      "loss/total_with_kl": 0.5028472542762756,
+      "metrics/advantage_var": 0.07222381234169006,
+      "regularization/advantage_var": 0.07222381234169006,
+      "regularization/kl": 0.0036111907102167606,
+      "rewards/chosen": 0.00937136253678655,
+      "rewards/margins": 0.005975195153221019,
+      "rewards/rejected": 0.003396167383565531,
+      "stability/repetition_rate_mean": 0.2878348231315613,
+      "stability/response_length_mean": 32.25,
+      "stability/response_length_std": 27.932315826416016,
+      "stability/response_length_var": 780.2142944335938,
+      "stability/token_entropy_mean": 4.72265625,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.9375177025794983,
+      "kl": 0.06844082474708557,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52935007.80368098,
+      "logits/rejected": -48337998.26751592,
+      "logps/chosen": -376.0254265720859,
+      "logps/rejected": -211.25706608280254,
+      "loss": 0.5028,
+      "loss/base_kto": 0.49898138642311096,
+      "loss/total_with_kl": 0.5028926134109497,
+      "metrics/advantage_var": 0.07822488993406296,
+      "regularization/advantage_var": 0.07822488993406296,
+      "regularization/kl": 0.003911244682967663,
+      "rewards/chosen": 0.012041092650290647,
+      "rewards/margins": 0.007926249172421917,
+      "rewards/rejected": 0.00411484347786873,
+      "stability/repetition_rate_mean": 0.162109375,
+      "stability/response_length_mean": 40.375,
+      "stability/response_length_std": 32.61874008178711,
+      "stability/response_length_var": 1063.982177734375,
+      "stability/token_entropy_mean": 3.357421875,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.935251772403717,
+      "kl": 0.0741070955991745,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54936120.88888889,
+      "logits/rejected": -49878696.50632911,
+      "logps/chosen": -311.8755787037037,
+      "logps/rejected": -196.9727675039557,
+      "loss": 0.5028,
+      "loss/base_kto": 0.4992446005344391,
+      "loss/total_with_kl": 0.503296971321106,
+      "metrics/advantage_var": 0.08104676008224487,
+      "regularization/advantage_var": 0.08104676008224487,
+      "regularization/kl": 0.004052338190376759,
+      "rewards/chosen": 0.012579193821659795,
+      "rewards/margins": 0.005940466211035505,
+      "rewards/rejected": 0.0066387276106242895,
+      "stability/repetition_rate_mean": 0.18797221779823303,
+      "stability/response_length_mean": 24.125,
+      "stability/response_length_std": 29.019390106201172,
+      "stability/response_length_var": 842.125,
+      "stability/token_entropy_mean": 3.87109375,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.986784815788269,
+      "kl": 0.06636197119951248,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53416365.22155689,
+      "logits/rejected": -50075306.666666664,
+      "logps/chosen": -392.29271145209583,
+      "logps/rejected": -211.18162785947712,
+      "loss": 0.5028,
+      "loss/base_kto": 0.4987066388130188,
+      "loss/total_with_kl": 0.5025004744529724,
+      "metrics/advantage_var": 0.07587640732526779,
+      "regularization/advantage_var": 0.07587640732526779,
+      "regularization/kl": 0.0037938205059617758,
+      "rewards/chosen": 0.011261776535810824,
+      "rewards/margins": 0.009564409612522242,
+      "rewards/rejected": 0.0016973669232885822,
+      "stability/repetition_rate_mean": 0.1822916716337204,
+      "stability/response_length_mean": 17.625,
+      "stability/response_length_std": 28.680194854736328,
+      "stability/response_length_var": 822.5535888671875,
+      "stability/token_entropy_mean": 4.8212890625,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 1.358068823814392,
+      "kl": 0.05749697610735893,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60522562.06451613,
+      "logits/rejected": -54295080.339393936,
+      "logps/chosen": -392.64884072580645,
+      "logps/rejected": -197.9084872159091,
+      "loss": 0.5029,
+      "loss/base_kto": 0.49840664863586426,
+      "loss/total_with_kl": 0.5019518733024597,
+      "metrics/advantage_var": 0.07090426236391068,
+      "regularization/advantage_var": 0.07090426236391068,
+      "regularization/kl": 0.0035452134907245636,
+      "rewards/chosen": 0.013610498366817351,
+      "rewards/margins": 0.012526261972542033,
+      "rewards/rejected": 0.0010842363942753185,
+      "stability/repetition_rate_mean": 0.4140625,
+      "stability/response_length_mean": 48.125,
+      "stability/response_length_std": 29.39600372314453,
+      "stability/response_length_var": 864.125,
+      "stability/token_entropy_mean": 4.4609375,
+      "step": 195
+    },
+    {
+      "epoch": 1.0,
+      "step": 196,
+      "total_flos": 3.1861866257511875e+18,
+      "train_loss": 0.5030260469232287,
+      "train_runtime": 3154.7919,
+      "train_samples_per_second": 31.698,
+      "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_kl-beta0.1/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3dc25d257c329c7b76aee6e6676887abb30f25f66f6b540e02a495bf57fdef9
+size 8017