Upload folder using huggingface_hub

mistral7b_kto_kl-beta0.001/README.md

@@ -0,0 +1,64 @@
+---
+library_name: peft
+license: other
+base_model: mistralai/Mistral-7B-v0.1
+tags:
+- base_model:adapter:mistralai/Mistral-7B-v0.1
+- llama-factory
+- lora
+- transformers
+pipeline_tag: text-generation
+model-index:
+- name: mistral7b_kto_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-beta0.001/adapter_config.json

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

mistral7b_kto_kl-beta0.001/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:34e8b1f840c5b7b24f368d0e77ffd4f1f66945cfdd6c614f678260170dd90107
+size 42002584

mistral7b_kto_kl-beta0.001/all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.5000301818458401,
+    "train_runtime": 3151.2022,
+    "train_samples_per_second": 31.734,
+    "train_steps_per_second": 0.062
+}

mistral7b_kto_kl-beta0.001/chat_template.jinja

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

mistral7b_kto_kl-beta0.001/checkpoint-196/README.md

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

mistral7b_kto_kl-beta0.001/checkpoint-196/adapter_config.json

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

mistral7b_kto_kl-beta0.001/checkpoint-196/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:34e8b1f840c5b7b24f368d0e77ffd4f1f66945cfdd6c614f678260170dd90107
+size 42002584

mistral7b_kto_kl-beta0.001/checkpoint-196/chat_template.jinja

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

mistral7b_kto_kl-beta0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7fb8acfb0fb481466b6930ab74fb392923069be8c54e90f37b2e692b163b8d27
+size 31467333

mistral7b_kto_kl-beta0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80b156ef586b65e067617db049657016dfb2abfe1357db353a5daf483096ec28
+size 31467333

mistral7b_kto_kl-beta0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03685cfba56177be7572d0a3a37454abdd04c7b12ddaedec1d62daa2649a9414
+size 31467333

mistral7b_kto_kl-beta0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f922ddba5bb14e2bf4b398608cb0550d02f7243eef368f199a3f0aac86e9b9b
+size 31467397

mistral7b_kto_kl-beta0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4ef157f88f18577f67b29c67c16403d7b9c5a03e35bf3aa296ac0ac90949e185
+size 31467397

mistral7b_kto_kl-beta0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:445ba3a850e93f7ca1e25fcc40e05083bdbe9b16ab9d181c11a3ef95ba73cef9
+size 31467397

mistral7b_kto_kl-beta0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7704c697acc3cf20ebfc5be7c9ed45f904fb8405bd4a3aa4219d804a9c569ddb
+size 31467397

mistral7b_kto_kl-beta0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa75a0f0644d3aac28af7b93cb0d7b37df0e215981d1a91a83eff5fc2426c13d
+size 31467397

mistral7b_kto_kl-beta0.001/checkpoint-196/global_step196/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e37d71a05bc69594acef42c9695971945e22839adf4596eb18b3a2e26955e999
+size 42288409

mistral7b_kto_kl-beta0.001/checkpoint-196/latest

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

mistral7b_kto_kl-beta0.001/checkpoint-196/rng_state_0.pth

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

mistral7b_kto_kl-beta0.001/checkpoint-196/rng_state_1.pth

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

mistral7b_kto_kl-beta0.001/checkpoint-196/rng_state_2.pth

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

mistral7b_kto_kl-beta0.001/checkpoint-196/rng_state_3.pth

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

mistral7b_kto_kl-beta0.001/checkpoint-196/rng_state_4.pth

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

mistral7b_kto_kl-beta0.001/checkpoint-196/rng_state_5.pth

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

mistral7b_kto_kl-beta0.001/checkpoint-196/rng_state_6.pth

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

mistral7b_kto_kl-beta0.001/checkpoint-196/rng_state_7.pth

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

mistral7b_kto_kl-beta0.001/checkpoint-196/scheduler.pt

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

mistral7b_kto_kl-beta0.001/checkpoint-196/special_tokens_map.json

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

mistral7b_kto_kl-beta0.001/checkpoint-196/tokenizer.model

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

mistral7b_kto_kl-beta0.001/checkpoint-196/tokenizer_config.json

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

mistral7b_kto_kl-beta0.001/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": 0.010321159847080708,
+      "kl": 0.009334363043308258,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59716823.578947365,
+      "logits/rejected": -50418389.333333336,
+      "logps/chosen": -439.0208675986842,
+      "logps/rejected": -198.4043433779762,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999978542327881,
+      "loss/total_with_kl": 0.5000227689743042,
+      "metrics/advantage_var": 0.04985074698925018,
+      "regularization/advantage_var": 0.04985074698925018,
+      "regularization/kl": 2.4925377147155814e-05,
+      "rewards/chosen": 2.4742100657405037e-05,
+      "rewards/margins": 1.9698174008609432e-05,
+      "rewards/rejected": 5.043926648795605e-06,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.01033781562000513,
+      "kl": 0.021588662639260292,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59296382.33766234,
+      "logits/rejected": -50010654.84337349,
+      "logps/chosen": -429.95926339285717,
+      "logps/rejected": -192.84822100903614,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000049471855164,
+      "loss/total_with_kl": 0.5000373721122742,
+      "metrics/advantage_var": 0.0648464560508728,
+      "regularization/advantage_var": 0.0648464560508728,
+      "regularization/kl": 3.242322782170959e-05,
+      "rewards/chosen": -2.0726006357120228e-05,
+      "rewards/margins": -4.044272520838349e-05,
+      "rewards/rejected": 1.9716718851263263e-05,
+      "stability/repetition_rate_mean": 0.3668472170829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.26171875,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.010668128728866577,
+      "kl": 0.016281843185424805,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56242951.75757576,
+      "logits/rejected": -51922435.30322581,
+      "logps/chosen": -404.10520833333334,
+      "logps/rejected": -200.6876764112903,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999719858169556,
+      "loss/total_with_kl": 0.5000293850898743,
+      "metrics/advantage_var": 0.0644526481628418,
+      "regularization/advantage_var": 0.0644526481628418,
+      "regularization/kl": 3.2226329494733363e-05,
+      "rewards/chosen": 2.561494369398464e-05,
+      "rewards/margins": 1.9232726033197994e-05,
+      "rewards/rejected": 6.382217660786644e-06,
+      "stability/repetition_rate_mean": 0.27336087822914124,
+      "stability/response_length_mean": 28.125,
+      "stability/response_length_std": 30.243947982788086,
+      "stability/response_length_var": 914.6964111328125,
+      "stability/token_entropy_mean": 3.7587890625,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.011172862723469734,
+      "kl": 0.02222653105854988,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51676390.4,
+      "logits/rejected": -48065916.8,
+      "logps/chosen": -343.68896484375,
+      "logps/rejected": -225.6365966796875,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000028014183044,
+      "loss/total_with_kl": 0.5000383257865906,
+      "metrics/advantage_var": 0.07106568664312363,
+      "regularization/advantage_var": 0.07106568664312363,
+      "regularization/kl": 3.5532844776753336e-05,
+      "rewards/chosen": 7.749405631329864e-06,
+      "rewards/margins": -1.9000891188625244e-05,
+      "rewards/rejected": 2.675029681995511e-05,
+      "stability/repetition_rate_mean": 0.1943204402923584,
+      "stability/response_length_mean": 42.125,
+      "stability/response_length_std": 27.01553726196289,
+      "stability/response_length_var": 729.8392944335938,
+      "stability/token_entropy_mean": 4.4296875,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.009160108864307404,
+      "kl": 0.013138818554580212,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56473031.50344828,
+      "logits/rejected": -52018123.337142855,
+      "logps/chosen": -378.4786099137931,
+      "logps/rejected": -202.77325892857144,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000008940696716,
+      "loss/total_with_kl": 0.5000317096710205,
+      "metrics/advantage_var": 0.061832915991544724,
+      "regularization/advantage_var": 0.061832915991544724,
+      "regularization/kl": 3.091646431130357e-05,
+      "rewards/chosen": 1.283879873567614e-05,
+      "rewards/margins": -6.746106088308278e-06,
+      "rewards/rejected": 1.958490482398442e-05,
+      "stability/repetition_rate_mean": 0.17997129261493683,
+      "stability/response_length_mean": 43.375,
+      "stability/response_length_std": 28.54538917541504,
+      "stability/response_length_var": 814.8392944335938,
+      "stability/token_entropy_mean": 4.7890625,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.010463796555995941,
+      "kl": 0.01257875096052885,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57743002.63087248,
+      "logits/rejected": -51520362.29239766,
+      "logps/chosen": -468.58819211409394,
+      "logps/rejected": -198.2330957602339,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000025629997253,
+      "loss/total_with_kl": 0.500034749507904,
+      "metrics/advantage_var": 0.06440309435129166,
+      "regularization/advantage_var": 0.06440309435129166,
+      "regularization/kl": 3.220155122107826e-05,
+      "rewards/chosen": -6.928566480743005e-06,
+      "rewards/margins": -2.1701127549079173e-05,
+      "rewards/rejected": 1.4772561068336168e-05,
+      "stability/repetition_rate_mean": 0.1953125,
+      "stability/response_length_mean": 48.375,
+      "stability/response_length_std": 28.93311309814453,
+      "stability/response_length_var": 837.125,
+      "stability/token_entropy_mean": 4.513671875,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.014569000340998173,
+      "kl": 0.022648466750979424,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57641219.24050633,
+      "logits/rejected": -53900932.74074074,
+      "logps/chosen": -385.8970530063291,
+      "logps/rejected": -195.1368152006173,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999943673610687,
+      "loss/total_with_kl": 0.5000303387641907,
+      "metrics/advantage_var": 0.07199499756097794,
+      "regularization/advantage_var": 0.07199499756097794,
+      "regularization/kl": 3.599749834393151e-05,
+      "rewards/chosen": 3.946519492170479e-05,
+      "rewards/margins": 4.9802977542996476e-05,
+      "rewards/rejected": -1.0337782621291684e-05,
+      "stability/repetition_rate_mean": 0.35856893658638,
+      "stability/response_length_mean": 53.375,
+      "stability/response_length_std": 21.14533042907715,
+      "stability/response_length_var": 447.125,
+      "stability/token_entropy_mean": 5.330078125,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.009226334281265736,
+      "kl": 0.017865750938653946,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56646547.2,
+      "logits/rejected": -50243718.4,
+      "logps/chosen": -421.89150390625,
+      "logps/rejected": -219.2673583984375,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000039339065552,
+      "loss/total_with_kl": 0.5000437498092651,
+      "metrics/advantage_var": 0.07968886941671371,
+      "regularization/advantage_var": 0.07968886941671371,
+      "regularization/kl": 3.984443537774496e-05,
+      "rewards/chosen": 1.1162247392348946e-05,
+      "rewards/margins": -3.0723534291610124e-05,
+      "rewards/rejected": 4.188578168395907e-05,
+      "stability/repetition_rate_mean": 0.142578125,
+      "stability/response_length_mean": 33.125,
+      "stability/response_length_std": 33.01704025268555,
+      "stability/response_length_var": 1090.125,
+      "stability/token_entropy_mean": 4.9609375,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.012385486625134945,
+      "kl": 0.03356611356139183,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52715269.15436242,
+      "logits/rejected": -50790807.204678364,
+      "logps/chosen": -352.28240352348996,
+      "logps/rejected": -243.81743421052633,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000016093254089,
+      "loss/total_with_kl": 0.5000400543212891,
+      "metrics/advantage_var": 0.07688915729522705,
+      "regularization/advantage_var": 0.07688915729522705,
+      "regularization/kl": 3.844457751256414e-05,
+      "rewards/chosen": 4.6062358223751884e-05,
+      "rewards/margins": -1.0668867753415165e-05,
+      "rewards/rejected": 5.673122597716705e-05,
+      "stability/repetition_rate_mean": 0.1013549193739891,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.87324333190918,
+      "stability/response_length_var": 892.4107055664062,
+      "stability/token_entropy_mean": 4.775390625,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.00840709451586008,
+      "kl": 0.032784707844257355,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54045579.167785235,
+      "logits/rejected": -48300567.953216374,
+      "logps/chosen": -375.1681260486577,
+      "logps/rejected": -180.20900721856725,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999998211860657,
+      "loss/total_with_kl": 0.5000237226486206,
+      "metrics/advantage_var": 0.04776905104517937,
+      "regularization/advantage_var": 0.04776905104517937,
+      "regularization/kl": 2.388452776358463e-05,
+      "rewards/chosen": 3.098852732437569e-05,
+      "rewards/margins": -6.186709812701927e-07,
+      "rewards/rejected": 3.160719830564588e-05,
+      "stability/repetition_rate_mean": 0.3831075429916382,
+      "stability/response_length_mean": 53.375,
+      "stability/response_length_std": 21.480472564697266,
+      "stability/response_length_var": 461.41070556640625,
+      "stability/token_entropy_mean": 3.857421875,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.011991314589977264,
+      "kl": 0.03365620970726013,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56956018.48447205,
+      "logits/rejected": -49576486.641509436,
+      "logps/chosen": -344.5125679347826,
+      "logps/rejected": -230.43519064465409,
+      "loss": 0.5,
+      "loss/base_kto": 0.500001847743988,
+      "loss/total_with_kl": 0.5000297427177429,
+      "metrics/advantage_var": 0.055924512445926666,
+      "regularization/advantage_var": 0.055924512445926666,
+      "regularization/kl": 2.796225635393057e-05,
+      "rewards/chosen": -5.357490956598187e-07,
+      "rewards/margins": -1.33645037668121e-05,
+      "rewards/rejected": 1.2828754671152282e-05,
+      "stability/repetition_rate_mean": 0.11199301481246948,
+      "stability/response_length_mean": 22.375,
+      "stability/response_length_std": 30.236862182617188,
+      "stability/response_length_var": 914.2678833007812,
+      "stability/token_entropy_mean": 4.333984375,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.00939466804265976,
+      "kl": 0.024347607046365738,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52675459.878787875,
+      "logits/rejected": -47832952.56774194,
+      "logps/chosen": -359.6363636363636,
+      "logps/rejected": -178.1663054435484,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999696016311646,
+      "loss/total_with_kl": 0.5000283122062683,
+      "metrics/advantage_var": 0.06277040392160416,
+      "regularization/advantage_var": 0.06277040392160416,
+      "regularization/kl": 3.138520332868211e-05,
+      "rewards/chosen": 8.442694264830965e-05,
+      "rewards/margins": 2.3823355270853723e-05,
+      "rewards/rejected": 6.060358737745593e-05,
+      "stability/repetition_rate_mean": 0.23800872266292572,
+      "stability/response_length_mean": 53.625,
+      "stability/response_length_std": 22.11617088317871,
+      "stability/response_length_var": 489.125,
+      "stability/token_entropy_mean": 5.255859375,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.008829456754028797,
+      "kl": 0.03643302991986275,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57575225.80645161,
+      "logits/rejected": -47288897.163636364,
+      "logps/chosen": -381.90148689516127,
+      "logps/rejected": -183.55746922348484,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999722838401794,
+      "loss/total_with_kl": 0.5000311732292175,
+      "metrics/advantage_var": 0.06777071952819824,
+      "regularization/advantage_var": 0.06777071952819824,
+      "regularization/kl": 3.388536424608901e-05,
+      "rewards/chosen": 4.530613941531027e-05,
+      "rewards/margins": 2.0425156931431993e-05,
+      "rewards/rejected": 2.488098248387828e-05,
+      "stability/repetition_rate_mean": 0.1857219785451889,
+      "stability/response_length_mean": 21.5,
+      "stability/response_length_std": 26.635099411010742,
+      "stability/response_length_var": 709.4285888671875,
+      "stability/token_entropy_mean": 5.392578125,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.007817639969289303,
+      "kl": 0.04110654443502426,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52791732.02580645,
+      "logits/rejected": -49063154.03636364,
+      "logps/chosen": -379.0649193548387,
+      "logps/rejected": -220.3504734848485,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000015497207642,
+      "loss/total_with_kl": 0.5000338554382324,
+      "metrics/advantage_var": 0.06463103741407394,
+      "regularization/advantage_var": 0.06463103741407394,
+      "regularization/kl": 3.2315518183168024e-05,
+      "rewards/chosen": 6.355317609925424e-05,
+      "rewards/margins": -8.118214524503094e-06,
+      "rewards/rejected": 7.167139062375734e-05,
+      "stability/repetition_rate_mean": 0.171875,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.6875,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.009152984246611595,
+      "kl": 0.03520666062831879,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49518003.89189189,
+      "logits/rejected": -47914368.0,
+      "logps/chosen": -314.1993771114865,
+      "logps/rejected": -228.88335755813952,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999912679195404,
+      "loss/total_with_kl": 0.50002521276474,
+      "metrics/advantage_var": 0.06784234941005707,
+      "regularization/advantage_var": 0.06784234941005707,
+      "regularization/kl": 3.3921172871487215e-05,
+      "rewards/chosen": 7.1432428887567e-05,
+      "rewards/margins": 6.99527711958152e-05,
+      "rewards/rejected": 1.4796576917518016e-06,
+      "stability/repetition_rate_mean": 0.14558292925357819,
+      "stability/response_length_mean": 24.25,
+      "stability/response_length_std": 30.093900680541992,
+      "stability/response_length_var": 905.6428833007812,
+      "stability/token_entropy_mean": 2.78125,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.010503264144062996,
+      "kl": 0.051506686955690384,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58671863.52317881,
+      "logits/rejected": -53488839.95266272,
+      "logps/chosen": -403.9934292218543,
+      "logps/rejected": -201.02734375,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999984800815582,
+      "loss/total_with_kl": 0.5000309348106384,
+      "metrics/advantage_var": 0.06490852683782578,
+      "regularization/advantage_var": 0.06490852683782578,
+      "regularization/kl": 3.245426341891289e-05,
+      "rewards/chosen": 3.66592804426389e-05,
+      "rewards/margins": 8.198070828566575e-06,
+      "rewards/rejected": 2.8461209614072326e-05,
+      "stability/repetition_rate_mean": 0.19088542461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.234375,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.009465193375945091,
+      "kl": 0.03408239036798477,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51497151.58709677,
+      "logits/rejected": -47171549.86666667,
+      "logps/chosen": -358.2992943548387,
+      "logps/rejected": -230.4330018939394,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999475479125977,
+      "loss/total_with_kl": 0.5000323057174683,
+      "metrics/advantage_var": 0.07506581395864487,
+      "regularization/advantage_var": 0.07506581395864487,
+      "regularization/kl": 3.753290729946457e-05,
+      "rewards/chosen": 3.3219540191273537e-05,
+      "rewards/margins": 4.1400590949743726e-05,
+      "rewards/rejected": -8.181050758470188e-06,
+      "stability/repetition_rate_mean": 0.15742863714694977,
+      "stability/response_length_mean": 24.875,
+      "stability/response_length_std": 27.424898147583008,
+      "stability/response_length_var": 752.125,
+      "stability/token_entropy_mean": 4.921875,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.012723367661237717,
+      "kl": 0.044009942561388016,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58462437.743589744,
+      "logits/rejected": -50612860.87804878,
+      "logps/chosen": -380.76332131410254,
+      "logps/rejected": -187.201064691311,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999934434890747,
+      "loss/total_with_kl": 0.5000306963920593,
+      "metrics/advantage_var": 0.06280595809221268,
+      "regularization/advantage_var": 0.06280595809221268,
+      "regularization/kl": 3.140298213111237e-05,
+      "rewards/chosen": 2.6056644482872424e-05,
+      "rewards/margins": 3.132729029854866e-07,
+      "rewards/rejected": 2.5743371579886937e-05,
+      "stability/repetition_rate_mean": 0.2192234843969345,
+      "stability/response_length_mean": 33.625,
+      "stability/response_length_std": 29.442134857177734,
+      "stability/response_length_var": 866.8392944335938,
+      "stability/token_entropy_mean": 3.70263671875,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.011232409626245499,
+      "kl": 0.043333958834409714,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50442868.658227846,
+      "logits/rejected": -49033930.27160494,
+      "logps/chosen": -314.9471172863924,
+      "logps/rejected": -207.13204089506172,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999579787254333,
+      "loss/total_with_kl": 0.5000306367874146,
+      "metrics/advantage_var": 0.06963532418012619,
+      "regularization/advantage_var": 0.06963532418012619,
+      "regularization/kl": 3.48176654370036e-05,
+      "rewards/chosen": 4.625187514797796e-05,
+      "rewards/margins": 3.456790598741307e-05,
+      "rewards/rejected": 1.1683969160564887e-05,
+      "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": 3.74609375,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.010943830944597721,
+      "kl": 0.04752064868807793,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54759797.62162162,
+      "logits/rejected": -50968784.37209302,
+      "logps/chosen": -359.7860272381757,
+      "logps/rejected": -216.5623410247093,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999910295009613,
+      "loss/total_with_kl": 0.500036895275116,
+      "metrics/advantage_var": 0.09180708229541779,
+      "regularization/advantage_var": 0.09180708229541779,
+      "regularization/kl": 4.5903540012659505e-05,
+      "rewards/chosen": 8.41418545491792e-05,
+      "rewards/margins": 6.757945212752732e-05,
+      "rewards/rejected": 1.6562402421651884e-05,
+      "stability/repetition_rate_mean": 0.24177631735801697,
+      "stability/response_length_mean": 38.0,
+      "stability/response_length_std": 30.10457992553711,
+      "stability/response_length_var": 906.2857055664062,
+      "stability/token_entropy_mean": 4.23828125,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.01067819632589817,
+      "kl": 0.04482289031147957,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59449306.98795181,
+      "logits/rejected": -53884203.22077922,
+      "logps/chosen": -399.820265436747,
+      "logps/rejected": -209.97304941152598,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999380111694336,
+      "loss/total_with_kl": 0.5000303983688354,
+      "metrics/advantage_var": 0.073161780834198,
+      "regularization/advantage_var": 0.073161780834198,
+      "regularization/kl": 3.658089190139435e-05,
+      "rewards/chosen": 8.220696844250323e-05,
+      "rewards/margins": 5.143789996083652e-05,
+      "rewards/rejected": 3.0769068481666704e-05,
+      "stability/repetition_rate_mean": 0.2014508843421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.384765625,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.00925524067133665,
+      "kl": 0.050605691969394684,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57465559.90361446,
+      "logits/rejected": -53690846.753246754,
+      "logps/chosen": -378.5371799698795,
+      "logps/rejected": -206.05924056412337,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000004768371582,
+      "loss/total_with_kl": 0.5000382661819458,
+      "metrics/advantage_var": 0.07546661794185638,
+      "regularization/advantage_var": 0.07546661794185638,
+      "regularization/kl": 3.773330900003202e-05,
+      "rewards/chosen": 2.6093964206885144e-05,
+      "rewards/margins": -2.7925838551514204e-06,
+      "rewards/rejected": 2.8886548062036564e-05,
+      "stability/repetition_rate_mean": 0.30853456258773804,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.911476135253906,
+      "stability/response_length_var": 894.6964111328125,
+      "stability/token_entropy_mean": 4.412109375,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.009009272791445255,
+      "kl": 0.04474036023020744,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60689090.43037975,
+      "logits/rejected": -53511787.45679013,
+      "logps/chosen": -364.50944422468353,
+      "logps/rejected": -201.17226080246914,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999332427978516,
+      "loss/total_with_kl": 0.5000336766242981,
+      "metrics/advantage_var": 0.08062459528446198,
+      "regularization/advantage_var": 0.08062459528446198,
+      "regularization/kl": 4.031229764223099e-05,
+      "rewards/chosen": 8.311558844922464e-05,
+      "rewards/margins": 5.8828055319233806e-05,
+      "rewards/rejected": 2.4287533129990837e-05,
+      "stability/repetition_rate_mean": 0.2506510615348816,
+      "stability/response_length_mean": 33.125,
+      "stability/response_length_std": 33.0516357421875,
+      "stability/response_length_var": 1092.4107666015625,
+      "stability/token_entropy_mean": 4.142578125,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.008836813271045685,
+      "kl": 0.03717491030693054,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57836122.35294118,
+      "logits/rejected": -54507458.68263473,
+      "logps/chosen": -384.19738051470586,
+      "logps/rejected": -212.1802020958084,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999862909317017,
+      "loss/total_with_kl": 0.5000400543212891,
+      "metrics/advantage_var": 0.08291424810886383,
+      "regularization/advantage_var": 0.08291424810886383,
+      "regularization/kl": 4.145712227909826e-05,
+      "rewards/chosen": 4.753693207806232e-05,
+      "rewards/margins": 1.1379732890869236e-05,
+      "rewards/rejected": 3.6157199187193086e-05,
+      "stability/repetition_rate_mean": 0.17961646616458893,
+      "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": 0.009799664840102196,
+      "kl": 0.03611461818218231,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60827171.06849315,
+      "logits/rejected": -55819781.88505747,
+      "logps/chosen": -417.35030500856163,
+      "logps/rejected": -216.33441091954023,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000001192092896,
+      "loss/total_with_kl": 0.5000359416007996,
+      "metrics/advantage_var": 0.0716974139213562,
+      "regularization/advantage_var": 0.0716974139213562,
+      "regularization/kl": 3.5848712286679074e-05,
+      "rewards/chosen": 2.899690059154001e-05,
+      "rewards/margins": -3.782787809415023e-06,
+      "rewards/rejected": 3.2779688400955034e-05,
+      "stability/repetition_rate_mean": 0.16328126192092896,
+      "stability/response_length_mean": 15.25,
+      "stability/response_length_std": 24.93276596069336,
+      "stability/response_length_var": 621.6428833007812,
+      "stability/token_entropy_mean": 3.625,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.009806789457798004,
+      "kl": 0.04913714528083801,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57520075.1483871,
+      "logits/rejected": -53990958.54545455,
+      "logps/chosen": -392.6003024193548,
+      "logps/rejected": -183.9016690340909,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999481439590454,
+      "loss/total_with_kl": 0.5000341534614563,
+      "metrics/advantage_var": 0.0785689577460289,
+      "regularization/advantage_var": 0.0785689577460289,
+      "regularization/kl": 3.92844776797574e-05,
+      "rewards/chosen": 5.7653367759719974e-05,
+      "rewards/margins": 4.0403727117929525e-05,
+      "rewards/rejected": 1.7249640641790448e-05,
+      "stability/repetition_rate_mean": 0.22703124582767487,
+      "stability/response_length_mean": 43.25,
+      "stability/response_length_std": 29.620214462280273,
+      "stability/response_length_var": 877.3571166992188,
+      "stability/token_entropy_mean": 3.57421875,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.009469456970691681,
+      "kl": 0.05869415029883385,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59993232.49079755,
+      "logits/rejected": -55193215.18471338,
+      "logps/chosen": -401.71668903374234,
+      "logps/rejected": -247.83942078025478,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999094009399414,
+      "loss/total_with_kl": 0.5000352263450623,
+      "metrics/advantage_var": 0.08847339451313019,
+      "regularization/advantage_var": 0.08847339451313019,
+      "regularization/kl": 4.423669815878384e-05,
+      "rewards/chosen": 9.945813177919095e-05,
+      "rewards/margins": 6.812238380639579e-05,
+      "rewards/rejected": 3.1335747972795155e-05,
+      "stability/repetition_rate_mean": 0.2230902761220932,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 30.20406723022461,
+      "stability/response_length_var": 912.2857055664062,
+      "stability/token_entropy_mean": 3.427734375,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.008573731407523155,
+      "kl": 0.0394296832382679,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57256320.77575757,
+      "logits/rejected": -50278221.62580645,
+      "logps/chosen": -367.1009943181818,
+      "logps/rejected": -190.01449092741936,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999907910823822,
+      "loss/total_with_kl": 0.5000235438346863,
+      "metrics/advantage_var": 0.0654965192079544,
+      "regularization/advantage_var": 0.0654965192079544,
+      "regularization/kl": 3.274825940025039e-05,
+      "rewards/chosen": 8.873008756023464e-05,
+      "rewards/margins": 6.875481660444132e-05,
+      "rewards/rejected": 1.997527095579332e-05,
+      "stability/repetition_rate_mean": 0.34765625,
+      "stability/response_length_mean": 56.25,
+      "stability/response_length_std": 21.920310974121094,
+      "stability/response_length_var": 480.5,
+      "stability/token_entropy_mean": 4.7548828125,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.010250517167150974,
+      "kl": 0.051040973514318466,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51322105.8313253,
+      "logits/rejected": -51878599.48051948,
+      "logps/chosen": -330.19470067771084,
+      "logps/rejected": -199.5067091112013,
+      "loss": 0.5,
+      "loss/base_kto": 0.49998751282691956,
+      "loss/total_with_kl": 0.5000160336494446,
+      "metrics/advantage_var": 0.05701936408877373,
+      "regularization/advantage_var": 0.05701936408877373,
+      "regularization/kl": 2.8509681214927696e-05,
+      "rewards/chosen": 7.712891527328147e-05,
+      "rewards/margins": 9.591767240252724e-05,
+      "rewards/rejected": -1.8788757129245763e-05,
+      "stability/repetition_rate_mean": 0.25424888730049133,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.73423957824707,
+      "stability/response_length_var": 884.125,
+      "stability/token_entropy_mean": 5.306640625,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.01047827210277319,
+      "kl": 0.056698884814977646,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55602344.65882353,
+      "logits/rejected": -50412625.92,
+      "logps/chosen": -397.7995863970588,
+      "logps/rejected": -207.72515625,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999827444553375,
+      "loss/total_with_kl": 0.5000240206718445,
+      "metrics/advantage_var": 0.08249103277921677,
+      "regularization/advantage_var": 0.08249103277921677,
+      "regularization/kl": 4.1245522879762575e-05,
+      "rewards/chosen": 0.00013485204647569095,
+      "rewards/margins": 0.00013406771958769098,
+      "rewards/rejected": 7.843268879999717e-07,
+      "stability/repetition_rate_mean": 0.20331649482250214,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.193626403808594,
+      "stability/response_length_var": 852.2678833007812,
+      "stability/token_entropy_mean": 4.59765625,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.009275847114622593,
+      "kl": 0.050125040113925934,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54248122.729411766,
+      "logits/rejected": -49590394.88,
+      "logps/chosen": -360.94453125,
+      "logps/rejected": -191.30770833333332,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999949038028717,
+      "loss/total_with_kl": 0.5000295639038086,
+      "metrics/advantage_var": 0.06931816786527634,
+      "regularization/advantage_var": 0.06931816786527634,
+      "regularization/kl": 3.465908230282366e-05,
+      "rewards/chosen": 4.4789784314001306e-05,
+      "rewards/margins": 4.690334328687658e-05,
+      "rewards/rejected": -2.113558972875277e-06,
+      "stability/repetition_rate_mean": 0.28984373807907104,
+      "stability/response_length_mean": 53.125,
+      "stability/response_length_std": 22.67432403564453,
+      "stability/response_length_var": 514.125,
+      "stability/token_entropy_mean": 4.5673828125,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.009821859188377857,
+      "kl": 0.049432478845119476,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55498038.85714286,
+      "logits/rejected": -49266189.473684214,
+      "logps/chosen": -416.84291294642856,
+      "logps/rejected": -186.8680998149671,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999743700027466,
+      "loss/total_with_kl": 0.5000368356704712,
+      "metrics/advantage_var": 0.07872576266527176,
+      "regularization/advantage_var": 0.07872576266527176,
+      "regularization/kl": 3.936288339900784e-05,
+      "rewards/chosen": 5.241769499012402e-05,
+      "rewards/margins": 2.316671556660107e-05,
+      "rewards/rejected": 2.925097942352295e-05,
+      "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": 3.828125,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.008929993025958538,
+      "kl": 0.04661773145198822,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54674326.06896552,
+      "logits/rejected": -50117249.75342466,
+      "logps/chosen": -370.8756061422414,
+      "logps/rejected": -212.5759578339041,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999971389770508,
+      "loss/total_with_kl": 0.5000324845314026,
+      "metrics/advantage_var": 0.0706528052687645,
+      "regularization/advantage_var": 0.0706528052687645,
+      "regularization/kl": 3.532640403136611e-05,
+      "rewards/chosen": 4.114243255435735e-05,
+      "rewards/margins": 2.4976168516181596e-05,
+      "rewards/rejected": 1.6166264038175754e-05,
+      "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": 4.83984375,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.008566767908632755,
+      "kl": 0.062040429562330246,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59409420.1183432,
+      "logits/rejected": -52976277.19205298,
+      "logps/chosen": -441.2774593195266,
+      "logps/rejected": -191.24706384519868,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999937117099762,
+      "loss/total_with_kl": 0.5000309944152832,
+      "metrics/advantage_var": 0.07455558329820633,
+      "regularization/advantage_var": 0.07455558329820633,
+      "regularization/kl": 3.727779403561726e-05,
+      "rewards/chosen": 6.438495822735792e-05,
+      "rewards/margins": 5.361681233712991e-05,
+      "rewards/rejected": 1.0768145890228007e-05,
+      "stability/repetition_rate_mean": 0.20866228640079498,
+      "stability/response_length_mean": 28.25,
+      "stability/response_length_std": 30.844078063964844,
+      "stability/response_length_var": 951.3571166992188,
+      "stability/token_entropy_mean": 3.31884765625,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.010120969265699387,
+      "kl": 0.06263448297977448,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56958445.4939759,
+      "logits/rejected": -48456025.766233765,
+      "logps/chosen": -356.2844973644578,
+      "logps/rejected": -194.80245535714286,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999809265136719,
+      "loss/total_with_kl": 0.5000235438346863,
+      "metrics/advantage_var": 0.08520467579364777,
+      "regularization/advantage_var": 0.08520467579364777,
+      "regularization/kl": 4.260233981767669e-05,
+      "rewards/chosen": 0.00015270808734089495,
+      "rewards/margins": 0.0001476406334579356,
+      "rewards/rejected": 5.0674538829593685e-06,
+      "stability/repetition_rate_mean": 0.2936941981315613,
+      "stability/response_length_mean": 32.25,
+      "stability/response_length_std": 27.932315826416016,
+      "stability/response_length_var": 780.2142944335938,
+      "stability/token_entropy_mean": 4.7060546875,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.010403833352029324,
+      "kl": 0.07130970805883408,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52957824.78527607,
+      "logits/rejected": -48375214.47133758,
+      "logps/chosen": -376.07349980828224,
+      "logps/rejected": -211.2522641321656,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999971091747284,
+      "loss/total_with_kl": 0.5000388622283936,
+      "metrics/advantage_var": 0.08360155671834946,
+      "regularization/advantage_var": 0.08360155671834946,
+      "regularization/kl": 4.180078030913137e-05,
+      "rewards/chosen": 7.232671683551343e-05,
+      "rewards/margins": 2.6380315060250514e-05,
+      "rewards/rejected": 4.594640177526292e-05,
+      "stability/repetition_rate_mean": 0.251953125,
+      "stability/response_length_mean": 40.375,
+      "stability/response_length_std": 32.61874008178711,
+      "stability/response_length_var": 1063.982177734375,
+      "stability/token_entropy_mean": 3.263671875,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.01109651941806078,
+      "kl": 0.07014961540699005,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54954995.358024694,
+      "logits/rejected": -49891590.48101266,
+      "logps/chosen": -311.9007040895062,
+      "logps/rejected": -197.01834454113924,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999902844429016,
+      "loss/total_with_kl": 0.5000271201133728,
+      "metrics/advantage_var": 0.07348966598510742,
+      "regularization/advantage_var": 0.07348966598510742,
+      "regularization/kl": 3.674483377835713e-05,
+      "rewards/chosen": 0.00010068489261615424,
+      "rewards/margins": 7.98731516469762e-05,
+      "rewards/rejected": 2.081174096917804e-05,
+      "stability/repetition_rate_mean": 0.18998579680919647,
+      "stability/response_length_mean": 22.25,
+      "stability/response_length_std": 27.78874397277832,
+      "stability/response_length_var": 772.2142944335938,
+      "stability/token_entropy_mean": 3.8359375,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.010649079456925392,
+      "kl": 0.06624338775873184,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53431853.98802395,
+      "logits/rejected": -50099728.732026145,
+      "logps/chosen": -392.27919161676647,
+      "logps/rejected": -211.17078993055554,
+      "loss": 0.5,
+      "loss/base_kto": 0.49998751282691956,
+      "loss/total_with_kl": 0.50002521276474,
+      "metrics/advantage_var": 0.07544363290071487,
+      "regularization/advantage_var": 0.07544363290071487,
+      "regularization/kl": 3.772181662498042e-05,
+      "rewards/chosen": 0.0001261207068751672,
+      "rewards/margins": 9.833026342690258e-05,
+      "rewards/rejected": 2.7790443448264613e-05,
+      "stability/repetition_rate_mean": 0.1106770858168602,
+      "stability/response_length_mean": 11.5,
+      "stability/response_length_std": 21.758413314819336,
+      "stability/response_length_var": 473.4285583496094,
+      "stability/token_entropy_mean": 5.3671875,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.01244101021438837,
+      "kl": 0.058560874313116074,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60543167.58709677,
+      "logits/rejected": -54317372.50909091,
+      "logps/chosen": -392.6783266129032,
+      "logps/rejected": -197.91171875,
+      "loss": 0.5,
+      "loss/base_kto": 0.49998798966407776,
+      "loss/total_with_kl": 0.5000260472297668,
+      "metrics/advantage_var": 0.07610367983579636,
+      "regularization/advantage_var": 0.07610367983579636,
+      "regularization/kl": 3.805184314842336e-05,
+      "rewards/chosen": 0.00010660523849148904,
+      "rewards/margins": 9.897794790260078e-05,
+      "rewards/rejected": 7.627290588888255e-06,
+      "stability/repetition_rate_mean": 0.3782551884651184,
+      "stability/response_length_mean": 38.875,
+      "stability/response_length_std": 30.786998748779297,
+      "stability/response_length_var": 947.8392944335938,
+      "stability/token_entropy_mean": 4.490234375,
+      "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-beta0.001/checkpoint-196/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e74897fa70a3117fd3b115a078e94a832dd2526a53fc52afc283e411a44afb4
+size 8017

mistral7b_kto_kl-beta0.001/checkpoint-196/zero_to_fp32.py

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

mistral7b_kto_kl-beta0.001/special_tokens_map.json

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

mistral7b_kto_kl-beta0.001/stability_metrics.jsonl

@@ -0,0 +1,38 @@
+{"stability/response_length_mean": 46.125, "stability/response_length_std": 27.231481552124023, "stability/response_length_var": 741.5535888671875, "stability/token_entropy_mean": 3.26171875, "stability/repetition_rate_mean": 0.3668472170829773, "step": 10}
+{"stability/response_length_mean": 28.125, "stability/response_length_std": 30.243947982788086, "stability/response_length_var": 914.6964111328125, "stability/token_entropy_mean": 3.7587890625, "stability/repetition_rate_mean": 0.27336087822914124, "step": 15}
+{"stability/response_length_mean": 42.125, "stability/response_length_std": 27.01553726196289, "stability/response_length_var": 729.8392944335938, "stability/token_entropy_mean": 4.4296875, "stability/repetition_rate_mean": 0.1943204402923584, "step": 20}
+{"stability/response_length_mean": 43.375, "stability/response_length_std": 28.54538917541504, "stability/response_length_var": 814.8392944335938, "stability/token_entropy_mean": 4.7890625, "stability/repetition_rate_mean": 0.17997129261493683, "step": 25}
+{"stability/response_length_mean": 48.375, "stability/response_length_std": 28.93311309814453, "stability/response_length_var": 837.125, "stability/token_entropy_mean": 4.513671875, "stability/repetition_rate_mean": 0.1953125, "step": 30}
+{"stability/response_length_mean": 53.375, "stability/response_length_std": 21.14533042907715, "stability/response_length_var": 447.125, "stability/token_entropy_mean": 5.330078125, "stability/repetition_rate_mean": 0.35856893658638, "step": 35}
+{"stability/response_length_mean": 33.125, "stability/response_length_std": 33.01704025268555, "stability/response_length_var": 1090.125, "stability/token_entropy_mean": 4.9609375, "stability/repetition_rate_mean": 0.142578125, "step": 40}
+{"stability/response_length_mean": 28.875, "stability/response_length_std": 29.87324333190918, "stability/response_length_var": 892.4107055664062, "stability/token_entropy_mean": 4.775390625, "stability/repetition_rate_mean": 0.1013549193739891, "step": 45}
+{"stability/response_length_mean": 53.375, "stability/response_length_std": 21.480472564697266, "stability/response_length_var": 461.41070556640625, "stability/token_entropy_mean": 3.857421875, "stability/repetition_rate_mean": 0.3831075429916382, "step": 50}
+{"stability/response_length_mean": 22.375, "stability/response_length_std": 30.236862182617188, "stability/response_length_var": 914.2678833007812, "stability/token_entropy_mean": 4.333984375, "stability/repetition_rate_mean": 0.11199301481246948, "step": 55}
+{"stability/response_length_mean": 53.625, "stability/response_length_std": 22.11617088317871, "stability/response_length_var": 489.125, "stability/token_entropy_mean": 5.255859375, "stability/repetition_rate_mean": 0.23800872266292572, "step": 60}
+{"stability/response_length_mean": 21.5, "stability/response_length_std": 26.635099411010742, "stability/response_length_var": 709.4285888671875, "stability/token_entropy_mean": 5.392578125, "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": 5.6875, "stability/repetition_rate_mean": 0.171875, "step": 70}
+{"stability/response_length_mean": 24.25, "stability/response_length_std": 30.093900680541992, "stability/response_length_var": 905.6428833007812, "stability/token_entropy_mean": 2.78125, "stability/repetition_rate_mean": 0.14558292925357819, "step": 75}
+{"stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.234375, "stability/repetition_rate_mean": 0.19088542461395264, "step": 80}
+{"stability/response_length_mean": 24.875, "stability/response_length_std": 27.424898147583008, "stability/response_length_var": 752.125, "stability/token_entropy_mean": 4.921875, "stability/repetition_rate_mean": 0.15742863714694977, "step": 85}
+{"stability/response_length_mean": 33.625, "stability/response_length_std": 29.442134857177734, "stability/response_length_var": 866.8392944335938, "stability/token_entropy_mean": 3.70263671875, "stability/repetition_rate_mean": 0.2192234843969345, "step": 90}
+{"stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.74609375, "stability/repetition_rate_mean": 0.19374233484268188, "step": 95}
+{"stability/response_length_mean": 38.0, "stability/response_length_std": 30.10457992553711, "stability/response_length_var": 906.2857055664062, "stability/token_entropy_mean": 4.23828125, "stability/repetition_rate_mean": 0.24177631735801697, "step": 100}
+{"stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.384765625, "stability/repetition_rate_mean": 0.2014508843421936, "step": 105}
+{"stability/response_length_mean": 28.875, "stability/response_length_std": 29.911476135253906, "stability/response_length_var": 894.6964111328125, "stability/token_entropy_mean": 4.412109375, "stability/repetition_rate_mean": 0.30853456258773804, "step": 110}
+{"stability/response_length_mean": 33.125, "stability/response_length_std": 33.0516357421875, "stability/response_length_var": 1092.4107666015625, "stability/token_entropy_mean": 4.142578125, "stability/repetition_rate_mean": 0.2506510615348816, "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": 0.17961646616458893, "step": 120}
+{"stability/response_length_mean": 15.25, "stability/response_length_std": 24.93276596069336, "stability/response_length_var": 621.6428833007812, "stability/token_entropy_mean": 3.625, "stability/repetition_rate_mean": 0.16328126192092896, "step": 125}
+{"stability/response_length_mean": 43.25, "stability/response_length_std": 29.620214462280273, "stability/response_length_var": 877.3571166992188, "stability/token_entropy_mean": 3.57421875, "stability/repetition_rate_mean": 0.22703124582767487, "step": 130}
+{"stability/response_length_mean": 29.0, "stability/response_length_std": 30.20406723022461, "stability/response_length_var": 912.2857055664062, "stability/token_entropy_mean": 3.427734375, "stability/repetition_rate_mean": 0.2230902761220932, "step": 135}
+{"stability/response_length_mean": 56.25, "stability/response_length_std": 21.920310974121094, "stability/response_length_var": 480.5, "stability/token_entropy_mean": 4.7548828125, "stability/repetition_rate_mean": 0.34765625, "step": 140}
+{"stability/response_length_mean": 28.875, "stability/response_length_std": 29.73423957824707, "stability/response_length_var": 884.125, "stability/token_entropy_mean": 5.306640625, "stability/repetition_rate_mean": 0.25424888730049133, "step": 145}
+{"stability/response_length_mean": 29.375, "stability/response_length_std": 29.193626403808594, "stability/response_length_var": 852.2678833007812, "stability/token_entropy_mean": 4.59765625, "stability/repetition_rate_mean": 0.20331649482250214, "step": 150}
+{"stability/response_length_mean": 53.125, "stability/response_length_std": 22.67432403564453, "stability/response_length_var": 514.125, "stability/token_entropy_mean": 4.5673828125, "stability/repetition_rate_mean": 0.28984373807907104, "step": 155}
+{"stability/response_length_mean": 26.25, "stability/response_length_std": 31.535694122314453, "stability/response_length_var": 994.5, "stability/token_entropy_mean": 3.828125, "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": 4.83984375, "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": 3.31884765625, "stability/repetition_rate_mean": 0.20866228640079498, "step": 170}
+{"stability/response_length_mean": 32.25, "stability/response_length_std": 27.932315826416016, "stability/response_length_var": 780.2142944335938, "stability/token_entropy_mean": 4.7060546875, "stability/repetition_rate_mean": 0.2936941981315613, "step": 175}
+{"stability/response_length_mean": 40.375, "stability/response_length_std": 32.61874008178711, "stability/response_length_var": 1063.982177734375, "stability/token_entropy_mean": 3.263671875, "stability/repetition_rate_mean": 0.251953125, "step": 180}
+{"stability/response_length_mean": 22.25, "stability/response_length_std": 27.78874397277832, "stability/response_length_var": 772.2142944335938, "stability/token_entropy_mean": 3.8359375, "stability/repetition_rate_mean": 0.18998579680919647, "step": 185}
+{"stability/response_length_mean": 11.5, "stability/response_length_std": 21.758413314819336, "stability/response_length_var": 473.4285583496094, "stability/token_entropy_mean": 5.3671875, "stability/repetition_rate_mean": 0.1106770858168602, "step": 190}
+{"stability/response_length_mean": 38.875, "stability/response_length_std": 30.786998748779297, "stability/response_length_var": 947.8392944335938, "stability/token_entropy_mean": 4.490234375, "stability/repetition_rate_mean": 0.3782551884651184, "step": 195}

mistral7b_kto_kl-beta0.001/tokenizer.model

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

mistral7b_kto_kl-beta0.001/tokenizer_config.json

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

mistral7b_kto_kl-beta0.001/train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.5000301818458401,
+    "train_runtime": 3151.2022,
+    "train_samples_per_second": 31.734,
+    "train_steps_per_second": 0.062
+}

mistral7b_kto_kl-beta0.001/trainer_log.jsonl

@@ -0,0 +1,40 @@
+{"current_steps": 5, "total_steps": 196, "loss": 0.5, "lr": 4e-08, "epoch": 0.025591810620601407, "percentage": 2.55, "elapsed_time": "0:01:18", "remaining_time": "0:49:47", "rewards/chosen": 2.4742100657405037e-05, "logps/chosen": -439.0208675986842, "logits/chosen": -59716823.578947365, "rewards/rejected": 5.043926648795605e-06, "logps/rejected": -198.4043433779762, "logits/rejected": -50418389.333333336, "rewards/margins": 1.9698174008609432e-05, "regularization/kl": 2.4925377147155814e-05, "regularization/advantage_var": 0.04985074698925018, "kl": 0.009334363043308258}
+{"current_steps": 10, "total_steps": 196, "loss": 0.5, "lr": 9e-08, "epoch": 0.05118362124120281, "percentage": 5.1, "elapsed_time": "0:02:38", "remaining_time": "0:49:15", "rewards/chosen": -2.0726006357120228e-05, "logps/chosen": -429.95926339285717, "logits/chosen": -59296382.33766234, "rewards/rejected": 1.9716718851263263e-05, "logps/rejected": -192.84822100903614, "logits/rejected": -50010654.84337349, "rewards/margins": -4.044272520838349e-05, "regularization/kl": 3.242322782170959e-05, "regularization/advantage_var": 0.0648464560508728, "kl": 0.021588662639260292, "stability/response_length_mean": 46.125, "stability/response_length_std": 27.231481552124023, "stability/response_length_var": 741.5535888671875, "stability/token_entropy_mean": 3.26171875, "stability/repetition_rate_mean": 0.3668472170829773}
+{"current_steps": 15, "total_steps": 196, "loss": 0.5, "lr": 1.4e-07, "epoch": 0.07677543186180422, "percentage": 7.65, "elapsed_time": "0:03:58", "remaining_time": "0:47:59", "rewards/chosen": 2.561494369398464e-05, "logps/chosen": -404.10520833333334, "logits/chosen": -56242951.75757576, "rewards/rejected": 6.382217660786644e-06, "logps/rejected": -200.6876764112903, "logits/rejected": -51922435.30322581, "rewards/margins": 1.9232726033197994e-05, "regularization/kl": 3.2226329494733363e-05, "regularization/advantage_var": 0.0644526481628418, "kl": 0.016281843185424805, "stability/response_length_mean": 28.125, "stability/response_length_std": 30.243947982788086, "stability/response_length_var": 914.6964111328125, "stability/token_entropy_mean": 3.7587890625, "stability/repetition_rate_mean": 0.27336087822914124}
+{"current_steps": 20, "total_steps": 196, "loss": 0.5, "lr": 1.8999999999999998e-07, "epoch": 0.10236724248240563, "percentage": 10.2, "elapsed_time": "0:05:20", "remaining_time": "0:46:56", "rewards/chosen": 7.749405631329864e-06, "logps/chosen": -343.68896484375, "logits/chosen": -51676390.4, "rewards/rejected": 2.675029681995511e-05, "logps/rejected": -225.6365966796875, "logits/rejected": -48065916.8, "rewards/margins": -1.9000891188625244e-05, "regularization/kl": 3.5532844776753336e-05, "regularization/advantage_var": 0.07106568664312363, "kl": 0.02222653105854988, "stability/response_length_mean": 42.125, "stability/response_length_std": 27.01553726196289, "stability/response_length_var": 729.8392944335938, "stability/token_entropy_mean": 4.4296875, "stability/repetition_rate_mean": 0.1943204402923584}
+{"current_steps": 25, "total_steps": 196, "loss": 0.5, "lr": 2.4e-07, "epoch": 0.12795905310300704, "percentage": 12.76, "elapsed_time": "0:06:39", "remaining_time": "0:45:34", "rewards/chosen": 1.283879873567614e-05, "logps/chosen": -378.4786099137931, "logits/chosen": -56473031.50344828, "rewards/rejected": 1.958490482398442e-05, "logps/rejected": -202.77325892857144, "logits/rejected": -52018123.337142855, "rewards/margins": -6.746106088308278e-06, "regularization/kl": 3.091646431130357e-05, "regularization/advantage_var": 0.061832915991544724, "kl": 0.013138818554580212, "stability/response_length_mean": 43.375, "stability/response_length_std": 28.54538917541504, "stability/response_length_var": 814.8392944335938, "stability/token_entropy_mean": 4.7890625, "stability/repetition_rate_mean": 0.17997129261493683}
+{"current_steps": 30, "total_steps": 196, "loss": 0.5, "lr": 2.9e-07, "epoch": 0.15355086372360843, "percentage": 15.31, "elapsed_time": "0:07:59", "remaining_time": "0:44:14", "rewards/chosen": -6.928566480743005e-06, "logps/chosen": -468.58819211409394, "logits/chosen": -57743002.63087248, "rewards/rejected": 1.4772561068336168e-05, "logps/rejected": -198.2330957602339, "logits/rejected": -51520362.29239766, "rewards/margins": -2.1701127549079173e-05, "regularization/kl": 3.220155122107826e-05, "regularization/advantage_var": 0.06440309435129166, "kl": 0.01257875096052885, "stability/response_length_mean": 48.375, "stability/response_length_std": 28.93311309814453, "stability/response_length_var": 837.125, "stability/token_entropy_mean": 4.513671875, "stability/repetition_rate_mean": 0.1953125}
+{"current_steps": 35, "total_steps": 196, "loss": 0.5, "lr": 3.4000000000000003e-07, "epoch": 0.17914267434420986, "percentage": 17.86, "elapsed_time": "0:09:20", "remaining_time": "0:43:00", "rewards/chosen": 3.946519492170479e-05, "logps/chosen": -385.8970530063291, "logits/chosen": -57641219.24050633, "rewards/rejected": -1.0337782621291684e-05, "logps/rejected": -195.1368152006173, "logits/rejected": -53900932.74074074, "rewards/margins": 4.9802977542996476e-05, "regularization/kl": 3.599749834393151e-05, "regularization/advantage_var": 0.07199499756097794, "kl": 0.022648466750979424, "stability/response_length_mean": 53.375, "stability/response_length_std": 21.14533042907715, "stability/response_length_var": 447.125, "stability/token_entropy_mean": 5.330078125, "stability/repetition_rate_mean": 0.35856893658638}
+{"current_steps": 40, "total_steps": 196, "loss": 0.5, "lr": 3.8999999999999997e-07, "epoch": 0.20473448496481125, "percentage": 20.41, "elapsed_time": "0:10:41", "remaining_time": "0:41:42", "rewards/chosen": 1.1162247392348946e-05, "logps/chosen": -421.89150390625, "logits/chosen": -56646547.2, "rewards/rejected": 4.188578168395907e-05, "logps/rejected": -219.2673583984375, "logits/rejected": -50243718.4, "rewards/margins": -3.0723534291610124e-05, "regularization/kl": 3.984443537774496e-05, "regularization/advantage_var": 0.07968886941671371, "kl": 0.017865750938653946, "stability/response_length_mean": 33.125, "stability/response_length_std": 33.01704025268555, "stability/response_length_var": 1090.125, "stability/token_entropy_mean": 4.9609375, "stability/repetition_rate_mean": 0.142578125}
+{"current_steps": 45, "total_steps": 196, "loss": 0.5, "lr": 4.3999999999999997e-07, "epoch": 0.23032629558541268, "percentage": 22.96, "elapsed_time": "0:12:02", "remaining_time": "0:40:23", "rewards/chosen": 4.6062358223751884e-05, "logps/chosen": -352.28240352348996, "logits/chosen": -52715269.15436242, "rewards/rejected": 5.673122597716705e-05, "logps/rejected": -243.81743421052633, "logits/rejected": -50790807.204678364, "rewards/margins": -1.0668867753415165e-05, "regularization/kl": 3.844457751256414e-05, "regularization/advantage_var": 0.07688915729522705, "kl": 0.03356611356139183, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.87324333190918, "stability/response_length_var": 892.4107055664062, "stability/token_entropy_mean": 4.775390625, "stability/repetition_rate_mean": 0.1013549193739891}
+{"current_steps": 50, "total_steps": 196, "loss": 0.5, "lr": 4.9e-07, "epoch": 0.2559181062060141, "percentage": 25.51, "elapsed_time": "0:13:22", "remaining_time": "0:39:02", "rewards/chosen": 3.098852732437569e-05, "logps/chosen": -375.1681260486577, "logits/chosen": -54045579.167785235, "rewards/rejected": 3.160719830564588e-05, "logps/rejected": -180.20900721856725, "logits/rejected": -48300567.953216374, "rewards/margins": -6.186709812701927e-07, "regularization/kl": 2.388452776358463e-05, "regularization/advantage_var": 0.04776905104517937, "kl": 0.032784707844257355, "stability/response_length_mean": 53.375, "stability/response_length_std": 21.480472564697266, "stability/response_length_var": 461.41070556640625, "stability/token_entropy_mean": 3.857421875, "stability/repetition_rate_mean": 0.3831075429916382}
+{"current_steps": 55, "total_steps": 196, "loss": 0.5, "lr": 5.4e-07, "epoch": 0.28150991682661547, "percentage": 28.06, "elapsed_time": "0:14:41", "remaining_time": "0:37:40", "rewards/chosen": -5.357490956598187e-07, "logps/chosen": -344.5125679347826, "logits/chosen": -56956018.48447205, "rewards/rejected": 1.2828754671152282e-05, "logps/rejected": -230.43519064465409, "logits/rejected": -49576486.641509436, "rewards/margins": -1.33645037668121e-05, "regularization/kl": 2.796225635393057e-05, "regularization/advantage_var": 0.055924512445926666, "kl": 0.03365620970726013, "stability/response_length_mean": 22.375, "stability/response_length_std": 30.236862182617188, "stability/response_length_var": 914.2678833007812, "stability/token_entropy_mean": 4.333984375, "stability/repetition_rate_mean": 0.11199301481246948}
+{"current_steps": 60, "total_steps": 196, "loss": 0.5, "lr": 5.9e-07, "epoch": 0.30710172744721687, "percentage": 30.61, "elapsed_time": "0:16:03", "remaining_time": "0:36:23", "rewards/chosen": 8.442694264830965e-05, "logps/chosen": -359.6363636363636, "logits/chosen": -52675459.878787875, "rewards/rejected": 6.060358737745593e-05, "logps/rejected": -178.1663054435484, "logits/rejected": -47832952.56774194, "rewards/margins": 2.3823355270853723e-05, "regularization/kl": 3.138520332868211e-05, "regularization/advantage_var": 0.06277040392160416, "kl": 0.024347607046365738, "stability/response_length_mean": 53.625, "stability/response_length_std": 22.11617088317871, "stability/response_length_var": 489.125, "stability/token_entropy_mean": 5.255859375, "stability/repetition_rate_mean": 0.23800872266292572}
+{"current_steps": 65, "total_steps": 196, "loss": 0.5, "lr": 6.4e-07, "epoch": 0.3326935380678183, "percentage": 33.16, "elapsed_time": "0:17:23", "remaining_time": "0:35:03", "rewards/chosen": 4.530613941531027e-05, "logps/chosen": -381.90148689516127, "logits/chosen": -57575225.80645161, "rewards/rejected": 2.488098248387828e-05, "logps/rejected": -183.55746922348484, "logits/rejected": -47288897.163636364, "rewards/margins": 2.0425156931431993e-05, "regularization/kl": 3.388536424608901e-05, "regularization/advantage_var": 0.06777071952819824, "kl": 0.03643302991986275, "stability/response_length_mean": 21.5, "stability/response_length_std": 26.635099411010742, "stability/response_length_var": 709.4285888671875, "stability/token_entropy_mean": 5.392578125, "stability/repetition_rate_mean": 0.1857219785451889}
+{"current_steps": 70, "total_steps": 196, "loss": 0.5, "lr": 6.9e-07, "epoch": 0.3582853486884197, "percentage": 35.71, "elapsed_time": "0:18:43", "remaining_time": "0:33:42", "rewards/chosen": 6.355317609925424e-05, "logps/chosen": -379.0649193548387, "logits/chosen": -52791732.02580645, "rewards/rejected": 7.167139062375734e-05, "logps/rejected": -220.3504734848485, "logits/rejected": -49063154.03636364, "rewards/margins": -8.118214524503094e-06, "regularization/kl": 3.2315518183168024e-05, "regularization/advantage_var": 0.06463103741407394, "kl": 0.04110654443502426, "stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.6875, "stability/repetition_rate_mean": 0.171875}
+{"current_steps": 75, "total_steps": 196, "loss": 0.5, "lr": 7.4e-07, "epoch": 0.3838771593090211, "percentage": 38.27, "elapsed_time": "0:20:04", "remaining_time": "0:32:23", "rewards/chosen": 7.1432428887567e-05, "logps/chosen": -314.1993771114865, "logits/chosen": -49518003.89189189, "rewards/rejected": 1.4796576917518016e-06, "logps/rejected": -228.88335755813952, "logits/rejected": -47914368.0, "rewards/margins": 6.99527711958152e-05, "regularization/kl": 3.3921172871487215e-05, "regularization/advantage_var": 0.06784234941005707, "kl": 0.03520666062831879, "stability/response_length_mean": 24.25, "stability/response_length_std": 30.093900680541992, "stability/response_length_var": 905.6428833007812, "stability/token_entropy_mean": 2.78125, "stability/repetition_rate_mean": 0.14558292925357819}
+{"current_steps": 80, "total_steps": 196, "loss": 0.5, "lr": 7.9e-07, "epoch": 0.4094689699296225, "percentage": 40.82, "elapsed_time": "0:21:25", "remaining_time": "0:31:04", "rewards/chosen": 3.66592804426389e-05, "logps/chosen": -403.9934292218543, "logits/chosen": -58671863.52317881, "rewards/rejected": 2.8461209614072326e-05, "logps/rejected": -201.02734375, "logits/rejected": -53488839.95266272, "rewards/margins": 8.198070828566575e-06, "regularization/kl": 3.245426341891289e-05, "regularization/advantage_var": 0.06490852683782578, "kl": 0.051506686955690384, "stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.234375, "stability/repetition_rate_mean": 0.19088542461395264}
+{"current_steps": 85, "total_steps": 196, "loss": 0.5, "lr": 8.399999999999999e-07, "epoch": 0.4350607805502239, "percentage": 43.37, "elapsed_time": "0:22:45", "remaining_time": "0:29:43", "rewards/chosen": 3.3219540191273537e-05, "logps/chosen": -358.2992943548387, "logits/chosen": -51497151.58709677, "rewards/rejected": -8.181050758470188e-06, "logps/rejected": -230.4330018939394, "logits/rejected": -47171549.86666667, "rewards/margins": 4.1400590949743726e-05, "regularization/kl": 3.753290729946457e-05, "regularization/advantage_var": 0.07506581395864487, "kl": 0.03408239036798477, "stability/response_length_mean": 24.875, "stability/response_length_std": 27.424898147583008, "stability/response_length_var": 752.125, "stability/token_entropy_mean": 4.921875, "stability/repetition_rate_mean": 0.15742863714694977}
+{"current_steps": 90, "total_steps": 196, "loss": 0.5, "lr": 8.9e-07, "epoch": 0.46065259117082535, "percentage": 45.92, "elapsed_time": "0:24:06", "remaining_time": "0:28:24", "rewards/chosen": 2.6056644482872424e-05, "logps/chosen": -380.76332131410254, "logits/chosen": -58462437.743589744, "rewards/rejected": 2.5743371579886937e-05, "logps/rejected": -187.201064691311, "logits/rejected": -50612860.87804878, "rewards/margins": 3.132729029854866e-07, "regularization/kl": 3.140298213111237e-05, "regularization/advantage_var": 0.06280595809221268, "kl": 0.044009942561388016, "stability/response_length_mean": 33.625, "stability/response_length_std": 29.442134857177734, "stability/response_length_var": 866.8392944335938, "stability/token_entropy_mean": 3.70263671875, "stability/repetition_rate_mean": 0.2192234843969345}
+{"current_steps": 95, "total_steps": 196, "loss": 0.5, "lr": 9.399999999999999e-07, "epoch": 0.48624440179142675, "percentage": 48.47, "elapsed_time": "0:25:26", "remaining_time": "0:27:03", "rewards/chosen": 4.625187514797796e-05, "logps/chosen": -314.9471172863924, "logits/chosen": -50442868.658227846, "rewards/rejected": 1.1683969160564887e-05, "logps/rejected": -207.13204089506172, "logits/rejected": -49033930.27160494, "rewards/margins": 3.456790598741307e-05, "regularization/kl": 3.48176654370036e-05, "regularization/advantage_var": 0.06963532418012619, "kl": 0.043333958834409714, "stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.74609375, "stability/repetition_rate_mean": 0.19374233484268188}
+{"current_steps": 100, "total_steps": 196, "loss": 0.5, "lr": 9.9e-07, "epoch": 0.5118362124120281, "percentage": 51.02, "elapsed_time": "0:26:47", "remaining_time": "0:25:42", "rewards/chosen": 8.41418545491792e-05, "logps/chosen": -359.7860272381757, "logits/chosen": -54759797.62162162, "rewards/rejected": 1.6562402421651884e-05, "logps/rejected": -216.5623410247093, "logits/rejected": -50968784.37209302, "rewards/margins": 6.757945212752732e-05, "regularization/kl": 4.5903540012659505e-05, "regularization/advantage_var": 0.09180708229541779, "kl": 0.04752064868807793, "stability/response_length_mean": 38.0, "stability/response_length_std": 30.10457992553711, "stability/response_length_var": 906.2857055664062, "stability/token_entropy_mean": 4.23828125, "stability/repetition_rate_mean": 0.24177631735801697}
+{"current_steps": 105, "total_steps": 196, "loss": 0.5, "lr": 9.999815554351224e-07, "epoch": 0.5374280230326296, "percentage": 53.57, "elapsed_time": "0:28:07", "remaining_time": "0:24:22", "rewards/chosen": 8.220696844250323e-05, "logps/chosen": -399.820265436747, "logits/chosen": -59449306.98795181, "rewards/rejected": 3.0769068481666704e-05, "logps/rejected": -209.97304941152598, "logits/rejected": -53884203.22077922, "rewards/margins": 5.143789996083652e-05, "regularization/kl": 3.658089190139435e-05, "regularization/advantage_var": 0.073161780834198, "kl": 0.04482289031147957, "stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.384765625, "stability/repetition_rate_mean": 0.2014508843421936}
+{"current_steps": 110, "total_steps": 196, "loss": 0.5, "lr": 9.999066267225444e-07, "epoch": 0.5630198336532309, "percentage": 56.12, "elapsed_time": "0:29:29", "remaining_time": "0:23:03", "rewards/chosen": 2.6093964206885144e-05, "logps/chosen": -378.5371799698795, "logits/chosen": -57465559.90361446, "rewards/rejected": 2.8886548062036564e-05, "logps/rejected": -206.05924056412337, "logits/rejected": -53690846.753246754, "rewards/margins": -2.7925838551514204e-06, "regularization/kl": 3.773330900003202e-05, "regularization/advantage_var": 0.07546661794185638, "kl": 0.050605691969394684, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.911476135253906, "stability/response_length_var": 894.6964111328125, "stability/token_entropy_mean": 4.412109375, "stability/repetition_rate_mean": 0.30853456258773804}
+{"current_steps": 115, "total_steps": 196, "loss": 0.5, "lr": 9.997740697079592e-07, "epoch": 0.5886116442738324, "percentage": 58.67, "elapsed_time": "0:30:50", "remaining_time": "0:21:43", "rewards/chosen": 8.311558844922464e-05, "logps/chosen": -364.50944422468353, "logits/chosen": -60689090.43037975, "rewards/rejected": 2.4287533129990837e-05, "logps/rejected": -201.17226080246914, "logits/rejected": -53511787.45679013, "rewards/margins": 5.8828055319233806e-05, "regularization/kl": 4.031229764223099e-05, "regularization/advantage_var": 0.08062459528446198, "kl": 0.04474036023020744, "stability/response_length_mean": 33.125, "stability/response_length_std": 33.0516357421875, "stability/response_length_var": 1092.4107666015625, "stability/token_entropy_mean": 4.142578125, "stability/repetition_rate_mean": 0.2506510615348816}
+{"current_steps": 120, "total_steps": 196, "loss": 0.5, "lr": 9.995838996722914e-07, "epoch": 0.6142034548944337, "percentage": 61.22, "elapsed_time": "0:32:09", "remaining_time": "0:20:22", "rewards/chosen": 4.753693207806232e-05, "logps/chosen": -384.19738051470586, "logits/chosen": -57836122.35294118, "rewards/rejected": 3.6157199187193086e-05, "logps/rejected": -212.1802020958084, "logits/rejected": -54507458.68263473, "rewards/margins": 1.1379732890869236e-05, "regularization/kl": 4.145712227909826e-05, "regularization/advantage_var": 0.08291424810886383, "kl": 0.03717491030693054, "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": 0.17961646616458893}
+{"current_steps": 125, "total_steps": 196, "loss": 0.5, "lr": 9.993361385379874e-07, "epoch": 0.6397952655150352, "percentage": 63.78, "elapsed_time": "0:33:30", "remaining_time": "0:19:02", "rewards/chosen": 2.899690059154001e-05, "logps/chosen": -417.35030500856163, "logits/chosen": -60827171.06849315, "rewards/rejected": 3.2779688400955034e-05, "logps/rejected": -216.33441091954023, "logits/rejected": -55819781.88505747, "rewards/margins": -3.782787809415023e-06, "regularization/kl": 3.5848712286679074e-05, "regularization/advantage_var": 0.0716974139213562, "kl": 0.03611461818218231, "stability/response_length_mean": 15.25, "stability/response_length_std": 24.93276596069336, "stability/response_length_var": 621.6428833007812, "stability/token_entropy_mean": 3.625, "stability/repetition_rate_mean": 0.16328126192092896}
+{"current_steps": 130, "total_steps": 196, "loss": 0.5, "lr": 9.99030814866488e-07, "epoch": 0.6653870761356366, "percentage": 66.33, "elapsed_time": "0:34:51", "remaining_time": "0:17:41", "rewards/chosen": 5.7653367759719974e-05, "logps/chosen": -392.6003024193548, "logits/chosen": -57520075.1483871, "rewards/rejected": 1.7249640641790448e-05, "logps/rejected": -183.9016690340909, "logits/rejected": -53990958.54545455, "rewards/margins": 4.0403727117929525e-05, "regularization/kl": 3.92844776797574e-05, "regularization/advantage_var": 0.0785689577460289, "kl": 0.04913714528083801, "stability/response_length_mean": 43.25, "stability/response_length_std": 29.620214462280273, "stability/response_length_var": 877.3571166992188, "stability/token_entropy_mean": 3.57421875, "stability/repetition_rate_mean": 0.22703124582767487}
+{"current_steps": 135, "total_steps": 196, "loss": 0.5, "lr": 9.986679638549358e-07, "epoch": 0.690978886756238, "percentage": 68.88, "elapsed_time": "0:36:11", "remaining_time": "0:16:20", "rewards/chosen": 9.945813177919095e-05, "logps/chosen": -401.71668903374234, "logits/chosen": -59993232.49079755, "rewards/rejected": 3.1335747972795155e-05, "logps/rejected": -247.83942078025478, "logits/rejected": -55193215.18471338, "rewards/margins": 6.812238380639579e-05, "regularization/kl": 4.423669815878384e-05, "regularization/advantage_var": 0.08847339451313019, "kl": 0.05869415029883385, "stability/response_length_mean": 29.0, "stability/response_length_std": 30.20406723022461, "stability/response_length_var": 912.2857055664062, "stability/token_entropy_mean": 3.427734375, "stability/repetition_rate_mean": 0.2230902761220932}
+{"current_steps": 140, "total_steps": 196, "loss": 0.5, "lr": 9.982476273321175e-07, "epoch": 0.7165706973768394, "percentage": 71.43, "elapsed_time": "0:37:31", "remaining_time": "0:15:00", "rewards/chosen": 8.873008756023464e-05, "logps/chosen": -367.1009943181818, "logits/chosen": -57256320.77575757, "rewards/rejected": 1.997527095579332e-05, "logps/rejected": -190.01449092741936, "logits/rejected": -50278221.62580645, "rewards/margins": 6.875481660444132e-05, "regularization/kl": 3.274825940025039e-05, "regularization/advantage_var": 0.0654965192079544, "kl": 0.0394296832382679, "stability/response_length_mean": 56.25, "stability/response_length_std": 21.920310974121094, "stability/response_length_var": 480.5, "stability/token_entropy_mean": 4.7548828125, "stability/repetition_rate_mean": 0.34765625}
+{"current_steps": 145, "total_steps": 196, "loss": 0.5, "lr": 9.977698537536417e-07, "epoch": 0.7421625079974408, "percentage": 73.98, "elapsed_time": "0:38:51", "remaining_time": "0:13:40", "rewards/chosen": 7.712891527328147e-05, "logps/chosen": -330.19470067771084, "logits/chosen": -51322105.8313253, "rewards/rejected": -1.8788757129245763e-05, "logps/rejected": -199.5067091112013, "logits/rejected": -51878599.48051948, "rewards/margins": 9.591767240252724e-05, "regularization/kl": 2.8509681214927696e-05, "regularization/advantage_var": 0.05701936408877373, "kl": 0.051040973514318466, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.73423957824707, "stability/response_length_var": 884.125, "stability/token_entropy_mean": 5.306640625, "stability/repetition_rate_mean": 0.25424888730049133}
+{"current_steps": 150, "total_steps": 196, "loss": 0.5, "lr": 9.972346981963546e-07, "epoch": 0.7677543186180422, "percentage": 76.53, "elapsed_time": "0:40:12", "remaining_time": "0:12:19", "rewards/chosen": 0.00013485204647569095, "logps/chosen": -397.7995863970588, "logits/chosen": -55602344.65882353, "rewards/rejected": 7.843268879999717e-07, "logps/rejected": -207.72515625, "logits/rejected": -50412625.92, "rewards/margins": 0.00013406771958769098, "regularization/kl": 4.1245522879762575e-05, "regularization/advantage_var": 0.08249103277921677, "kl": 0.056698884814977646, "stability/response_length_mean": 29.375, "stability/response_length_std": 29.193626403808594, "stability/response_length_var": 852.2678833007812, "stability/token_entropy_mean": 4.59765625, "stability/repetition_rate_mean": 0.20331649482250214}
+{"current_steps": 155, "total_steps": 196, "loss": 0.5, "lr": 9.966422223519885e-07, "epoch": 0.7933461292386437, "percentage": 79.08, "elapsed_time": "0:41:34", "remaining_time": "0:10:59", "rewards/chosen": 4.4789784314001306e-05, "logps/chosen": -360.94453125, "logits/chosen": -54248122.729411766, "rewards/rejected": -2.113558972875277e-06, "logps/rejected": -191.30770833333332, "logits/rejected": -49590394.88, "rewards/margins": 4.690334328687658e-05, "regularization/kl": 3.465908230282366e-05, "regularization/advantage_var": 0.06931816786527634, "kl": 0.050125040113925934, "stability/response_length_mean": 53.125, "stability/response_length_std": 22.67432403564453, "stability/response_length_var": 514.125, "stability/token_entropy_mean": 4.5673828125, "stability/repetition_rate_mean": 0.28984373807907104}
+{"current_steps": 160, "total_steps": 196, "loss": 0.5, "lr": 9.959924945200523e-07, "epoch": 0.818937939859245, "percentage": 81.63, "elapsed_time": "0:42:54", "remaining_time": "0:09:39", "rewards/chosen": 5.241769499012402e-05, "logps/chosen": -416.84291294642856, "logits/chosen": -55498038.85714286, "rewards/rejected": 2.925097942352295e-05, "logps/rejected": -186.8680998149671, "logits/rejected": -49266189.473684214, "rewards/margins": 2.316671556660107e-05, "regularization/kl": 3.936288339900784e-05, "regularization/advantage_var": 0.07872576266527176, "kl": 0.049432478845119476, "stability/response_length_mean": 26.25, "stability/response_length_std": 31.535694122314453, "stability/response_length_var": 994.5, "stability/token_entropy_mean": 3.828125, "stability/repetition_rate_mean": 0.291015625}
+{"current_steps": 165, "total_steps": 196, "loss": 0.5, "lr": 9.952855895999568e-07, "epoch": 0.8445297504798465, "percentage": 84.18, "elapsed_time": "0:44:15", "remaining_time": "0:08:18", "rewards/chosen": 4.114243255435735e-05, "logps/chosen": -370.8756061422414, "logits/chosen": -54674326.06896552, "rewards/rejected": 1.6166264038175754e-05, "logps/rejected": -212.5759578339041, "logits/rejected": -50117249.75342466, "rewards/margins": 2.4976168516181596e-05, "regularization/kl": 3.532640403136611e-05, "regularization/advantage_var": 0.0706528052687645, "kl": 0.04661773145198822, "stability/response_length_mean": 12.625, "stability/response_length_std": 21.460512161254883, "stability/response_length_var": 460.5535583496094, "stability/token_entropy_mean": 4.83984375, "stability/repetition_rate_mean": 0.20468750596046448}
+{"current_steps": 170, "total_steps": 196, "loss": 0.5, "lr": 9.9452158908238e-07, "epoch": 0.8701215611004478, "percentage": 86.73, "elapsed_time": "0:45:36", "remaining_time": "0:06:58", "rewards/chosen": 6.438495822735792e-05, "logps/chosen": -441.2774593195266, "logits/chosen": -59409420.1183432, "rewards/rejected": 1.0768145890228007e-05, "logps/rejected": -191.24706384519868, "logits/rejected": -52976277.19205298, "rewards/margins": 5.361681233712991e-05, "regularization/kl": 3.727779403561726e-05, "regularization/advantage_var": 0.07455558329820633, "kl": 0.062040429562330246, "stability/response_length_mean": 28.25, "stability/response_length_std": 30.844078063964844, "stability/response_length_var": 951.3571166992188, "stability/token_entropy_mean": 3.31884765625, "stability/repetition_rate_mean": 0.20866228640079498}
+{"current_steps": 175, "total_steps": 196, "loss": 0.5, "lr": 9.937005810398745e-07, "epoch": 0.8957133717210493, "percentage": 89.29, "elapsed_time": "0:46:56", "remaining_time": "0:05:38", "rewards/chosen": 0.00015270808734089495, "logps/chosen": -356.2844973644578, "logits/chosen": -56958445.4939759, "rewards/rejected": 5.0674538829593685e-06, "logps/rejected": -194.80245535714286, "logits/rejected": -48456025.766233765, "rewards/margins": 0.0001476406334579356, "regularization/kl": 4.260233981767669e-05, "regularization/advantage_var": 0.08520467579364777, "kl": 0.06263448297977448, "stability/response_length_mean": 32.25, "stability/response_length_std": 27.932315826416016, "stability/response_length_var": 780.2142944335938, "stability/token_entropy_mean": 4.7060546875, "stability/repetition_rate_mean": 0.2936941981315613}
+{"current_steps": 180, "total_steps": 196, "loss": 0.5, "lr": 9.928226601167138e-07, "epoch": 0.9213051823416507, "percentage": 91.84, "elapsed_time": "0:48:17", "remaining_time": "0:04:17", "rewards/chosen": 7.232671683551343e-05, "logps/chosen": -376.07349980828224, "logits/chosen": -52957824.78527607, "rewards/rejected": 4.594640177526292e-05, "logps/rejected": -211.2522641321656, "logits/rejected": -48375214.47133758, "rewards/margins": 2.6380315060250514e-05, "regularization/kl": 4.180078030913137e-05, "regularization/advantage_var": 0.08360155671834946, "kl": 0.07130970805883408, "stability/response_length_mean": 40.375, "stability/response_length_std": 32.61874008178711, "stability/response_length_var": 1063.982177734375, "stability/token_entropy_mean": 3.263671875, "stability/repetition_rate_mean": 0.251953125}
+{"current_steps": 185, "total_steps": 196, "loss": 0.5, "lr": 9.918879275179817e-07, "epoch": 0.946896992962252, "percentage": 94.39, "elapsed_time": "0:49:36", "remaining_time": "0:02:57", "rewards/chosen": 0.00010068489261615424, "logps/chosen": -311.9007040895062, "logits/chosen": -54954995.358024694, "rewards/rejected": 2.081174096917804e-05, "logps/rejected": -197.01834454113924, "logits/rejected": -49891590.48101266, "rewards/margins": 7.98731516469762e-05, "regularization/kl": 3.674483377835713e-05, "regularization/advantage_var": 0.07348966598510742, "kl": 0.07014961540699005, "stability/response_length_mean": 22.25, "stability/response_length_std": 27.78874397277832, "stability/response_length_var": 772.2142944335938, "stability/token_entropy_mean": 3.8359375, "stability/repetition_rate_mean": 0.18998579680919647}
+{"current_steps": 190, "total_steps": 196, "loss": 0.5, "lr": 9.90896490997906e-07, "epoch": 0.9724888035828535, "percentage": 96.94, "elapsed_time": "0:50:56", "remaining_time": "0:01:36", "rewards/chosen": 0.0001261207068751672, "logps/chosen": -392.27919161676647, "logits/chosen": -53431853.98802395, "rewards/rejected": 2.7790443448264613e-05, "logps/rejected": -211.17078993055554, "logits/rejected": -50099728.732026145, "rewards/margins": 9.833026342690258e-05, "regularization/kl": 3.772181662498042e-05, "regularization/advantage_var": 0.07544363290071487, "kl": 0.06624338775873184, "stability/response_length_mean": 11.5, "stability/response_length_std": 21.758413314819336, "stability/response_length_var": 473.4285583496094, "stability/token_entropy_mean": 5.3671875, "stability/repetition_rate_mean": 0.1106770858168602}
+{"current_steps": 195, "total_steps": 196, "loss": 0.5, "lr": 9.898484648474362e-07, "epoch": 0.9980806142034548, "percentage": 99.49, "elapsed_time": "0:52:17", "remaining_time": "0:00:16", "rewards/chosen": 0.00010660523849148904, "logps/chosen": -392.6783266129032, "logits/chosen": -60543167.58709677, "rewards/rejected": 7.627290588888255e-06, "logps/rejected": -197.91171875, "logits/rejected": -54317372.50909091, "rewards/margins": 9.897794790260078e-05, "regularization/kl": 3.805184314842336e-05, "regularization/advantage_var": 0.07610367983579636, "kl": 0.058560874313116074, "stability/response_length_mean": 38.875, "stability/response_length_std": 30.786998748779297, "stability/response_length_var": 947.8392944335938, "stability/token_entropy_mean": 4.490234375, "stability/repetition_rate_mean": 0.3782551884651184}
+{"current_steps": 196, "total_steps": 196, "epoch": 1.0, "percentage": 100.0, "elapsed_time": "0:52:31", "remaining_time": "0:00:00"}

mistral7b_kto_kl-beta0.001/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": 0.010321159847080708,
+      "kl": 0.009334363043308258,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59716823.578947365,
+      "logits/rejected": -50418389.333333336,
+      "logps/chosen": -439.0208675986842,
+      "logps/rejected": -198.4043433779762,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999978542327881,
+      "loss/total_with_kl": 0.5000227689743042,
+      "metrics/advantage_var": 0.04985074698925018,
+      "regularization/advantage_var": 0.04985074698925018,
+      "regularization/kl": 2.4925377147155814e-05,
+      "rewards/chosen": 2.4742100657405037e-05,
+      "rewards/margins": 1.9698174008609432e-05,
+      "rewards/rejected": 5.043926648795605e-06,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.01033781562000513,
+      "kl": 0.021588662639260292,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59296382.33766234,
+      "logits/rejected": -50010654.84337349,
+      "logps/chosen": -429.95926339285717,
+      "logps/rejected": -192.84822100903614,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000049471855164,
+      "loss/total_with_kl": 0.5000373721122742,
+      "metrics/advantage_var": 0.0648464560508728,
+      "regularization/advantage_var": 0.0648464560508728,
+      "regularization/kl": 3.242322782170959e-05,
+      "rewards/chosen": -2.0726006357120228e-05,
+      "rewards/margins": -4.044272520838349e-05,
+      "rewards/rejected": 1.9716718851263263e-05,
+      "stability/repetition_rate_mean": 0.3668472170829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.26171875,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.010668128728866577,
+      "kl": 0.016281843185424805,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56242951.75757576,
+      "logits/rejected": -51922435.30322581,
+      "logps/chosen": -404.10520833333334,
+      "logps/rejected": -200.6876764112903,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999719858169556,
+      "loss/total_with_kl": 0.5000293850898743,
+      "metrics/advantage_var": 0.0644526481628418,
+      "regularization/advantage_var": 0.0644526481628418,
+      "regularization/kl": 3.2226329494733363e-05,
+      "rewards/chosen": 2.561494369398464e-05,
+      "rewards/margins": 1.9232726033197994e-05,
+      "rewards/rejected": 6.382217660786644e-06,
+      "stability/repetition_rate_mean": 0.27336087822914124,
+      "stability/response_length_mean": 28.125,
+      "stability/response_length_std": 30.243947982788086,
+      "stability/response_length_var": 914.6964111328125,
+      "stability/token_entropy_mean": 3.7587890625,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.011172862723469734,
+      "kl": 0.02222653105854988,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51676390.4,
+      "logits/rejected": -48065916.8,
+      "logps/chosen": -343.68896484375,
+      "logps/rejected": -225.6365966796875,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000028014183044,
+      "loss/total_with_kl": 0.5000383257865906,
+      "metrics/advantage_var": 0.07106568664312363,
+      "regularization/advantage_var": 0.07106568664312363,
+      "regularization/kl": 3.5532844776753336e-05,
+      "rewards/chosen": 7.749405631329864e-06,
+      "rewards/margins": -1.9000891188625244e-05,
+      "rewards/rejected": 2.675029681995511e-05,
+      "stability/repetition_rate_mean": 0.1943204402923584,
+      "stability/response_length_mean": 42.125,
+      "stability/response_length_std": 27.01553726196289,
+      "stability/response_length_var": 729.8392944335938,
+      "stability/token_entropy_mean": 4.4296875,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.009160108864307404,
+      "kl": 0.013138818554580212,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56473031.50344828,
+      "logits/rejected": -52018123.337142855,
+      "logps/chosen": -378.4786099137931,
+      "logps/rejected": -202.77325892857144,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000008940696716,
+      "loss/total_with_kl": 0.5000317096710205,
+      "metrics/advantage_var": 0.061832915991544724,
+      "regularization/advantage_var": 0.061832915991544724,
+      "regularization/kl": 3.091646431130357e-05,
+      "rewards/chosen": 1.283879873567614e-05,
+      "rewards/margins": -6.746106088308278e-06,
+      "rewards/rejected": 1.958490482398442e-05,
+      "stability/repetition_rate_mean": 0.17997129261493683,
+      "stability/response_length_mean": 43.375,
+      "stability/response_length_std": 28.54538917541504,
+      "stability/response_length_var": 814.8392944335938,
+      "stability/token_entropy_mean": 4.7890625,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.010463796555995941,
+      "kl": 0.01257875096052885,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57743002.63087248,
+      "logits/rejected": -51520362.29239766,
+      "logps/chosen": -468.58819211409394,
+      "logps/rejected": -198.2330957602339,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000025629997253,
+      "loss/total_with_kl": 0.500034749507904,
+      "metrics/advantage_var": 0.06440309435129166,
+      "regularization/advantage_var": 0.06440309435129166,
+      "regularization/kl": 3.220155122107826e-05,
+      "rewards/chosen": -6.928566480743005e-06,
+      "rewards/margins": -2.1701127549079173e-05,
+      "rewards/rejected": 1.4772561068336168e-05,
+      "stability/repetition_rate_mean": 0.1953125,
+      "stability/response_length_mean": 48.375,
+      "stability/response_length_std": 28.93311309814453,
+      "stability/response_length_var": 837.125,
+      "stability/token_entropy_mean": 4.513671875,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.014569000340998173,
+      "kl": 0.022648466750979424,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57641219.24050633,
+      "logits/rejected": -53900932.74074074,
+      "logps/chosen": -385.8970530063291,
+      "logps/rejected": -195.1368152006173,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999943673610687,
+      "loss/total_with_kl": 0.5000303387641907,
+      "metrics/advantage_var": 0.07199499756097794,
+      "regularization/advantage_var": 0.07199499756097794,
+      "regularization/kl": 3.599749834393151e-05,
+      "rewards/chosen": 3.946519492170479e-05,
+      "rewards/margins": 4.9802977542996476e-05,
+      "rewards/rejected": -1.0337782621291684e-05,
+      "stability/repetition_rate_mean": 0.35856893658638,
+      "stability/response_length_mean": 53.375,
+      "stability/response_length_std": 21.14533042907715,
+      "stability/response_length_var": 447.125,
+      "stability/token_entropy_mean": 5.330078125,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.009226334281265736,
+      "kl": 0.017865750938653946,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56646547.2,
+      "logits/rejected": -50243718.4,
+      "logps/chosen": -421.89150390625,
+      "logps/rejected": -219.2673583984375,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000039339065552,
+      "loss/total_with_kl": 0.5000437498092651,
+      "metrics/advantage_var": 0.07968886941671371,
+      "regularization/advantage_var": 0.07968886941671371,
+      "regularization/kl": 3.984443537774496e-05,
+      "rewards/chosen": 1.1162247392348946e-05,
+      "rewards/margins": -3.0723534291610124e-05,
+      "rewards/rejected": 4.188578168395907e-05,
+      "stability/repetition_rate_mean": 0.142578125,
+      "stability/response_length_mean": 33.125,
+      "stability/response_length_std": 33.01704025268555,
+      "stability/response_length_var": 1090.125,
+      "stability/token_entropy_mean": 4.9609375,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.012385486625134945,
+      "kl": 0.03356611356139183,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52715269.15436242,
+      "logits/rejected": -50790807.204678364,
+      "logps/chosen": -352.28240352348996,
+      "logps/rejected": -243.81743421052633,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000016093254089,
+      "loss/total_with_kl": 0.5000400543212891,
+      "metrics/advantage_var": 0.07688915729522705,
+      "regularization/advantage_var": 0.07688915729522705,
+      "regularization/kl": 3.844457751256414e-05,
+      "rewards/chosen": 4.6062358223751884e-05,
+      "rewards/margins": -1.0668867753415165e-05,
+      "rewards/rejected": 5.673122597716705e-05,
+      "stability/repetition_rate_mean": 0.1013549193739891,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.87324333190918,
+      "stability/response_length_var": 892.4107055664062,
+      "stability/token_entropy_mean": 4.775390625,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.00840709451586008,
+      "kl": 0.032784707844257355,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54045579.167785235,
+      "logits/rejected": -48300567.953216374,
+      "logps/chosen": -375.1681260486577,
+      "logps/rejected": -180.20900721856725,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999998211860657,
+      "loss/total_with_kl": 0.5000237226486206,
+      "metrics/advantage_var": 0.04776905104517937,
+      "regularization/advantage_var": 0.04776905104517937,
+      "regularization/kl": 2.388452776358463e-05,
+      "rewards/chosen": 3.098852732437569e-05,
+      "rewards/margins": -6.186709812701927e-07,
+      "rewards/rejected": 3.160719830564588e-05,
+      "stability/repetition_rate_mean": 0.3831075429916382,
+      "stability/response_length_mean": 53.375,
+      "stability/response_length_std": 21.480472564697266,
+      "stability/response_length_var": 461.41070556640625,
+      "stability/token_entropy_mean": 3.857421875,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.011991314589977264,
+      "kl": 0.03365620970726013,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56956018.48447205,
+      "logits/rejected": -49576486.641509436,
+      "logps/chosen": -344.5125679347826,
+      "logps/rejected": -230.43519064465409,
+      "loss": 0.5,
+      "loss/base_kto": 0.500001847743988,
+      "loss/total_with_kl": 0.5000297427177429,
+      "metrics/advantage_var": 0.055924512445926666,
+      "regularization/advantage_var": 0.055924512445926666,
+      "regularization/kl": 2.796225635393057e-05,
+      "rewards/chosen": -5.357490956598187e-07,
+      "rewards/margins": -1.33645037668121e-05,
+      "rewards/rejected": 1.2828754671152282e-05,
+      "stability/repetition_rate_mean": 0.11199301481246948,
+      "stability/response_length_mean": 22.375,
+      "stability/response_length_std": 30.236862182617188,
+      "stability/response_length_var": 914.2678833007812,
+      "stability/token_entropy_mean": 4.333984375,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.00939466804265976,
+      "kl": 0.024347607046365738,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52675459.878787875,
+      "logits/rejected": -47832952.56774194,
+      "logps/chosen": -359.6363636363636,
+      "logps/rejected": -178.1663054435484,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999696016311646,
+      "loss/total_with_kl": 0.5000283122062683,
+      "metrics/advantage_var": 0.06277040392160416,
+      "regularization/advantage_var": 0.06277040392160416,
+      "regularization/kl": 3.138520332868211e-05,
+      "rewards/chosen": 8.442694264830965e-05,
+      "rewards/margins": 2.3823355270853723e-05,
+      "rewards/rejected": 6.060358737745593e-05,
+      "stability/repetition_rate_mean": 0.23800872266292572,
+      "stability/response_length_mean": 53.625,
+      "stability/response_length_std": 22.11617088317871,
+      "stability/response_length_var": 489.125,
+      "stability/token_entropy_mean": 5.255859375,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.008829456754028797,
+      "kl": 0.03643302991986275,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57575225.80645161,
+      "logits/rejected": -47288897.163636364,
+      "logps/chosen": -381.90148689516127,
+      "logps/rejected": -183.55746922348484,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999722838401794,
+      "loss/total_with_kl": 0.5000311732292175,
+      "metrics/advantage_var": 0.06777071952819824,
+      "regularization/advantage_var": 0.06777071952819824,
+      "regularization/kl": 3.388536424608901e-05,
+      "rewards/chosen": 4.530613941531027e-05,
+      "rewards/margins": 2.0425156931431993e-05,
+      "rewards/rejected": 2.488098248387828e-05,
+      "stability/repetition_rate_mean": 0.1857219785451889,
+      "stability/response_length_mean": 21.5,
+      "stability/response_length_std": 26.635099411010742,
+      "stability/response_length_var": 709.4285888671875,
+      "stability/token_entropy_mean": 5.392578125,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.007817639969289303,
+      "kl": 0.04110654443502426,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52791732.02580645,
+      "logits/rejected": -49063154.03636364,
+      "logps/chosen": -379.0649193548387,
+      "logps/rejected": -220.3504734848485,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000015497207642,
+      "loss/total_with_kl": 0.5000338554382324,
+      "metrics/advantage_var": 0.06463103741407394,
+      "regularization/advantage_var": 0.06463103741407394,
+      "regularization/kl": 3.2315518183168024e-05,
+      "rewards/chosen": 6.355317609925424e-05,
+      "rewards/margins": -8.118214524503094e-06,
+      "rewards/rejected": 7.167139062375734e-05,
+      "stability/repetition_rate_mean": 0.171875,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.6875,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.009152984246611595,
+      "kl": 0.03520666062831879,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49518003.89189189,
+      "logits/rejected": -47914368.0,
+      "logps/chosen": -314.1993771114865,
+      "logps/rejected": -228.88335755813952,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999912679195404,
+      "loss/total_with_kl": 0.50002521276474,
+      "metrics/advantage_var": 0.06784234941005707,
+      "regularization/advantage_var": 0.06784234941005707,
+      "regularization/kl": 3.3921172871487215e-05,
+      "rewards/chosen": 7.1432428887567e-05,
+      "rewards/margins": 6.99527711958152e-05,
+      "rewards/rejected": 1.4796576917518016e-06,
+      "stability/repetition_rate_mean": 0.14558292925357819,
+      "stability/response_length_mean": 24.25,
+      "stability/response_length_std": 30.093900680541992,
+      "stability/response_length_var": 905.6428833007812,
+      "stability/token_entropy_mean": 2.78125,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.010503264144062996,
+      "kl": 0.051506686955690384,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58671863.52317881,
+      "logits/rejected": -53488839.95266272,
+      "logps/chosen": -403.9934292218543,
+      "logps/rejected": -201.02734375,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999984800815582,
+      "loss/total_with_kl": 0.5000309348106384,
+      "metrics/advantage_var": 0.06490852683782578,
+      "regularization/advantage_var": 0.06490852683782578,
+      "regularization/kl": 3.245426341891289e-05,
+      "rewards/chosen": 3.66592804426389e-05,
+      "rewards/margins": 8.198070828566575e-06,
+      "rewards/rejected": 2.8461209614072326e-05,
+      "stability/repetition_rate_mean": 0.19088542461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.234375,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.009465193375945091,
+      "kl": 0.03408239036798477,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51497151.58709677,
+      "logits/rejected": -47171549.86666667,
+      "logps/chosen": -358.2992943548387,
+      "logps/rejected": -230.4330018939394,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999475479125977,
+      "loss/total_with_kl": 0.5000323057174683,
+      "metrics/advantage_var": 0.07506581395864487,
+      "regularization/advantage_var": 0.07506581395864487,
+      "regularization/kl": 3.753290729946457e-05,
+      "rewards/chosen": 3.3219540191273537e-05,
+      "rewards/margins": 4.1400590949743726e-05,
+      "rewards/rejected": -8.181050758470188e-06,
+      "stability/repetition_rate_mean": 0.15742863714694977,
+      "stability/response_length_mean": 24.875,
+      "stability/response_length_std": 27.424898147583008,
+      "stability/response_length_var": 752.125,
+      "stability/token_entropy_mean": 4.921875,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.012723367661237717,
+      "kl": 0.044009942561388016,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58462437.743589744,
+      "logits/rejected": -50612860.87804878,
+      "logps/chosen": -380.76332131410254,
+      "logps/rejected": -187.201064691311,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999934434890747,
+      "loss/total_with_kl": 0.5000306963920593,
+      "metrics/advantage_var": 0.06280595809221268,
+      "regularization/advantage_var": 0.06280595809221268,
+      "regularization/kl": 3.140298213111237e-05,
+      "rewards/chosen": 2.6056644482872424e-05,
+      "rewards/margins": 3.132729029854866e-07,
+      "rewards/rejected": 2.5743371579886937e-05,
+      "stability/repetition_rate_mean": 0.2192234843969345,
+      "stability/response_length_mean": 33.625,
+      "stability/response_length_std": 29.442134857177734,
+      "stability/response_length_var": 866.8392944335938,
+      "stability/token_entropy_mean": 3.70263671875,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.011232409626245499,
+      "kl": 0.043333958834409714,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50442868.658227846,
+      "logits/rejected": -49033930.27160494,
+      "logps/chosen": -314.9471172863924,
+      "logps/rejected": -207.13204089506172,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999579787254333,
+      "loss/total_with_kl": 0.5000306367874146,
+      "metrics/advantage_var": 0.06963532418012619,
+      "regularization/advantage_var": 0.06963532418012619,
+      "regularization/kl": 3.48176654370036e-05,
+      "rewards/chosen": 4.625187514797796e-05,
+      "rewards/margins": 3.456790598741307e-05,
+      "rewards/rejected": 1.1683969160564887e-05,
+      "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": 3.74609375,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.010943830944597721,
+      "kl": 0.04752064868807793,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54759797.62162162,
+      "logits/rejected": -50968784.37209302,
+      "logps/chosen": -359.7860272381757,
+      "logps/rejected": -216.5623410247093,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999910295009613,
+      "loss/total_with_kl": 0.500036895275116,
+      "metrics/advantage_var": 0.09180708229541779,
+      "regularization/advantage_var": 0.09180708229541779,
+      "regularization/kl": 4.5903540012659505e-05,
+      "rewards/chosen": 8.41418545491792e-05,
+      "rewards/margins": 6.757945212752732e-05,
+      "rewards/rejected": 1.6562402421651884e-05,
+      "stability/repetition_rate_mean": 0.24177631735801697,
+      "stability/response_length_mean": 38.0,
+      "stability/response_length_std": 30.10457992553711,
+      "stability/response_length_var": 906.2857055664062,
+      "stability/token_entropy_mean": 4.23828125,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.01067819632589817,
+      "kl": 0.04482289031147957,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59449306.98795181,
+      "logits/rejected": -53884203.22077922,
+      "logps/chosen": -399.820265436747,
+      "logps/rejected": -209.97304941152598,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999380111694336,
+      "loss/total_with_kl": 0.5000303983688354,
+      "metrics/advantage_var": 0.073161780834198,
+      "regularization/advantage_var": 0.073161780834198,
+      "regularization/kl": 3.658089190139435e-05,
+      "rewards/chosen": 8.220696844250323e-05,
+      "rewards/margins": 5.143789996083652e-05,
+      "rewards/rejected": 3.0769068481666704e-05,
+      "stability/repetition_rate_mean": 0.2014508843421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.384765625,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.00925524067133665,
+      "kl": 0.050605691969394684,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57465559.90361446,
+      "logits/rejected": -53690846.753246754,
+      "logps/chosen": -378.5371799698795,
+      "logps/rejected": -206.05924056412337,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000004768371582,
+      "loss/total_with_kl": 0.5000382661819458,
+      "metrics/advantage_var": 0.07546661794185638,
+      "regularization/advantage_var": 0.07546661794185638,
+      "regularization/kl": 3.773330900003202e-05,
+      "rewards/chosen": 2.6093964206885144e-05,
+      "rewards/margins": -2.7925838551514204e-06,
+      "rewards/rejected": 2.8886548062036564e-05,
+      "stability/repetition_rate_mean": 0.30853456258773804,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.911476135253906,
+      "stability/response_length_var": 894.6964111328125,
+      "stability/token_entropy_mean": 4.412109375,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.009009272791445255,
+      "kl": 0.04474036023020744,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60689090.43037975,
+      "logits/rejected": -53511787.45679013,
+      "logps/chosen": -364.50944422468353,
+      "logps/rejected": -201.17226080246914,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999332427978516,
+      "loss/total_with_kl": 0.5000336766242981,
+      "metrics/advantage_var": 0.08062459528446198,
+      "regularization/advantage_var": 0.08062459528446198,
+      "regularization/kl": 4.031229764223099e-05,
+      "rewards/chosen": 8.311558844922464e-05,
+      "rewards/margins": 5.8828055319233806e-05,
+      "rewards/rejected": 2.4287533129990837e-05,
+      "stability/repetition_rate_mean": 0.2506510615348816,
+      "stability/response_length_mean": 33.125,
+      "stability/response_length_std": 33.0516357421875,
+      "stability/response_length_var": 1092.4107666015625,
+      "stability/token_entropy_mean": 4.142578125,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.008836813271045685,
+      "kl": 0.03717491030693054,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57836122.35294118,
+      "logits/rejected": -54507458.68263473,
+      "logps/chosen": -384.19738051470586,
+      "logps/rejected": -212.1802020958084,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999862909317017,
+      "loss/total_with_kl": 0.5000400543212891,
+      "metrics/advantage_var": 0.08291424810886383,
+      "regularization/advantage_var": 0.08291424810886383,
+      "regularization/kl": 4.145712227909826e-05,
+      "rewards/chosen": 4.753693207806232e-05,
+      "rewards/margins": 1.1379732890869236e-05,
+      "rewards/rejected": 3.6157199187193086e-05,
+      "stability/repetition_rate_mean": 0.17961646616458893,
+      "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": 0.009799664840102196,
+      "kl": 0.03611461818218231,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60827171.06849315,
+      "logits/rejected": -55819781.88505747,
+      "logps/chosen": -417.35030500856163,
+      "logps/rejected": -216.33441091954023,
+      "loss": 0.5,
+      "loss/base_kto": 0.5000001192092896,
+      "loss/total_with_kl": 0.5000359416007996,
+      "metrics/advantage_var": 0.0716974139213562,
+      "regularization/advantage_var": 0.0716974139213562,
+      "regularization/kl": 3.5848712286679074e-05,
+      "rewards/chosen": 2.899690059154001e-05,
+      "rewards/margins": -3.782787809415023e-06,
+      "rewards/rejected": 3.2779688400955034e-05,
+      "stability/repetition_rate_mean": 0.16328126192092896,
+      "stability/response_length_mean": 15.25,
+      "stability/response_length_std": 24.93276596069336,
+      "stability/response_length_var": 621.6428833007812,
+      "stability/token_entropy_mean": 3.625,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.009806789457798004,
+      "kl": 0.04913714528083801,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57520075.1483871,
+      "logits/rejected": -53990958.54545455,
+      "logps/chosen": -392.6003024193548,
+      "logps/rejected": -183.9016690340909,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999481439590454,
+      "loss/total_with_kl": 0.5000341534614563,
+      "metrics/advantage_var": 0.0785689577460289,
+      "regularization/advantage_var": 0.0785689577460289,
+      "regularization/kl": 3.92844776797574e-05,
+      "rewards/chosen": 5.7653367759719974e-05,
+      "rewards/margins": 4.0403727117929525e-05,
+      "rewards/rejected": 1.7249640641790448e-05,
+      "stability/repetition_rate_mean": 0.22703124582767487,
+      "stability/response_length_mean": 43.25,
+      "stability/response_length_std": 29.620214462280273,
+      "stability/response_length_var": 877.3571166992188,
+      "stability/token_entropy_mean": 3.57421875,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.009469456970691681,
+      "kl": 0.05869415029883385,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59993232.49079755,
+      "logits/rejected": -55193215.18471338,
+      "logps/chosen": -401.71668903374234,
+      "logps/rejected": -247.83942078025478,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999094009399414,
+      "loss/total_with_kl": 0.5000352263450623,
+      "metrics/advantage_var": 0.08847339451313019,
+      "regularization/advantage_var": 0.08847339451313019,
+      "regularization/kl": 4.423669815878384e-05,
+      "rewards/chosen": 9.945813177919095e-05,
+      "rewards/margins": 6.812238380639579e-05,
+      "rewards/rejected": 3.1335747972795155e-05,
+      "stability/repetition_rate_mean": 0.2230902761220932,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 30.20406723022461,
+      "stability/response_length_var": 912.2857055664062,
+      "stability/token_entropy_mean": 3.427734375,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.008573731407523155,
+      "kl": 0.0394296832382679,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57256320.77575757,
+      "logits/rejected": -50278221.62580645,
+      "logps/chosen": -367.1009943181818,
+      "logps/rejected": -190.01449092741936,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999907910823822,
+      "loss/total_with_kl": 0.5000235438346863,
+      "metrics/advantage_var": 0.0654965192079544,
+      "regularization/advantage_var": 0.0654965192079544,
+      "regularization/kl": 3.274825940025039e-05,
+      "rewards/chosen": 8.873008756023464e-05,
+      "rewards/margins": 6.875481660444132e-05,
+      "rewards/rejected": 1.997527095579332e-05,
+      "stability/repetition_rate_mean": 0.34765625,
+      "stability/response_length_mean": 56.25,
+      "stability/response_length_std": 21.920310974121094,
+      "stability/response_length_var": 480.5,
+      "stability/token_entropy_mean": 4.7548828125,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.010250517167150974,
+      "kl": 0.051040973514318466,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51322105.8313253,
+      "logits/rejected": -51878599.48051948,
+      "logps/chosen": -330.19470067771084,
+      "logps/rejected": -199.5067091112013,
+      "loss": 0.5,
+      "loss/base_kto": 0.49998751282691956,
+      "loss/total_with_kl": 0.5000160336494446,
+      "metrics/advantage_var": 0.05701936408877373,
+      "regularization/advantage_var": 0.05701936408877373,
+      "regularization/kl": 2.8509681214927696e-05,
+      "rewards/chosen": 7.712891527328147e-05,
+      "rewards/margins": 9.591767240252724e-05,
+      "rewards/rejected": -1.8788757129245763e-05,
+      "stability/repetition_rate_mean": 0.25424888730049133,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.73423957824707,
+      "stability/response_length_var": 884.125,
+      "stability/token_entropy_mean": 5.306640625,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.01047827210277319,
+      "kl": 0.056698884814977646,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55602344.65882353,
+      "logits/rejected": -50412625.92,
+      "logps/chosen": -397.7995863970588,
+      "logps/rejected": -207.72515625,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999827444553375,
+      "loss/total_with_kl": 0.5000240206718445,
+      "metrics/advantage_var": 0.08249103277921677,
+      "regularization/advantage_var": 0.08249103277921677,
+      "regularization/kl": 4.1245522879762575e-05,
+      "rewards/chosen": 0.00013485204647569095,
+      "rewards/margins": 0.00013406771958769098,
+      "rewards/rejected": 7.843268879999717e-07,
+      "stability/repetition_rate_mean": 0.20331649482250214,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.193626403808594,
+      "stability/response_length_var": 852.2678833007812,
+      "stability/token_entropy_mean": 4.59765625,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.009275847114622593,
+      "kl": 0.050125040113925934,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54248122.729411766,
+      "logits/rejected": -49590394.88,
+      "logps/chosen": -360.94453125,
+      "logps/rejected": -191.30770833333332,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999949038028717,
+      "loss/total_with_kl": 0.5000295639038086,
+      "metrics/advantage_var": 0.06931816786527634,
+      "regularization/advantage_var": 0.06931816786527634,
+      "regularization/kl": 3.465908230282366e-05,
+      "rewards/chosen": 4.4789784314001306e-05,
+      "rewards/margins": 4.690334328687658e-05,
+      "rewards/rejected": -2.113558972875277e-06,
+      "stability/repetition_rate_mean": 0.28984373807907104,
+      "stability/response_length_mean": 53.125,
+      "stability/response_length_std": 22.67432403564453,
+      "stability/response_length_var": 514.125,
+      "stability/token_entropy_mean": 4.5673828125,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.009821859188377857,
+      "kl": 0.049432478845119476,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55498038.85714286,
+      "logits/rejected": -49266189.473684214,
+      "logps/chosen": -416.84291294642856,
+      "logps/rejected": -186.8680998149671,
+      "loss": 0.5,
+      "loss/base_kto": 0.49999743700027466,
+      "loss/total_with_kl": 0.5000368356704712,
+      "metrics/advantage_var": 0.07872576266527176,
+      "regularization/advantage_var": 0.07872576266527176,
+      "regularization/kl": 3.936288339900784e-05,
+      "rewards/chosen": 5.241769499012402e-05,
+      "rewards/margins": 2.316671556660107e-05,
+      "rewards/rejected": 2.925097942352295e-05,
+      "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": 3.828125,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.008929993025958538,
+      "kl": 0.04661773145198822,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54674326.06896552,
+      "logits/rejected": -50117249.75342466,
+      "logps/chosen": -370.8756061422414,
+      "logps/rejected": -212.5759578339041,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999971389770508,
+      "loss/total_with_kl": 0.5000324845314026,
+      "metrics/advantage_var": 0.0706528052687645,
+      "regularization/advantage_var": 0.0706528052687645,
+      "regularization/kl": 3.532640403136611e-05,
+      "rewards/chosen": 4.114243255435735e-05,
+      "rewards/margins": 2.4976168516181596e-05,
+      "rewards/rejected": 1.6166264038175754e-05,
+      "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": 4.83984375,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.008566767908632755,
+      "kl": 0.062040429562330246,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59409420.1183432,
+      "logits/rejected": -52976277.19205298,
+      "logps/chosen": -441.2774593195266,
+      "logps/rejected": -191.24706384519868,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999937117099762,
+      "loss/total_with_kl": 0.5000309944152832,
+      "metrics/advantage_var": 0.07455558329820633,
+      "regularization/advantage_var": 0.07455558329820633,
+      "regularization/kl": 3.727779403561726e-05,
+      "rewards/chosen": 6.438495822735792e-05,
+      "rewards/margins": 5.361681233712991e-05,
+      "rewards/rejected": 1.0768145890228007e-05,
+      "stability/repetition_rate_mean": 0.20866228640079498,
+      "stability/response_length_mean": 28.25,
+      "stability/response_length_std": 30.844078063964844,
+      "stability/response_length_var": 951.3571166992188,
+      "stability/token_entropy_mean": 3.31884765625,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.010120969265699387,
+      "kl": 0.06263448297977448,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56958445.4939759,
+      "logits/rejected": -48456025.766233765,
+      "logps/chosen": -356.2844973644578,
+      "logps/rejected": -194.80245535714286,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999809265136719,
+      "loss/total_with_kl": 0.5000235438346863,
+      "metrics/advantage_var": 0.08520467579364777,
+      "regularization/advantage_var": 0.08520467579364777,
+      "regularization/kl": 4.260233981767669e-05,
+      "rewards/chosen": 0.00015270808734089495,
+      "rewards/margins": 0.0001476406334579356,
+      "rewards/rejected": 5.0674538829593685e-06,
+      "stability/repetition_rate_mean": 0.2936941981315613,
+      "stability/response_length_mean": 32.25,
+      "stability/response_length_std": 27.932315826416016,
+      "stability/response_length_var": 780.2142944335938,
+      "stability/token_entropy_mean": 4.7060546875,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.010403833352029324,
+      "kl": 0.07130970805883408,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52957824.78527607,
+      "logits/rejected": -48375214.47133758,
+      "logps/chosen": -376.07349980828224,
+      "logps/rejected": -211.2522641321656,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999971091747284,
+      "loss/total_with_kl": 0.5000388622283936,
+      "metrics/advantage_var": 0.08360155671834946,
+      "regularization/advantage_var": 0.08360155671834946,
+      "regularization/kl": 4.180078030913137e-05,
+      "rewards/chosen": 7.232671683551343e-05,
+      "rewards/margins": 2.6380315060250514e-05,
+      "rewards/rejected": 4.594640177526292e-05,
+      "stability/repetition_rate_mean": 0.251953125,
+      "stability/response_length_mean": 40.375,
+      "stability/response_length_std": 32.61874008178711,
+      "stability/response_length_var": 1063.982177734375,
+      "stability/token_entropy_mean": 3.263671875,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.01109651941806078,
+      "kl": 0.07014961540699005,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54954995.358024694,
+      "logits/rejected": -49891590.48101266,
+      "logps/chosen": -311.9007040895062,
+      "logps/rejected": -197.01834454113924,
+      "loss": 0.5,
+      "loss/base_kto": 0.4999902844429016,
+      "loss/total_with_kl": 0.5000271201133728,
+      "metrics/advantage_var": 0.07348966598510742,
+      "regularization/advantage_var": 0.07348966598510742,
+      "regularization/kl": 3.674483377835713e-05,
+      "rewards/chosen": 0.00010068489261615424,
+      "rewards/margins": 7.98731516469762e-05,
+      "rewards/rejected": 2.081174096917804e-05,
+      "stability/repetition_rate_mean": 0.18998579680919647,
+      "stability/response_length_mean": 22.25,
+      "stability/response_length_std": 27.78874397277832,
+      "stability/response_length_var": 772.2142944335938,
+      "stability/token_entropy_mean": 3.8359375,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.010649079456925392,
+      "kl": 0.06624338775873184,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53431853.98802395,
+      "logits/rejected": -50099728.732026145,
+      "logps/chosen": -392.27919161676647,
+      "logps/rejected": -211.17078993055554,
+      "loss": 0.5,
+      "loss/base_kto": 0.49998751282691956,
+      "loss/total_with_kl": 0.50002521276474,
+      "metrics/advantage_var": 0.07544363290071487,
+      "regularization/advantage_var": 0.07544363290071487,
+      "regularization/kl": 3.772181662498042e-05,
+      "rewards/chosen": 0.0001261207068751672,
+      "rewards/margins": 9.833026342690258e-05,
+      "rewards/rejected": 2.7790443448264613e-05,
+      "stability/repetition_rate_mean": 0.1106770858168602,
+      "stability/response_length_mean": 11.5,
+      "stability/response_length_std": 21.758413314819336,
+      "stability/response_length_var": 473.4285583496094,
+      "stability/token_entropy_mean": 5.3671875,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.01244101021438837,
+      "kl": 0.058560874313116074,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60543167.58709677,
+      "logits/rejected": -54317372.50909091,
+      "logps/chosen": -392.6783266129032,
+      "logps/rejected": -197.91171875,
+      "loss": 0.5,
+      "loss/base_kto": 0.49998798966407776,
+      "loss/total_with_kl": 0.5000260472297668,
+      "metrics/advantage_var": 0.07610367983579636,
+      "regularization/advantage_var": 0.07610367983579636,
+      "regularization/kl": 3.805184314842336e-05,
+      "rewards/chosen": 0.00010660523849148904,
+      "rewards/margins": 9.897794790260078e-05,
+      "rewards/rejected": 7.627290588888255e-06,
+      "stability/repetition_rate_mean": 0.3782551884651184,
+      "stability/response_length_mean": 38.875,
+      "stability/response_length_std": 30.786998748779297,
+      "stability/response_length_var": 947.8392944335938,
+      "stability/token_entropy_mean": 4.490234375,
+      "step": 195
+    },
+    {
+      "epoch": 1.0,
+      "step": 196,
+      "total_flos": 3.1861866257511875e+18,
+      "train_loss": 0.5000301818458401,
+      "train_runtime": 3151.2022,
+      "train_samples_per_second": 31.734,
+      "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-beta0.001/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e74897fa70a3117fd3b115a078e94a832dd2526a53fc52afc283e411a44afb4
+size 8017