Upload folder using huggingface_hub

mistral7b_kto_simple/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_simple
+  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_simple
+
+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_simple/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": [
+    "down_proj",
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "k_proj",
+    "o_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_simple/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5544a3a84bd8dc923678c6961615880b549a38cafdde54b75d73e67b650e2329
+size 42002584

mistral7b_kto_simple/all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.4956757848968311,
+    "train_runtime": 3145.5892,
+    "train_samples_per_second": 31.791,
+    "train_steps_per_second": 0.062
+}

mistral7b_kto_simple/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_simple/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_simple/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": [
+    "down_proj",
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "k_proj",
+    "o_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_simple/checkpoint-196/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5544a3a84bd8dc923678c6961615880b549a38cafdde54b75d73e67b650e2329
+size 42002584

mistral7b_kto_simple/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_simple/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 sha256:2155d12f095dd96e8f7d2c9a6034f6d4c096981be3dc093ad067a71826fb7d5b
+size 31467333

mistral7b_kto_simple/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 sha256:2037c2dcd8461d550ba8b4abb1ed488ddf6263dd115cc5d99d426fa1e863f2a7
+size 31467333

mistral7b_kto_simple/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:c3826dad26098f571242e2ab2d1d1ae834e43185dfa47141182b28d59898c782
+size 31467333

mistral7b_kto_simple/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:b4c9a0df6c03f60b32ebbc4e8e8f5c05e649e3c9e883d4709540561f809b2da0
+size 31467397

mistral7b_kto_simple/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:eb5e5f1369a513586db7ebcf22c4662857127b6988d9cfc4bf3eb1aeb06ae243
+size 31467397

mistral7b_kto_simple/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 sha256:753857dbf7d831dcd4426226cd7c4a39d798a148fa90fdc7b8baf5399502af0f
+size 31467397

mistral7b_kto_simple/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 sha256:2baee89de6d255f1404c1525d84cd7b8037c15be5b006dd2476fd0a36fd10b83
+size 31467397

mistral7b_kto_simple/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 sha256:29bb2b13fde10c3a7085f09558d67817b02cd50df65e0d63bfad34bebf768543
+size 31467397

mistral7b_kto_simple/checkpoint-196/global_step196/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5cf055cacbb4c8d05fedc1cd84ed557bac1056867f5cdd3ed85b9defb0e5d4a6
+size 42288409

mistral7b_kto_simple/checkpoint-196/latest

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

mistral7b_kto_simple/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_simple/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_simple/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_simple/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_simple/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_simple/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_simple/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_simple/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_simple/checkpoint-196/scheduler.pt

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

mistral7b_kto_simple/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_simple/checkpoint-196/tokenizer.model

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

mistral7b_kto_simple/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_simple/checkpoint-196/trainer_state.json

@@ -0,0 +1,848 @@
+{
+  "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": 0.5772594809532166,
+      "kl": 0.006305641960352659,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59718379.78947368,
+      "logits/rejected": -50413504.0,
+      "logps/chosen": -439.03207236842104,
+      "logps/rejected": -198.4214332217262,
+      "loss": 0.5,
+      "loss/base_kto": 0.49969324469566345,
+      "rewards/chosen": 0.0013566404385002035,
+      "rewards/margins": 0.0025594215466638558,
+      "rewards/rejected": -0.001202781108163652,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.6031567454338074,
+      "kl": 0.028230074793100357,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59308962.90909091,
+      "logits/rejected": -50030080.0,
+      "logps/chosen": -429.9140625,
+      "logps/rejected": -192.86823465737953,
+      "loss": 0.5001,
+      "loss/base_kto": 0.49971267580986023,
+      "rewards/chosen": 0.002450463833746972,
+      "rewards/margins": 0.0024800021542366484,
+      "rewards/rejected": -2.95383204896766e-05,
+      "stability/repetition_rate_mean": 0.3664148449897766,
+      "stability/response_length_mean": 44.25,
+      "stability/response_length_std": 28.13614845275879,
+      "stability/response_length_var": 791.6428833007812,
+      "stability/token_entropy_mean": 3.4755859375,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.609616219997406,
+      "kl": 0.012302148155868053,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56221720.82424243,
+      "logits/rejected": -51918762.11612903,
+      "logps/chosen": -404.1202651515151,
+      "logps/rejected": -200.68004032258065,
+      "loss": 0.4998,
+      "loss/base_kto": 0.500029981136322,
+      "rewards/chosen": 0.001057231787479285,
+      "rewards/margins": -0.00034482885199441484,
+      "rewards/rejected": 0.0014020606394736998,
+      "stability/repetition_rate_mean": 0.2929687201976776,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.92341423034668,
+      "stability/response_length_var": 895.4107055664062,
+      "stability/token_entropy_mean": 3.7001953125,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.6197210550308228,
+      "kl": 0.016423773020505905,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51663750.4,
+      "logits/rejected": -48055072.0,
+      "logps/chosen": -343.6933349609375,
+      "logps/rejected": -225.6587890625,
+      "loss": 0.5004,
+      "loss/base_kto": 0.5000466704368591,
+      "rewards/chosen": 0.00033793365582823753,
+      "rewards/margins": -0.00011792583391070368,
+      "rewards/rejected": 0.0004558594897389412,
+      "stability/repetition_rate_mean": 0.2155257910490036,
+      "stability/response_length_mean": 44.0,
+      "stability/response_length_std": 28.0611572265625,
+      "stability/response_length_var": 787.4285888671875,
+      "stability/token_entropy_mean": 4.42578125,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.5772079229354858,
+      "kl": 0.01958128809928894,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56465217.32413793,
+      "logits/rejected": -52019422.35428572,
+      "logps/chosen": -378.4753232758621,
+      "logps/rejected": -202.7753125,
+      "loss": 0.5001,
+      "loss/base_kto": 0.49998876452445984,
+      "rewards/chosen": 0.0016139640890318772,
+      "rewards/margins": -0.00013830620079792184,
+      "rewards/rejected": 0.001752270289829799,
+      "stability/repetition_rate_mean": 0.19711539149284363,
+      "stability/response_length_mean": 50.25,
+      "stability/response_length_std": 25.549671173095703,
+      "stability/response_length_var": 652.7857055664062,
+      "stability/token_entropy_mean": 4.5234375,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.6031031012535095,
+      "kl": 0.020890694111585617,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57728749.10067114,
+      "logits/rejected": -51514302.12865497,
+      "logps/chosen": -468.5621854026846,
+      "logps/rejected": -198.23862390350877,
+      "loss": 0.5002,
+      "loss/base_kto": 0.49984827637672424,
+      "rewards/chosen": 0.0019035963404098613,
+      "rewards/margins": 0.000978329737129632,
+      "rewards/rejected": 0.0009252666032802292,
+      "stability/repetition_rate_mean": 0.1525297611951828,
+      "stability/response_length_mean": 43.0,
+      "stability/response_length_std": 29.602123260498047,
+      "stability/response_length_var": 876.2857055664062,
+      "stability/token_entropy_mean": 4.568359375,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6770267486572266,
+      "kl": 0.031000053510069847,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57633195.74683544,
+      "logits/rejected": -53898088.2962963,
+      "logps/chosen": -385.922245846519,
+      "logps/rejected": -195.10961612654322,
+      "loss": 0.4998,
+      "loss/base_kto": 0.5000401139259338,
+      "rewards/chosen": 0.0014279391380805004,
+      "rewards/margins": -0.0002580981845407863,
+      "rewards/rejected": 0.0016860373226212867,
+      "stability/repetition_rate_mean": 0.3030133843421936,
+      "stability/response_length_mean": 56.875,
+      "stability/response_length_std": 20.152544021606445,
+      "stability/response_length_var": 406.125,
+      "stability/token_entropy_mean": 5.474609375,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.587213933467865,
+      "kl": 0.0245906300842762,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56631648.0,
+      "logits/rejected": -50232096.0,
+      "logps/chosen": -421.8708984375,
+      "logps/rejected": -219.26767578125,
+      "loss": 0.4998,
+      "loss/base_kto": 0.50019770860672,
+      "rewards/chosen": 0.003170560672879219,
+      "rewards/margins": -0.0009862679988145826,
+      "rewards/rejected": 0.004156828671693802,
+      "stability/repetition_rate_mean": 0.15522204339504242,
+      "stability/response_length_mean": 35.0,
+      "stability/response_length_std": 31.518701553344727,
+      "stability/response_length_var": 993.4285888671875,
+      "stability/token_entropy_mean": 4.7880859375,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.640202522277832,
+      "kl": 0.038881849497556686,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52691122.684563756,
+      "logits/rejected": -50773704.60818713,
+      "logps/chosen": -352.2867030201342,
+      "logps/rejected": -243.83422423245614,
+      "loss": 0.4998,
+      "loss/base_kto": 0.4999396800994873,
+      "rewards/chosen": 0.004177153910566497,
+      "rewards/margins": 0.0001828821414053836,
+      "rewards/rejected": 0.003994271769161113,
+      "stability/repetition_rate_mean": 0.0859469324350357,
+      "stability/response_length_mean": 22.375,
+      "stability/response_length_std": 26.52189064025879,
+      "stability/response_length_var": 703.4107055664062,
+      "stability/token_entropy_mean": 4.9140625,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.5786681175231934,
+      "kl": 0.04716196283698082,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54047022.389261745,
+      "logits/rejected": -48305211.883040935,
+      "logps/chosen": -375.11949454697987,
+      "logps/rejected": -180.19741410818713,
+      "loss": 0.4999,
+      "loss/base_kto": 0.499483197927475,
+      "rewards/chosen": 0.007963991005148663,
+      "rewards/margins": 0.0036445832186064505,
+      "rewards/rejected": 0.004319407786542212,
+      "stability/repetition_rate_mean": 0.4108419120311737,
+      "stability/response_length_mean": 55.375,
+      "stability/response_length_std": 16.361431121826172,
+      "stability/response_length_var": 267.6964416503906,
+      "stability/token_entropy_mean": 3.81640625,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.663858950138092,
+      "kl": 0.0537404902279377,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56946357.26708075,
+      "logits/rejected": -49559825.710691825,
+      "logps/chosen": -344.4793769409938,
+      "logps/rejected": -230.40548840408806,
+      "loss": 0.4996,
+      "loss/base_kto": 0.5001100897789001,
+      "rewards/chosen": 0.003263072937912082,
+      "rewards/margins": -0.0009928640175938652,
+      "rewards/rejected": 0.004255936955505947,
+      "stability/repetition_rate_mean": 0.107421875,
+      "stability/response_length_mean": 24.5,
+      "stability/response_length_std": 32.710853576660156,
+      "stability/response_length_var": 1070.0,
+      "stability/token_entropy_mean": 4.41015625,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.5810667276382446,
+      "kl": 0.05833861976861954,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52670401.93939394,
+      "logits/rejected": -47818137.6,
+      "logps/chosen": -359.6366477272727,
+      "logps/rejected": -178.17121975806452,
+      "loss": 0.4997,
+      "loss/base_kto": 0.499610960483551,
+      "rewards/chosen": 0.008416923609646883,
+      "rewards/margins": 0.002847369465310552,
+      "rewards/rejected": 0.005569554144336331,
+      "stability/repetition_rate_mean": 0.18834353983402252,
+      "stability/response_length_mean": 44.625,
+      "stability/response_length_std": 27.8820858001709,
+      "stability/response_length_var": 777.4107055664062,
+      "stability/token_entropy_mean": 4.666015625,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.5759316682815552,
+      "kl": 0.07869795709848404,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57552697.80645161,
+      "logits/rejected": -47262831.70909091,
+      "logps/chosen": -381.796875,
+      "logps/rejected": -183.53030303030303,
+      "loss": 0.4996,
+      "loss/base_kto": 0.4987983703613281,
+      "rewards/chosen": 0.014990917328865297,
+      "rewards/margins": 0.009785529548704684,
+      "rewards/rejected": 0.005205387780160615,
+      "stability/repetition_rate_mean": 0.19731052219867706,
+      "stability/response_length_mean": 21.125,
+      "stability/response_length_std": 28.027729034423828,
+      "stability/response_length_var": 785.5535888671875,
+      "stability/token_entropy_mean": 5.58984375,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5528621673583984,
+      "kl": 0.08887721598148346,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52767777.03225806,
+      "logits/rejected": -49031406.93333333,
+      "logps/chosen": -379.00811491935485,
+      "logps/rejected": -220.3328125,
+      "loss": 0.4992,
+      "loss/base_kto": 0.49963781237602234,
+      "rewards/chosen": 0.012032072005733366,
+      "rewards/margins": 0.0030974022221122426,
+      "rewards/rejected": 0.008934669783621124,
+      "stability/repetition_rate_mean": 0.185546875,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.7265625,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.6098544001579285,
+      "kl": 0.09266681969165802,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49499744.86486486,
+      "logits/rejected": -47894179.72093023,
+      "logps/chosen": -314.1058910472973,
+      "logps/rejected": -228.81817768895348,
+      "loss": 0.4993,
+      "loss/base_kto": 0.49878960847854614,
+      "rewards/chosen": 0.016490372451576026,
+      "rewards/margins": 0.009822355097453599,
+      "rewards/rejected": 0.006668017354122428,
+      "stability/repetition_rate_mean": 0.08308292925357819,
+      "stability/response_length_mean": 22.75,
+      "stability/response_length_std": 31.07479476928711,
+      "stability/response_length_var": 965.6428833007812,
+      "stability/token_entropy_mean": 2.146484375,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.6027852892875671,
+      "kl": 0.12336266040802002,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58629587.9205298,
+      "logits/rejected": -53451757.82248521,
+      "logps/chosen": -403.85350269039736,
+      "logps/rejected": -200.96463572485206,
+      "loss": 0.4991,
+      "loss/base_kto": 0.4989641308784485,
+      "rewards/chosen": 0.017660609933714204,
+      "rewards/margins": 0.008543980056680181,
+      "rewards/rejected": 0.009116629877034023,
+      "stability/repetition_rate_mean": 0.18541666865348816,
+      "stability/response_length_mean": 39.0,
+      "stability/response_length_std": 29.159658432006836,
+      "stability/response_length_var": 850.2857055664062,
+      "stability/token_entropy_mean": 5.22265625,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.6124152541160583,
+      "kl": 0.125388041138649,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51481851.04516129,
+      "logits/rejected": -47137016.24242424,
+      "logps/chosen": -358.1038306451613,
+      "logps/rejected": -230.3265625,
+      "loss": 0.4988,
+      "loss/base_kto": 0.4984590709209442,
+      "rewards/chosen": 0.022868017996511152,
+      "rewards/margins": 0.013043898081849399,
+      "rewards/rejected": 0.009824119914661754,
+      "stability/repetition_rate_mean": 0.12302795797586441,
+      "stability/response_length_mean": 27.375,
+      "stability/response_length_std": 30.39707374572754,
+      "stability/response_length_var": 923.9821166992188,
+      "stability/token_entropy_mean": 4.92578125,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.6379120945930481,
+      "kl": 0.17190246284008026,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58409432.615384616,
+      "logits/rejected": -50583776.780487806,
+      "logps/chosen": -380.54447115384613,
+      "logps/rejected": -187.12595274390245,
+      "loss": 0.4984,
+      "loss/base_kto": 0.4981650412082672,
+      "rewards/chosen": 0.024491310119628906,
+      "rewards/margins": 0.014406566939702848,
+      "rewards/rejected": 0.010084743179926059,
+      "stability/repetition_rate_mean": 0.2374526560306549,
+      "stability/response_length_mean": 35.625,
+      "stability/response_length_std": 31.057262420654297,
+      "stability/response_length_var": 964.5535888671875,
+      "stability/token_entropy_mean": 3.7021484375,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.6159020066261292,
+      "kl": 0.15860331058502197,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50396399.79746836,
+      "logits/rejected": -48982038.12345679,
+      "logps/chosen": -314.7759592563291,
+      "logps/rejected": -207.06117380401236,
+      "loss": 0.4982,
+      "loss/base_kto": 0.49839383363723755,
+      "rewards/chosen": 0.021742636644387546,
+      "rewards/margins": 0.013488207203948062,
+      "rewards/rejected": 0.008254429440439484,
+      "stability/repetition_rate_mean": 0.09133065491914749,
+      "stability/response_length_mean": 21.875,
+      "stability/response_length_std": 27.304826736450195,
+      "stability/response_length_var": 745.5535888671875,
+      "stability/token_entropy_mean": 4.076171875,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.6105856895446777,
+      "kl": 0.2014942169189453,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54708237.83783784,
+      "logits/rejected": -50898551.069767445,
+      "logps/chosen": -359.57057643581084,
+      "logps/rejected": -216.38419785610466,
+      "loss": 0.4981,
+      "loss/base_kto": 0.49866095185279846,
+      "rewards/chosen": 0.029960999617705476,
+      "rewards/margins": 0.010491025964843637,
+      "rewards/rejected": 0.01946997365286184,
+      "stability/repetition_rate_mean": 0.22838345170021057,
+      "stability/response_length_mean": 37.875,
+      "stability/response_length_std": 30.248672485351562,
+      "stability/response_length_var": 914.9821166992188,
+      "stability/token_entropy_mean": 4.185546875,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6785591244697571,
+      "kl": 0.22638840973377228,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59369200.578313254,
+      "logits/rejected": -53790852.98701299,
+      "logps/chosen": -399.54828689759034,
+      "logps/rejected": -209.8457538555195,
+      "loss": 0.4975,
+      "loss/base_kto": 0.49753338098526,
+      "rewards/chosen": 0.03542007021157138,
+      "rewards/margins": 0.01961183469632369,
+      "rewards/rejected": 0.015808235515247692,
+      "stability/repetition_rate_mean": 0.1897321343421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.349609375,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.6502929925918579,
+      "kl": 0.266899973154068,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57387927.13253012,
+      "logits/rejected": -53615536.20779221,
+      "logps/chosen": -378.2028661521084,
+      "logps/rejected": -205.8787413758117,
+      "loss": 0.4975,
+      "loss/base_kto": 0.49790263175964355,
+      "rewards/chosen": 0.03603898760784103,
+      "rewards/margins": 0.015099021306975758,
+      "rewards/rejected": 0.02093996630086527,
+      "stability/repetition_rate_mean": 0.31048768758773804,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.911476135253906,
+      "stability/response_length_var": 894.6964111328125,
+      "stability/token_entropy_mean": 4.357421875,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.6486251950263977,
+      "kl": 0.32395443320274353,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60606632.50632911,
+      "logits/rejected": -53405677.03703704,
+      "logps/chosen": -364.0806467563291,
+      "logps/rejected": -200.9707995756173,
+      "loss": 0.4965,
+      "loss/base_kto": 0.4963931143283844,
+      "rewards/chosen": 0.05118938035602811,
+      "rewards/margins": 0.0286137355568074,
+      "rewards/rejected": 0.02257564479922071,
+      "stability/repetition_rate_mean": 0.2749515473842621,
+      "stability/response_length_mean": 38.75,
+      "stability/response_length_std": 29.936838150024414,
+      "stability/response_length_var": 896.2142944335938,
+      "stability/token_entropy_mean": 4.966796875,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.6199327111244202,
+      "kl": 0.3587552011013031,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57752502.37908497,
+      "logits/rejected": -54366698.53892215,
+      "logps/chosen": -383.6620710784314,
+      "logps/rejected": -211.94790886976048,
+      "loss": 0.4959,
+      "loss/base_kto": 0.4960261285305023,
+      "rewards/chosen": 0.05828505871342678,
+      "rewards/margins": 0.03143982054748571,
+      "rewards/rejected": 0.026845238165941067,
+      "stability/repetition_rate_mean": 0.1723366528749466,
+      "stability/response_length_mean": 27.875,
+      "stability/response_length_std": 30.31943130493164,
+      "stability/response_length_var": 919.2678833007812,
+      "stability/token_entropy_mean": 4.640625,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.6118870377540588,
+      "kl": 0.3919747769832611,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60696351.56164384,
+      "logits/rejected": -55669807.08045977,
+      "logps/chosen": -416.76867508561645,
+      "logps/rejected": -216.09132543103448,
+      "loss": 0.4961,
+      "loss/base_kto": 0.4960012435913086,
+      "rewards/chosen": 0.061065817532474045,
+      "rewards/margins": 0.033479519105111495,
+      "rewards/rejected": 0.02758629842736255,
+      "stability/repetition_rate_mean": 0.15937499701976776,
+      "stability/response_length_mean": 15.875,
+      "stability/response_length_std": 24.584766387939453,
+      "stability/response_length_var": 604.4107055664062,
+      "stability/token_entropy_mean": 3.49609375,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.5947321653366089,
+      "kl": 0.45393142104148865,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57384028.49032258,
+      "logits/rejected": -53841212.50909091,
+      "logps/chosen": -391.96595262096776,
+      "logps/rejected": -183.6536221590909,
+      "loss": 0.4954,
+      "loss/base_kto": 0.49458155035972595,
+      "rewards/chosen": 0.06919794390278478,
+      "rewards/margins": 0.042669391818526664,
+      "rewards/rejected": 0.026528552084258108,
+      "stability/repetition_rate_mean": 0.25242185592651367,
+      "stability/response_length_mean": 43.25,
+      "stability/response_length_std": 29.620214462280273,
+      "stability/response_length_var": 877.3571166992188,
+      "stability/token_entropy_mean": 3.7138671875,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5941182971000671,
+      "kl": 0.501304566860199,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59851254.57668712,
+      "logits/rejected": -55000540.12738854,
+      "logps/chosen": -400.9979629984663,
+      "logps/rejected": -247.54722332802547,
+      "loss": 0.4945,
+      "loss/base_kto": 0.4937683045864105,
+      "rewards/chosen": 0.08181942898802962,
+      "rewards/margins": 0.049467831888482444,
+      "rewards/rejected": 0.03235159709954717,
+      "stability/repetition_rate_mean": 0.2230902761220932,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.936838150024414,
+      "stability/response_length_var": 896.2142944335938,
+      "stability/token_entropy_mean": 4.072265625,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.7290944457054138,
+      "kl": 0.4828576147556305,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57085976.82424243,
+      "logits/rejected": -50076896.61935484,
+      "logps/chosen": -366.3819128787879,
+      "logps/rejected": -189.91817036290323,
+      "loss": 0.4937,
+      "loss/base_kto": 0.4917580187320709,
+      "rewards/chosen": 0.08078298857717803,
+      "rewards/margins": 0.06915265418450736,
+      "rewards/rejected": 0.011630334392670662,
+      "stability/repetition_rate_mean": 0.2265625,
+      "stability/response_length_mean": 56.25,
+      "stability/response_length_std": 21.920310974121094,
+      "stability/response_length_var": 480.5,
+      "stability/token_entropy_mean": 5.150390625,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.6742141246795654,
+      "kl": 0.5725060105323792,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51148522.409638554,
+      "logits/rejected": -51662778.18181818,
+      "logps/chosen": -329.4726327183735,
+      "logps/rejected": -199.14629413555195,
+      "loss": 0.4932,
+      "loss/base_kto": 0.49420061707496643,
+      "rewards/chosen": 0.07991740789758153,
+      "rewards/margins": 0.04575571016309556,
+      "rewards/rejected": 0.03416169773448597,
+      "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": 5.2734375,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.6416594386100769,
+      "kl": 0.638058602809906,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55423500.04705882,
+      "logits/rejected": -50200978.77333333,
+      "logps/chosen": -396.80946691176473,
+      "logps/rejected": -207.36325520833333,
+      "loss": 0.493,
+      "loss/base_kto": 0.49001631140708923,
+      "rewards/chosen": 0.11249845168169807,
+      "rewards/margins": 0.0762304780997482,
+      "rewards/rejected": 0.03626797358194987,
+      "stability/repetition_rate_mean": 0.23066024482250214,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.193626403808594,
+      "stability/response_length_var": 852.2678833007812,
+      "stability/token_entropy_mean": 4.6015625,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.6336647272109985,
+      "kl": 0.6403493881225586,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54021613.92941176,
+      "logits/rejected": -49351734.61333334,
+      "logps/chosen": -360.1190487132353,
+      "logps/rejected": -191.01639322916665,
+      "loss": 0.4925,
+      "loss/base_kto": 0.4930504858493805,
+      "rewards/chosen": 0.0870250758002786,
+      "rewards/margins": 0.058103303161321895,
+      "rewards/rejected": 0.028921772638956705,
+      "stability/repetition_rate_mean": 0.3113190531730652,
+      "stability/response_length_mean": 61.375,
+      "stability/response_length_std": 7.424621105194092,
+      "stability/response_length_var": 55.125,
+      "stability/token_entropy_mean": 4.529296875,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.6168467402458191,
+      "kl": 0.7161628603935242,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55242325.333333336,
+      "logits/rejected": -48961121.684210524,
+      "logps/chosen": -415.55738467261904,
+      "logps/rejected": -186.4566072162829,
+      "loss": 0.4911,
+      "loss/base_kto": 0.4885805547237396,
+      "rewards/chosen": 0.13379353568667457,
+      "rewards/margins": 0.08971994144277165,
+      "rewards/rejected": 0.04407359424390291,
+      "stability/repetition_rate_mean": 0.30851560831069946,
+      "stability/response_length_mean": 32.5,
+      "stability/response_length_std": 30.528675079345703,
+      "stability/response_length_var": 932.0,
+      "stability/token_entropy_mean": 4.474609375,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.7232491374015808,
+      "kl": 0.7498390078544617,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54396980.96551724,
+      "logits/rejected": -49794756.38356164,
+      "logps/chosen": -369.75480423850576,
+      "logps/rejected": -212.1316486515411,
+      "loss": 0.4894,
+      "loss/base_kto": 0.49128732085227966,
+      "rewards/chosen": 0.11619376040053093,
+      "rewards/margins": 0.07014617738226721,
+      "rewards/rejected": 0.04604758301826373,
+      "stability/repetition_rate_mean": 0.2411106526851654,
+      "stability/response_length_mean": 22.5,
+      "stability/response_length_std": 28.33977508544922,
+      "stability/response_length_var": 803.1428833007812,
+      "stability/token_entropy_mean": 4.86328125,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.6223028302192688,
+      "kl": 0.9178180694580078,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59141307.834319524,
+      "logits/rejected": -52591744.84768212,
+      "logps/chosen": -439.69304733727813,
+      "logps/rejected": -190.61025455298014,
+      "loss": 0.4881,
+      "loss/base_kto": 0.48755979537963867,
+      "rewards/chosen": 0.16488214075212648,
+      "rewards/margins": 0.10012487574960424,
+      "rewards/rejected": 0.06475726500252224,
+      "stability/repetition_rate_mean": 0.3434279263019562,
+      "stability/response_length_mean": 44.125,
+      "stability/response_length_std": 28.43256378173828,
+      "stability/response_length_var": 808.4107055664062,
+      "stability/token_entropy_mean": 3.70166015625,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.7227226495742798,
+      "kl": 0.937430202960968,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56671336.86746988,
+      "logits/rejected": -48043027.948051944,
+      "logps/chosen": -354.8946018448795,
+      "logps/rejected": -194.3057528409091,
+      "loss": 0.4875,
+      "loss/base_kto": 0.48688632249832153,
+      "rewards/chosen": 0.154259325510048,
+      "rewards/margins": 0.10408077600360047,
+      "rewards/rejected": 0.05017854950644753,
+      "stability/repetition_rate_mean": 0.18430830538272858,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 25.773740768432617,
+      "stability/response_length_var": 664.2857055664062,
+      "stability/token_entropy_mean": 4.8564453125,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.6434662938117981,
+      "kl": 1.026900291442871,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52659602.06134969,
+      "logits/rejected": -47986926.06369427,
+      "logps/chosen": -374.42537384969324,
+      "logps/rejected": -210.55354299363057,
+      "loss": 0.4865,
+      "loss/base_kto": 0.4877110421657562,
+      "rewards/chosen": 0.17204565504577263,
+      "rewards/margins": 0.09757829369276705,
+      "rewards/rejected": 0.07446736135300558,
+      "stability/repetition_rate_mean": 0.16679687798023224,
+      "stability/response_length_mean": 34.75,
+      "stability/response_length_std": 31.54928207397461,
+      "stability/response_length_var": 995.3571166992188,
+      "stability/token_entropy_mean": 3.541015625,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.6501672863960266,
+      "kl": 1.079397439956665,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54617631.60493827,
+      "logits/rejected": -49526434.025316454,
+      "logps/chosen": -310.3054108796296,
+      "logps/rejected": -196.56382268591773,
+      "loss": 0.4857,
+      "loss/base_kto": 0.4866374433040619,
+      "rewards/chosen": 0.16959618933406878,
+      "rewards/margins": 0.12206393268857642,
+      "rewards/rejected": 0.04753225664549236,
+      "stability/repetition_rate_mean": 0.205078125,
+      "stability/response_length_mean": 32.875,
+      "stability/response_length_std": 33.30996322631836,
+      "stability/response_length_var": 1109.5535888671875,
+      "stability/token_entropy_mean": 3.732421875,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.6943851113319397,
+      "kl": 1.1196650266647339,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53094718.8502994,
+      "logits/rejected": -49691517.49019608,
+      "logps/chosen": -390.31998502994014,
+      "logps/rejected": -210.46075878267973,
+      "loss": 0.484,
+      "loss/base_kto": 0.48342210054397583,
+      "rewards/chosen": 0.2085319564727966,
+      "rewards/margins": 0.13474922263035827,
+      "rewards/rejected": 0.07378273384243834,
+      "stability/repetition_rate_mean": 0.18919041752815247,
+      "stability/response_length_mean": 18.625,
+      "stability/response_length_std": 26.52189064025879,
+      "stability/response_length_var": 703.4107055664062,
+      "stability/token_entropy_mean": 4.5703125,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.6349285244941711,
+      "kl": 1.1486577987670898,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60025334.090322584,
+      "logits/rejected": -53795740.7030303,
+      "logps/chosen": -390.8627268145161,
+      "logps/rejected": -197.078125,
+      "loss": 0.4843,
+      "loss/base_kto": 0.4863618016242981,
+      "rewards/chosen": 0.1922200725924584,
+      "rewards/margins": 0.10809968629540585,
+      "rewards/rejected": 0.08412038629705255,
+      "stability/repetition_rate_mean": 0.341796875,
+      "stability/response_length_mean": 48.5,
+      "stability/response_length_std": 28.705400466918945,
+      "stability/response_length_var": 824.0,
+      "stability/token_entropy_mean": 5.171875,
+      "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_simple/checkpoint-196/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cca02ae9ad4d71243fc18608eed865018fb06c94becd454c7ce13581852dd3e2
+size 8017

mistral7b_kto_simple/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_simple/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_simple/tokenizer.model

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

mistral7b_kto_simple/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_simple/train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.4956757848968311,
+    "train_runtime": 3145.5892,
+    "train_samples_per_second": 31.791,
+    "train_steps_per_second": 0.062
+}

mistral7b_kto_simple/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": "0:01:17", "remaining_time": "0:49:34", "rewards/chosen": 0.0013566404385002035, "logps/chosen": -439.03207236842104, "logits/chosen": -59718379.78947368, "rewards/rejected": -0.001202781108163652, "logps/rejected": -198.4214332217262, "logits/rejected": -50413504.0, "rewards/margins": 0.0025594215466638558, "kl": 0.006305641960352659}
+{"current_steps": 10, "total_steps": 196, "loss": 0.5001, "lr": 9e-08, "epoch": 0.05118362124120281, "percentage": 5.1, "elapsed_time": "0:02:38", "remaining_time": "0:49:09", "rewards/chosen": 0.002450463833746972, "logps/chosen": -429.9140625, "logits/chosen": -59308962.90909091, "rewards/rejected": -2.95383204896766e-05, "logps/rejected": -192.86823465737953, "logits/rejected": -50030080.0, "rewards/margins": 0.0024800021542366484, "kl": 0.028230074793100357, "stability/response_length_mean": 44.25, "stability/response_length_std": 28.13614845275879, "stability/response_length_var": 791.6428833007812, "stability/token_entropy_mean": 3.4755859375, "stability/repetition_rate_mean": 0.3664148449897766}
+{"current_steps": 15, "total_steps": 196, "loss": 0.4998, "lr": 1.4e-07, "epoch": 0.07677543186180422, "percentage": 7.65, "elapsed_time": "0:03:58", "remaining_time": "0:47:54", "rewards/chosen": 0.001057231787479285, "logps/chosen": -404.1202651515151, "logits/chosen": -56221720.82424243, "rewards/rejected": 0.0014020606394736998, "logps/rejected": -200.68004032258065, "logits/rejected": -51918762.11612903, "rewards/margins": -0.00034482885199441484, "kl": 0.012302148155868053, "stability/response_length_mean": 29.375, "stability/response_length_std": 29.92341423034668, "stability/response_length_var": 895.4107055664062, "stability/token_entropy_mean": 3.7001953125, "stability/repetition_rate_mean": 0.2929687201976776}
+{"current_steps": 20, "total_steps": 196, "loss": 0.5004, "lr": 1.8999999999999998e-07, "epoch": 0.10236724248240563, "percentage": 10.2, "elapsed_time": "0:05:19", "remaining_time": "0:46:51", "rewards/chosen": 0.00033793365582823753, "logps/chosen": -343.6933349609375, "logits/chosen": -51663750.4, "rewards/rejected": 0.0004558594897389412, "logps/rejected": -225.6587890625, "logits/rejected": -48055072.0, "rewards/margins": -0.00011792583391070368, "kl": 0.016423773020505905, "stability/response_length_mean": 44.0, "stability/response_length_std": 28.0611572265625, "stability/response_length_var": 787.4285888671875, "stability/token_entropy_mean": 4.42578125, "stability/repetition_rate_mean": 0.2155257910490036}
+{"current_steps": 25, "total_steps": 196, "loss": 0.5001, "lr": 2.4e-07, "epoch": 0.12795905310300704, "percentage": 12.76, "elapsed_time": "0:06:39", "remaining_time": "0:45:30", "rewards/chosen": 0.0016139640890318772, "logps/chosen": -378.4753232758621, "logits/chosen": -56465217.32413793, "rewards/rejected": 0.001752270289829799, "logps/rejected": -202.7753125, "logits/rejected": -52019422.35428572, "rewards/margins": -0.00013830620079792184, "kl": 0.01958128809928894, "stability/response_length_mean": 50.25, "stability/response_length_std": 25.549671173095703, "stability/response_length_var": 652.7857055664062, "stability/token_entropy_mean": 4.5234375, "stability/repetition_rate_mean": 0.19711539149284363}
+{"current_steps": 30, "total_steps": 196, "loss": 0.5002, "lr": 2.9e-07, "epoch": 0.15355086372360843, "percentage": 15.31, "elapsed_time": "0:07:59", "remaining_time": "0:44:11", "rewards/chosen": 0.0019035963404098613, "logps/chosen": -468.5621854026846, "logits/chosen": -57728749.10067114, "rewards/rejected": 0.0009252666032802292, "logps/rejected": -198.23862390350877, "logits/rejected": -51514302.12865497, "rewards/margins": 0.000978329737129632, "kl": 0.020890694111585617, "stability/response_length_mean": 43.0, "stability/response_length_std": 29.602123260498047, "stability/response_length_var": 876.2857055664062, "stability/token_entropy_mean": 4.568359375, "stability/repetition_rate_mean": 0.1525297611951828}
+{"current_steps": 35, "total_steps": 196, "loss": 0.4998, "lr": 3.4000000000000003e-07, "epoch": 0.17914267434420986, "percentage": 17.86, "elapsed_time": "0:09:20", "remaining_time": "0:42:57", "rewards/chosen": 0.0014279391380805004, "logps/chosen": -385.922245846519, "logits/chosen": -57633195.74683544, "rewards/rejected": 0.0016860373226212867, "logps/rejected": -195.10961612654322, "logits/rejected": -53898088.2962963, "rewards/margins": -0.0002580981845407863, "kl": 0.031000053510069847, "stability/response_length_mean": 56.875, "stability/response_length_std": 20.152544021606445, "stability/response_length_var": 406.125, "stability/token_entropy_mean": 5.474609375, "stability/repetition_rate_mean": 0.3030133843421936}
+{"current_steps": 40, "total_steps": 196, "loss": 0.4998, "lr": 3.8999999999999997e-07, "epoch": 0.20473448496481125, "percentage": 20.41, "elapsed_time": "0:10:40", "remaining_time": "0:41:39", "rewards/chosen": 0.003170560672879219, "logps/chosen": -421.8708984375, "logits/chosen": -56631648.0, "rewards/rejected": 0.004156828671693802, "logps/rejected": -219.26767578125, "logits/rejected": -50232096.0, "rewards/margins": -0.0009862679988145826, "kl": 0.0245906300842762, "stability/response_length_mean": 35.0, "stability/response_length_std": 31.518701553344727, "stability/response_length_var": 993.4285888671875, "stability/token_entropy_mean": 4.7880859375, "stability/repetition_rate_mean": 0.15522204339504242}
+{"current_steps": 45, "total_steps": 196, "loss": 0.4998, "lr": 4.3999999999999997e-07, "epoch": 0.23032629558541268, "percentage": 22.96, "elapsed_time": "0:12:01", "remaining_time": "0:40:20", "rewards/chosen": 0.004177153910566497, "logps/chosen": -352.2867030201342, "logits/chosen": -52691122.684563756, "rewards/rejected": 0.003994271769161113, "logps/rejected": -243.83422423245614, "logits/rejected": -50773704.60818713, "rewards/margins": 0.0001828821414053836, "kl": 0.038881849497556686, "stability/response_length_mean": 22.375, "stability/response_length_std": 26.52189064025879, "stability/response_length_var": 703.4107055664062, "stability/token_entropy_mean": 4.9140625, "stability/repetition_rate_mean": 0.0859469324350357}
+{"current_steps": 50, "total_steps": 196, "loss": 0.4999, "lr": 4.9e-07, "epoch": 0.2559181062060141, "percentage": 25.51, "elapsed_time": "0:13:21", "remaining_time": "0:38:59", "rewards/chosen": 0.007963991005148663, "logps/chosen": -375.11949454697987, "logits/chosen": -54047022.389261745, "rewards/rejected": 0.004319407786542212, "logps/rejected": -180.19741410818713, "logits/rejected": -48305211.883040935, "rewards/margins": 0.0036445832186064505, "kl": 0.04716196283698082, "stability/response_length_mean": 55.375, "stability/response_length_std": 16.361431121826172, "stability/response_length_var": 267.6964416503906, "stability/token_entropy_mean": 3.81640625, "stability/repetition_rate_mean": 0.4108419120311737}
+{"current_steps": 55, "total_steps": 196, "loss": 0.4996, "lr": 5.4e-07, "epoch": 0.28150991682661547, "percentage": 28.06, "elapsed_time": "0:14:40", "remaining_time": "0:37:38", "rewards/chosen": 0.003263072937912082, "logps/chosen": -344.4793769409938, "logits/chosen": -56946357.26708075, "rewards/rejected": 0.004255936955505947, "logps/rejected": -230.40548840408806, "logits/rejected": -49559825.710691825, "rewards/margins": -0.0009928640175938652, "kl": 0.0537404902279377, "stability/response_length_mean": 24.5, "stability/response_length_std": 32.710853576660156, "stability/response_length_var": 1070.0, "stability/token_entropy_mean": 4.41015625, "stability/repetition_rate_mean": 0.107421875}
+{"current_steps": 60, "total_steps": 196, "loss": 0.4997, "lr": 5.9e-07, "epoch": 0.30710172744721687, "percentage": 30.61, "elapsed_time": "0:16:02", "remaining_time": "0:36:20", "rewards/chosen": 0.008416923609646883, "logps/chosen": -359.6366477272727, "logits/chosen": -52670401.93939394, "rewards/rejected": 0.005569554144336331, "logps/rejected": -178.17121975806452, "logits/rejected": -47818137.6, "rewards/margins": 0.002847369465310552, "kl": 0.05833861976861954, "stability/response_length_mean": 44.625, "stability/response_length_std": 27.8820858001709, "stability/response_length_var": 777.4107055664062, "stability/token_entropy_mean": 4.666015625, "stability/repetition_rate_mean": 0.18834353983402252}
+{"current_steps": 65, "total_steps": 196, "loss": 0.4996, "lr": 6.4e-07, "epoch": 0.3326935380678183, "percentage": 33.16, "elapsed_time": "0:17:22", "remaining_time": "0:35:01", "rewards/chosen": 0.014990917328865297, "logps/chosen": -381.796875, "logits/chosen": -57552697.80645161, "rewards/rejected": 0.005205387780160615, "logps/rejected": -183.53030303030303, "logits/rejected": -47262831.70909091, "rewards/margins": 0.009785529548704684, "kl": 0.07869795709848404, "stability/response_length_mean": 21.125, "stability/response_length_std": 28.027729034423828, "stability/response_length_var": 785.5535888671875, "stability/token_entropy_mean": 5.58984375, "stability/repetition_rate_mean": 0.19731052219867706}
+{"current_steps": 70, "total_steps": 196, "loss": 0.4992, "lr": 6.9e-07, "epoch": 0.3582853486884197, "percentage": 35.71, "elapsed_time": "0:18:42", "remaining_time": "0:33:40", "rewards/chosen": 0.012032072005733366, "logps/chosen": -379.00811491935485, "logits/chosen": -52767777.03225806, "rewards/rejected": 0.008934669783621124, "logps/rejected": -220.3328125, "logits/rejected": -49031406.93333333, "rewards/margins": 0.0030974022221122426, "kl": 0.08887721598148346, "stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.7265625, "stability/repetition_rate_mean": 0.185546875}
+{"current_steps": 75, "total_steps": 196, "loss": 0.4993, "lr": 7.4e-07, "epoch": 0.3838771593090211, "percentage": 38.27, "elapsed_time": "0:20:03", "remaining_time": "0:32:21", "rewards/chosen": 0.016490372451576026, "logps/chosen": -314.1058910472973, "logits/chosen": -49499744.86486486, "rewards/rejected": 0.006668017354122428, "logps/rejected": -228.81817768895348, "logits/rejected": -47894179.72093023, "rewards/margins": 0.009822355097453599, "kl": 0.09266681969165802, "stability/response_length_mean": 22.75, "stability/response_length_std": 31.07479476928711, "stability/response_length_var": 965.6428833007812, "stability/token_entropy_mean": 2.146484375, "stability/repetition_rate_mean": 0.08308292925357819}
+{"current_steps": 80, "total_steps": 196, "loss": 0.4991, "lr": 7.9e-07, "epoch": 0.4094689699296225, "percentage": 40.82, "elapsed_time": "0:21:24", "remaining_time": "0:31:01", "rewards/chosen": 0.017660609933714204, "logps/chosen": -403.85350269039736, "logits/chosen": -58629587.9205298, "rewards/rejected": 0.009116629877034023, "logps/rejected": -200.96463572485206, "logits/rejected": -53451757.82248521, "rewards/margins": 0.008543980056680181, "kl": 0.12336266040802002, "stability/response_length_mean": 39.0, "stability/response_length_std": 29.159658432006836, "stability/response_length_var": 850.2857055664062, "stability/token_entropy_mean": 5.22265625, "stability/repetition_rate_mean": 0.18541666865348816}
+{"current_steps": 85, "total_steps": 196, "loss": 0.4988, "lr": 8.399999999999999e-07, "epoch": 0.4350607805502239, "percentage": 43.37, "elapsed_time": "0:22:44", "remaining_time": "0:29:41", "rewards/chosen": 0.022868017996511152, "logps/chosen": -358.1038306451613, "logits/chosen": -51481851.04516129, "rewards/rejected": 0.009824119914661754, "logps/rejected": -230.3265625, "logits/rejected": -47137016.24242424, "rewards/margins": 0.013043898081849399, "kl": 0.125388041138649, "stability/response_length_mean": 27.375, "stability/response_length_std": 30.39707374572754, "stability/response_length_var": 923.9821166992188, "stability/token_entropy_mean": 4.92578125, "stability/repetition_rate_mean": 0.12302795797586441}
+{"current_steps": 90, "total_steps": 196, "loss": 0.4984, "lr": 8.9e-07, "epoch": 0.46065259117082535, "percentage": 45.92, "elapsed_time": "0:24:05", "remaining_time": "0:28:21", "rewards/chosen": 0.024491310119628906, "logps/chosen": -380.54447115384613, "logits/chosen": -58409432.615384616, "rewards/rejected": 0.010084743179926059, "logps/rejected": -187.12595274390245, "logits/rejected": -50583776.780487806, "rewards/margins": 0.014406566939702848, "kl": 0.17190246284008026, "stability/response_length_mean": 35.625, "stability/response_length_std": 31.057262420654297, "stability/response_length_var": 964.5535888671875, "stability/token_entropy_mean": 3.7021484375, "stability/repetition_rate_mean": 0.2374526560306549}
+{"current_steps": 95, "total_steps": 196, "loss": 0.4982, "lr": 9.399999999999999e-07, "epoch": 0.48624440179142675, "percentage": 48.47, "elapsed_time": "0:25:24", "remaining_time": "0:27:01", "rewards/chosen": 0.021742636644387546, "logps/chosen": -314.7759592563291, "logits/chosen": -50396399.79746836, "rewards/rejected": 0.008254429440439484, "logps/rejected": -207.06117380401236, "logits/rejected": -48982038.12345679, "rewards/margins": 0.013488207203948062, "kl": 0.15860331058502197, "stability/response_length_mean": 21.875, "stability/response_length_std": 27.304826736450195, "stability/response_length_var": 745.5535888671875, "stability/token_entropy_mean": 4.076171875, "stability/repetition_rate_mean": 0.09133065491914749}
+{"current_steps": 100, "total_steps": 196, "loss": 0.4981, "lr": 9.9e-07, "epoch": 0.5118362124120281, "percentage": 51.02, "elapsed_time": "0:26:45", "remaining_time": "0:25:41", "rewards/chosen": 0.029960999617705476, "logps/chosen": -359.57057643581084, "logits/chosen": -54708237.83783784, "rewards/rejected": 0.01946997365286184, "logps/rejected": -216.38419785610466, "logits/rejected": -50898551.069767445, "rewards/margins": 0.010491025964843637, "kl": 0.2014942169189453, "stability/response_length_mean": 37.875, "stability/response_length_std": 30.248672485351562, "stability/response_length_var": 914.9821166992188, "stability/token_entropy_mean": 4.185546875, "stability/repetition_rate_mean": 0.22838345170021057}
+{"current_steps": 105, "total_steps": 196, "loss": 0.4975, "lr": 9.999815554351224e-07, "epoch": 0.5374280230326296, "percentage": 53.57, "elapsed_time": "0:28:05", "remaining_time": "0:24:21", "rewards/chosen": 0.03542007021157138, "logps/chosen": -399.54828689759034, "logits/chosen": -59369200.578313254, "rewards/rejected": 0.015808235515247692, "logps/rejected": -209.8457538555195, "logits/rejected": -53790852.98701299, "rewards/margins": 0.01961183469632369, "kl": 0.22638840973377228, "stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.349609375, "stability/repetition_rate_mean": 0.1897321343421936}
+{"current_steps": 110, "total_steps": 196, "loss": 0.4975, "lr": 9.999066267225444e-07, "epoch": 0.5630198336532309, "percentage": 56.12, "elapsed_time": "0:29:27", "remaining_time": "0:23:01", "rewards/chosen": 0.03603898760784103, "logps/chosen": -378.2028661521084, "logits/chosen": -57387927.13253012, "rewards/rejected": 0.02093996630086527, "logps/rejected": -205.8787413758117, "logits/rejected": -53615536.20779221, "rewards/margins": 0.015099021306975758, "kl": 0.266899973154068, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.911476135253906, "stability/response_length_var": 894.6964111328125, "stability/token_entropy_mean": 4.357421875, "stability/repetition_rate_mean": 0.31048768758773804}
+{"current_steps": 115, "total_steps": 196, "loss": 0.4965, "lr": 9.997740697079592e-07, "epoch": 0.5886116442738324, "percentage": 58.67, "elapsed_time": "0:30:48", "remaining_time": "0:21:41", "rewards/chosen": 0.05118938035602811, "logps/chosen": -364.0806467563291, "logits/chosen": -60606632.50632911, "rewards/rejected": 0.02257564479922071, "logps/rejected": -200.9707995756173, "logits/rejected": -53405677.03703704, "rewards/margins": 0.0286137355568074, "kl": 0.32395443320274353, "stability/response_length_mean": 38.75, "stability/response_length_std": 29.936838150024414, "stability/response_length_var": 896.2142944335938, "stability/token_entropy_mean": 4.966796875, "stability/repetition_rate_mean": 0.2749515473842621}
+{"current_steps": 120, "total_steps": 196, "loss": 0.4959, "lr": 9.995838996722914e-07, "epoch": 0.6142034548944337, "percentage": 61.22, "elapsed_time": "0:32:07", "remaining_time": "0:20:20", "rewards/chosen": 0.05828505871342678, "logps/chosen": -383.6620710784314, "logits/chosen": -57752502.37908497, "rewards/rejected": 0.026845238165941067, "logps/rejected": -211.94790886976048, "logits/rejected": -54366698.53892215, "rewards/margins": 0.03143982054748571, "kl": 0.3587552011013031, "stability/response_length_mean": 27.875, "stability/response_length_std": 30.31943130493164, "stability/response_length_var": 919.2678833007812, "stability/token_entropy_mean": 4.640625, "stability/repetition_rate_mean": 0.1723366528749466}
+{"current_steps": 125, "total_steps": 196, "loss": 0.4961, "lr": 9.993361385379874e-07, "epoch": 0.6397952655150352, "percentage": 63.78, "elapsed_time": "0:33:28", "remaining_time": "0:19:00", "rewards/chosen": 0.061065817532474045, "logps/chosen": -416.76867508561645, "logits/chosen": -60696351.56164384, "rewards/rejected": 0.02758629842736255, "logps/rejected": -216.09132543103448, "logits/rejected": -55669807.08045977, "rewards/margins": 0.033479519105111495, "kl": 0.3919747769832611, "stability/response_length_mean": 15.875, "stability/response_length_std": 24.584766387939453, "stability/response_length_var": 604.4107055664062, "stability/token_entropy_mean": 3.49609375, "stability/repetition_rate_mean": 0.15937499701976776}
+{"current_steps": 130, "total_steps": 196, "loss": 0.4954, "lr": 9.99030814866488e-07, "epoch": 0.6653870761356366, "percentage": 66.33, "elapsed_time": "0:34:48", "remaining_time": "0:17:40", "rewards/chosen": 0.06919794390278478, "logps/chosen": -391.96595262096776, "logits/chosen": -57384028.49032258, "rewards/rejected": 0.026528552084258108, "logps/rejected": -183.6536221590909, "logits/rejected": -53841212.50909091, "rewards/margins": 0.042669391818526664, "kl": 0.45393142104148865, "stability/response_length_mean": 43.25, "stability/response_length_std": 29.620214462280273, "stability/response_length_var": 877.3571166992188, "stability/token_entropy_mean": 3.7138671875, "stability/repetition_rate_mean": 0.25242185592651367}
+{"current_steps": 135, "total_steps": 196, "loss": 0.4945, "lr": 9.986679638549358e-07, "epoch": 0.690978886756238, "percentage": 68.88, "elapsed_time": "0:36:07", "remaining_time": "0:16:19", "rewards/chosen": 0.08181942898802962, "logps/chosen": -400.9979629984663, "logits/chosen": -59851254.57668712, "rewards/rejected": 0.03235159709954717, "logps/rejected": -247.54722332802547, "logits/rejected": -55000540.12738854, "rewards/margins": 0.049467831888482444, "kl": 0.501304566860199, "stability/response_length_mean": 29.25, "stability/response_length_std": 29.936838150024414, "stability/response_length_var": 896.2142944335938, "stability/token_entropy_mean": 4.072265625, "stability/repetition_rate_mean": 0.2230902761220932}
+{"current_steps": 140, "total_steps": 196, "loss": 0.4937, "lr": 9.982476273321175e-07, "epoch": 0.7165706973768394, "percentage": 71.43, "elapsed_time": "0:37:28", "remaining_time": "0:14:59", "rewards/chosen": 0.08078298857717803, "logps/chosen": -366.3819128787879, "logits/chosen": -57085976.82424243, "rewards/rejected": 0.011630334392670662, "logps/rejected": -189.91817036290323, "logits/rejected": -50076896.61935484, "rewards/margins": 0.06915265418450736, "kl": 0.4828576147556305, "stability/response_length_mean": 56.25, "stability/response_length_std": 21.920310974121094, "stability/response_length_var": 480.5, "stability/token_entropy_mean": 5.150390625, "stability/repetition_rate_mean": 0.2265625}
+{"current_steps": 145, "total_steps": 196, "loss": 0.4932, "lr": 9.977698537536417e-07, "epoch": 0.7421625079974408, "percentage": 73.98, "elapsed_time": "0:38:48", "remaining_time": "0:13:38", "rewards/chosen": 0.07991740789758153, "logps/chosen": -329.4726327183735, "logits/chosen": -51148522.409638554, "rewards/rejected": 0.03416169773448597, "logps/rejected": -199.14629413555195, "logits/rejected": -51662778.18181818, "rewards/margins": 0.04575571016309556, "kl": 0.5725060105323792, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.73423957824707, "stability/response_length_var": 884.125, "stability/token_entropy_mean": 5.2734375, "stability/repetition_rate_mean": 0.25424888730049133}
+{"current_steps": 150, "total_steps": 196, "loss": 0.493, "lr": 9.972346981963546e-07, "epoch": 0.7677543186180422, "percentage": 76.53, "elapsed_time": "0:40:08", "remaining_time": "0:12:18", "rewards/chosen": 0.11249845168169807, "logps/chosen": -396.80946691176473, "logits/chosen": -55423500.04705882, "rewards/rejected": 0.03626797358194987, "logps/rejected": -207.36325520833333, "logits/rejected": -50200978.77333333, "rewards/margins": 0.0762304780997482, "kl": 0.638058602809906, "stability/response_length_mean": 29.375, "stability/response_length_std": 29.193626403808594, "stability/response_length_var": 852.2678833007812, "stability/token_entropy_mean": 4.6015625, "stability/repetition_rate_mean": 0.23066024482250214}
+{"current_steps": 155, "total_steps": 196, "loss": 0.4925, "lr": 9.966422223519885e-07, "epoch": 0.7933461292386437, "percentage": 79.08, "elapsed_time": "0:41:30", "remaining_time": "0:10:58", "rewards/chosen": 0.0870250758002786, "logps/chosen": -360.1190487132353, "logits/chosen": -54021613.92941176, "rewards/rejected": 0.028921772638956705, "logps/rejected": -191.01639322916665, "logits/rejected": -49351734.61333334, "rewards/margins": 0.058103303161321895, "kl": 0.6403493881225586, "stability/response_length_mean": 61.375, "stability/response_length_std": 7.424621105194092, "stability/response_length_var": 55.125, "stability/token_entropy_mean": 4.529296875, "stability/repetition_rate_mean": 0.3113190531730652}
+{"current_steps": 160, "total_steps": 196, "loss": 0.4911, "lr": 9.959924945200523e-07, "epoch": 0.818937939859245, "percentage": 81.63, "elapsed_time": "0:42:50", "remaining_time": "0:09:38", "rewards/chosen": 0.13379353568667457, "logps/chosen": -415.55738467261904, "logits/chosen": -55242325.333333336, "rewards/rejected": 0.04407359424390291, "logps/rejected": -186.4566072162829, "logits/rejected": -48961121.684210524, "rewards/margins": 0.08971994144277165, "kl": 0.7161628603935242, "stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.474609375, "stability/repetition_rate_mean": 0.30851560831069946}
+{"current_steps": 165, "total_steps": 196, "loss": 0.4894, "lr": 9.952855895999568e-07, "epoch": 0.8445297504798465, "percentage": 84.18, "elapsed_time": "0:44:10", "remaining_time": "0:08:17", "rewards/chosen": 0.11619376040053093, "logps/chosen": -369.75480423850576, "logits/chosen": -54396980.96551724, "rewards/rejected": 0.04604758301826373, "logps/rejected": -212.1316486515411, "logits/rejected": -49794756.38356164, "rewards/margins": 0.07014617738226721, "kl": 0.7498390078544617, "stability/response_length_mean": 22.5, "stability/response_length_std": 28.33977508544922, "stability/response_length_var": 803.1428833007812, "stability/token_entropy_mean": 4.86328125, "stability/repetition_rate_mean": 0.2411106526851654}
+{"current_steps": 170, "total_steps": 196, "loss": 0.4881, "lr": 9.9452158908238e-07, "epoch": 0.8701215611004478, "percentage": 86.73, "elapsed_time": "0:45:31", "remaining_time": "0:06:57", "rewards/chosen": 0.16488214075212648, "logps/chosen": -439.69304733727813, "logits/chosen": -59141307.834319524, "rewards/rejected": 0.06475726500252224, "logps/rejected": -190.61025455298014, "logits/rejected": -52591744.84768212, "rewards/margins": 0.10012487574960424, "kl": 0.9178180694580078, "stability/response_length_mean": 44.125, "stability/response_length_std": 28.43256378173828, "stability/response_length_var": 808.4107055664062, "stability/token_entropy_mean": 3.70166015625, "stability/repetition_rate_mean": 0.3434279263019562}
+{"current_steps": 175, "total_steps": 196, "loss": 0.4875, "lr": 9.937005810398745e-07, "epoch": 0.8957133717210493, "percentage": 89.29, "elapsed_time": "0:46:51", "remaining_time": "0:05:37", "rewards/chosen": 0.154259325510048, "logps/chosen": -354.8946018448795, "logits/chosen": -56671336.86746988, "rewards/rejected": 0.05017854950644753, "logps/rejected": -194.3057528409091, "logits/rejected": -48043027.948051944, "rewards/margins": 0.10408077600360047, "kl": 0.937430202960968, "stability/response_length_mean": 27.0, "stability/response_length_std": 25.773740768432617, "stability/response_length_var": 664.2857055664062, "stability/token_entropy_mean": 4.8564453125, "stability/repetition_rate_mean": 0.18430830538272858}
+{"current_steps": 180, "total_steps": 196, "loss": 0.4865, "lr": 9.928226601167138e-07, "epoch": 0.9213051823416507, "percentage": 91.84, "elapsed_time": "0:48:12", "remaining_time": "0:04:17", "rewards/chosen": 0.17204565504577263, "logps/chosen": -374.42537384969324, "logits/chosen": -52659602.06134969, "rewards/rejected": 0.07446736135300558, "logps/rejected": -210.55354299363057, "logits/rejected": -47986926.06369427, "rewards/margins": 0.09757829369276705, "kl": 1.026900291442871, "stability/response_length_mean": 34.75, "stability/response_length_std": 31.54928207397461, "stability/response_length_var": 995.3571166992188, "stability/token_entropy_mean": 3.541015625, "stability/repetition_rate_mean": 0.16679687798023224}
+{"current_steps": 185, "total_steps": 196, "loss": 0.4857, "lr": 9.918879275179817e-07, "epoch": 0.946896992962252, "percentage": 94.39, "elapsed_time": "0:49:31", "remaining_time": "0:02:56", "rewards/chosen": 0.16959618933406878, "logps/chosen": -310.3054108796296, "logits/chosen": -54617631.60493827, "rewards/rejected": 0.04753225664549236, "logps/rejected": -196.56382268591773, "logits/rejected": -49526434.025316454, "rewards/margins": 0.12206393268857642, "kl": 1.079397439956665, "stability/response_length_mean": 32.875, "stability/response_length_std": 33.30996322631836, "stability/response_length_var": 1109.5535888671875, "stability/token_entropy_mean": 3.732421875, "stability/repetition_rate_mean": 0.205078125}
+{"current_steps": 190, "total_steps": 196, "loss": 0.484, "lr": 9.90896490997906e-07, "epoch": 0.9724888035828535, "percentage": 96.94, "elapsed_time": "0:50:51", "remaining_time": "0:01:36", "rewards/chosen": 0.2085319564727966, "logps/chosen": -390.31998502994014, "logits/chosen": -53094718.8502994, "rewards/rejected": 0.07378273384243834, "logps/rejected": -210.46075878267973, "logits/rejected": -49691517.49019608, "rewards/margins": 0.13474922263035827, "kl": 1.1196650266647339, "stability/response_length_mean": 18.625, "stability/response_length_std": 26.52189064025879, "stability/response_length_var": 703.4107055664062, "stability/token_entropy_mean": 4.5703125, "stability/repetition_rate_mean": 0.18919041752815247}
+{"current_steps": 195, "total_steps": 196, "loss": 0.4843, "lr": 9.898484648474362e-07, "epoch": 0.9980806142034548, "percentage": 99.49, "elapsed_time": "0:52:12", "remaining_time": "0:00:16", "rewards/chosen": 0.1922200725924584, "logps/chosen": -390.8627268145161, "logits/chosen": -60025334.090322584, "rewards/rejected": 0.08412038629705255, "logps/rejected": -197.078125, "logits/rejected": -53795740.7030303, "rewards/margins": 0.10809968629540585, "kl": 1.1486577987670898, "stability/response_length_mean": 48.5, "stability/response_length_std": 28.705400466918945, "stability/response_length_var": 824.0, "stability/token_entropy_mean": 5.171875, "stability/repetition_rate_mean": 0.341796875}
+{"current_steps": 196, "total_steps": 196, "epoch": 1.0, "percentage": 100.0, "elapsed_time": "0:52:25", "remaining_time": "0:00:00"}

mistral7b_kto_simple/trainer_state.json

@@ -0,0 +1,857 @@
+{
+  "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": 0.5772594809532166,
+      "kl": 0.006305641960352659,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59718379.78947368,
+      "logits/rejected": -50413504.0,
+      "logps/chosen": -439.03207236842104,
+      "logps/rejected": -198.4214332217262,
+      "loss": 0.5,
+      "loss/base_kto": 0.49969324469566345,
+      "rewards/chosen": 0.0013566404385002035,
+      "rewards/margins": 0.0025594215466638558,
+      "rewards/rejected": -0.001202781108163652,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.6031567454338074,
+      "kl": 0.028230074793100357,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59308962.90909091,
+      "logits/rejected": -50030080.0,
+      "logps/chosen": -429.9140625,
+      "logps/rejected": -192.86823465737953,
+      "loss": 0.5001,
+      "loss/base_kto": 0.49971267580986023,
+      "rewards/chosen": 0.002450463833746972,
+      "rewards/margins": 0.0024800021542366484,
+      "rewards/rejected": -2.95383204896766e-05,
+      "stability/repetition_rate_mean": 0.3664148449897766,
+      "stability/response_length_mean": 44.25,
+      "stability/response_length_std": 28.13614845275879,
+      "stability/response_length_var": 791.6428833007812,
+      "stability/token_entropy_mean": 3.4755859375,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.609616219997406,
+      "kl": 0.012302148155868053,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56221720.82424243,
+      "logits/rejected": -51918762.11612903,
+      "logps/chosen": -404.1202651515151,
+      "logps/rejected": -200.68004032258065,
+      "loss": 0.4998,
+      "loss/base_kto": 0.500029981136322,
+      "rewards/chosen": 0.001057231787479285,
+      "rewards/margins": -0.00034482885199441484,
+      "rewards/rejected": 0.0014020606394736998,
+      "stability/repetition_rate_mean": 0.2929687201976776,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.92341423034668,
+      "stability/response_length_var": 895.4107055664062,
+      "stability/token_entropy_mean": 3.7001953125,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.6197210550308228,
+      "kl": 0.016423773020505905,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51663750.4,
+      "logits/rejected": -48055072.0,
+      "logps/chosen": -343.6933349609375,
+      "logps/rejected": -225.6587890625,
+      "loss": 0.5004,
+      "loss/base_kto": 0.5000466704368591,
+      "rewards/chosen": 0.00033793365582823753,
+      "rewards/margins": -0.00011792583391070368,
+      "rewards/rejected": 0.0004558594897389412,
+      "stability/repetition_rate_mean": 0.2155257910490036,
+      "stability/response_length_mean": 44.0,
+      "stability/response_length_std": 28.0611572265625,
+      "stability/response_length_var": 787.4285888671875,
+      "stability/token_entropy_mean": 4.42578125,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.5772079229354858,
+      "kl": 0.01958128809928894,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56465217.32413793,
+      "logits/rejected": -52019422.35428572,
+      "logps/chosen": -378.4753232758621,
+      "logps/rejected": -202.7753125,
+      "loss": 0.5001,
+      "loss/base_kto": 0.49998876452445984,
+      "rewards/chosen": 0.0016139640890318772,
+      "rewards/margins": -0.00013830620079792184,
+      "rewards/rejected": 0.001752270289829799,
+      "stability/repetition_rate_mean": 0.19711539149284363,
+      "stability/response_length_mean": 50.25,
+      "stability/response_length_std": 25.549671173095703,
+      "stability/response_length_var": 652.7857055664062,
+      "stability/token_entropy_mean": 4.5234375,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.6031031012535095,
+      "kl": 0.020890694111585617,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57728749.10067114,
+      "logits/rejected": -51514302.12865497,
+      "logps/chosen": -468.5621854026846,
+      "logps/rejected": -198.23862390350877,
+      "loss": 0.5002,
+      "loss/base_kto": 0.49984827637672424,
+      "rewards/chosen": 0.0019035963404098613,
+      "rewards/margins": 0.000978329737129632,
+      "rewards/rejected": 0.0009252666032802292,
+      "stability/repetition_rate_mean": 0.1525297611951828,
+      "stability/response_length_mean": 43.0,
+      "stability/response_length_std": 29.602123260498047,
+      "stability/response_length_var": 876.2857055664062,
+      "stability/token_entropy_mean": 4.568359375,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6770267486572266,
+      "kl": 0.031000053510069847,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57633195.74683544,
+      "logits/rejected": -53898088.2962963,
+      "logps/chosen": -385.922245846519,
+      "logps/rejected": -195.10961612654322,
+      "loss": 0.4998,
+      "loss/base_kto": 0.5000401139259338,
+      "rewards/chosen": 0.0014279391380805004,
+      "rewards/margins": -0.0002580981845407863,
+      "rewards/rejected": 0.0016860373226212867,
+      "stability/repetition_rate_mean": 0.3030133843421936,
+      "stability/response_length_mean": 56.875,
+      "stability/response_length_std": 20.152544021606445,
+      "stability/response_length_var": 406.125,
+      "stability/token_entropy_mean": 5.474609375,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.587213933467865,
+      "kl": 0.0245906300842762,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56631648.0,
+      "logits/rejected": -50232096.0,
+      "logps/chosen": -421.8708984375,
+      "logps/rejected": -219.26767578125,
+      "loss": 0.4998,
+      "loss/base_kto": 0.50019770860672,
+      "rewards/chosen": 0.003170560672879219,
+      "rewards/margins": -0.0009862679988145826,
+      "rewards/rejected": 0.004156828671693802,
+      "stability/repetition_rate_mean": 0.15522204339504242,
+      "stability/response_length_mean": 35.0,
+      "stability/response_length_std": 31.518701553344727,
+      "stability/response_length_var": 993.4285888671875,
+      "stability/token_entropy_mean": 4.7880859375,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.640202522277832,
+      "kl": 0.038881849497556686,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52691122.684563756,
+      "logits/rejected": -50773704.60818713,
+      "logps/chosen": -352.2867030201342,
+      "logps/rejected": -243.83422423245614,
+      "loss": 0.4998,
+      "loss/base_kto": 0.4999396800994873,
+      "rewards/chosen": 0.004177153910566497,
+      "rewards/margins": 0.0001828821414053836,
+      "rewards/rejected": 0.003994271769161113,
+      "stability/repetition_rate_mean": 0.0859469324350357,
+      "stability/response_length_mean": 22.375,
+      "stability/response_length_std": 26.52189064025879,
+      "stability/response_length_var": 703.4107055664062,
+      "stability/token_entropy_mean": 4.9140625,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.5786681175231934,
+      "kl": 0.04716196283698082,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54047022.389261745,
+      "logits/rejected": -48305211.883040935,
+      "logps/chosen": -375.11949454697987,
+      "logps/rejected": -180.19741410818713,
+      "loss": 0.4999,
+      "loss/base_kto": 0.499483197927475,
+      "rewards/chosen": 0.007963991005148663,
+      "rewards/margins": 0.0036445832186064505,
+      "rewards/rejected": 0.004319407786542212,
+      "stability/repetition_rate_mean": 0.4108419120311737,
+      "stability/response_length_mean": 55.375,
+      "stability/response_length_std": 16.361431121826172,
+      "stability/response_length_var": 267.6964416503906,
+      "stability/token_entropy_mean": 3.81640625,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.663858950138092,
+      "kl": 0.0537404902279377,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56946357.26708075,
+      "logits/rejected": -49559825.710691825,
+      "logps/chosen": -344.4793769409938,
+      "logps/rejected": -230.40548840408806,
+      "loss": 0.4996,
+      "loss/base_kto": 0.5001100897789001,
+      "rewards/chosen": 0.003263072937912082,
+      "rewards/margins": -0.0009928640175938652,
+      "rewards/rejected": 0.004255936955505947,
+      "stability/repetition_rate_mean": 0.107421875,
+      "stability/response_length_mean": 24.5,
+      "stability/response_length_std": 32.710853576660156,
+      "stability/response_length_var": 1070.0,
+      "stability/token_entropy_mean": 4.41015625,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.5810667276382446,
+      "kl": 0.05833861976861954,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52670401.93939394,
+      "logits/rejected": -47818137.6,
+      "logps/chosen": -359.6366477272727,
+      "logps/rejected": -178.17121975806452,
+      "loss": 0.4997,
+      "loss/base_kto": 0.499610960483551,
+      "rewards/chosen": 0.008416923609646883,
+      "rewards/margins": 0.002847369465310552,
+      "rewards/rejected": 0.005569554144336331,
+      "stability/repetition_rate_mean": 0.18834353983402252,
+      "stability/response_length_mean": 44.625,
+      "stability/response_length_std": 27.8820858001709,
+      "stability/response_length_var": 777.4107055664062,
+      "stability/token_entropy_mean": 4.666015625,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.5759316682815552,
+      "kl": 0.07869795709848404,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57552697.80645161,
+      "logits/rejected": -47262831.70909091,
+      "logps/chosen": -381.796875,
+      "logps/rejected": -183.53030303030303,
+      "loss": 0.4996,
+      "loss/base_kto": 0.4987983703613281,
+      "rewards/chosen": 0.014990917328865297,
+      "rewards/margins": 0.009785529548704684,
+      "rewards/rejected": 0.005205387780160615,
+      "stability/repetition_rate_mean": 0.19731052219867706,
+      "stability/response_length_mean": 21.125,
+      "stability/response_length_std": 28.027729034423828,
+      "stability/response_length_var": 785.5535888671875,
+      "stability/token_entropy_mean": 5.58984375,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5528621673583984,
+      "kl": 0.08887721598148346,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52767777.03225806,
+      "logits/rejected": -49031406.93333333,
+      "logps/chosen": -379.00811491935485,
+      "logps/rejected": -220.3328125,
+      "loss": 0.4992,
+      "loss/base_kto": 0.49963781237602234,
+      "rewards/chosen": 0.012032072005733366,
+      "rewards/margins": 0.0030974022221122426,
+      "rewards/rejected": 0.008934669783621124,
+      "stability/repetition_rate_mean": 0.185546875,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.7265625,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.6098544001579285,
+      "kl": 0.09266681969165802,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49499744.86486486,
+      "logits/rejected": -47894179.72093023,
+      "logps/chosen": -314.1058910472973,
+      "logps/rejected": -228.81817768895348,
+      "loss": 0.4993,
+      "loss/base_kto": 0.49878960847854614,
+      "rewards/chosen": 0.016490372451576026,
+      "rewards/margins": 0.009822355097453599,
+      "rewards/rejected": 0.006668017354122428,
+      "stability/repetition_rate_mean": 0.08308292925357819,
+      "stability/response_length_mean": 22.75,
+      "stability/response_length_std": 31.07479476928711,
+      "stability/response_length_var": 965.6428833007812,
+      "stability/token_entropy_mean": 2.146484375,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.6027852892875671,
+      "kl": 0.12336266040802002,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58629587.9205298,
+      "logits/rejected": -53451757.82248521,
+      "logps/chosen": -403.85350269039736,
+      "logps/rejected": -200.96463572485206,
+      "loss": 0.4991,
+      "loss/base_kto": 0.4989641308784485,
+      "rewards/chosen": 0.017660609933714204,
+      "rewards/margins": 0.008543980056680181,
+      "rewards/rejected": 0.009116629877034023,
+      "stability/repetition_rate_mean": 0.18541666865348816,
+      "stability/response_length_mean": 39.0,
+      "stability/response_length_std": 29.159658432006836,
+      "stability/response_length_var": 850.2857055664062,
+      "stability/token_entropy_mean": 5.22265625,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.6124152541160583,
+      "kl": 0.125388041138649,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51481851.04516129,
+      "logits/rejected": -47137016.24242424,
+      "logps/chosen": -358.1038306451613,
+      "logps/rejected": -230.3265625,
+      "loss": 0.4988,
+      "loss/base_kto": 0.4984590709209442,
+      "rewards/chosen": 0.022868017996511152,
+      "rewards/margins": 0.013043898081849399,
+      "rewards/rejected": 0.009824119914661754,
+      "stability/repetition_rate_mean": 0.12302795797586441,
+      "stability/response_length_mean": 27.375,
+      "stability/response_length_std": 30.39707374572754,
+      "stability/response_length_var": 923.9821166992188,
+      "stability/token_entropy_mean": 4.92578125,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.6379120945930481,
+      "kl": 0.17190246284008026,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58409432.615384616,
+      "logits/rejected": -50583776.780487806,
+      "logps/chosen": -380.54447115384613,
+      "logps/rejected": -187.12595274390245,
+      "loss": 0.4984,
+      "loss/base_kto": 0.4981650412082672,
+      "rewards/chosen": 0.024491310119628906,
+      "rewards/margins": 0.014406566939702848,
+      "rewards/rejected": 0.010084743179926059,
+      "stability/repetition_rate_mean": 0.2374526560306549,
+      "stability/response_length_mean": 35.625,
+      "stability/response_length_std": 31.057262420654297,
+      "stability/response_length_var": 964.5535888671875,
+      "stability/token_entropy_mean": 3.7021484375,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.6159020066261292,
+      "kl": 0.15860331058502197,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50396399.79746836,
+      "logits/rejected": -48982038.12345679,
+      "logps/chosen": -314.7759592563291,
+      "logps/rejected": -207.06117380401236,
+      "loss": 0.4982,
+      "loss/base_kto": 0.49839383363723755,
+      "rewards/chosen": 0.021742636644387546,
+      "rewards/margins": 0.013488207203948062,
+      "rewards/rejected": 0.008254429440439484,
+      "stability/repetition_rate_mean": 0.09133065491914749,
+      "stability/response_length_mean": 21.875,
+      "stability/response_length_std": 27.304826736450195,
+      "stability/response_length_var": 745.5535888671875,
+      "stability/token_entropy_mean": 4.076171875,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.6105856895446777,
+      "kl": 0.2014942169189453,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54708237.83783784,
+      "logits/rejected": -50898551.069767445,
+      "logps/chosen": -359.57057643581084,
+      "logps/rejected": -216.38419785610466,
+      "loss": 0.4981,
+      "loss/base_kto": 0.49866095185279846,
+      "rewards/chosen": 0.029960999617705476,
+      "rewards/margins": 0.010491025964843637,
+      "rewards/rejected": 0.01946997365286184,
+      "stability/repetition_rate_mean": 0.22838345170021057,
+      "stability/response_length_mean": 37.875,
+      "stability/response_length_std": 30.248672485351562,
+      "stability/response_length_var": 914.9821166992188,
+      "stability/token_entropy_mean": 4.185546875,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6785591244697571,
+      "kl": 0.22638840973377228,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59369200.578313254,
+      "logits/rejected": -53790852.98701299,
+      "logps/chosen": -399.54828689759034,
+      "logps/rejected": -209.8457538555195,
+      "loss": 0.4975,
+      "loss/base_kto": 0.49753338098526,
+      "rewards/chosen": 0.03542007021157138,
+      "rewards/margins": 0.01961183469632369,
+      "rewards/rejected": 0.015808235515247692,
+      "stability/repetition_rate_mean": 0.1897321343421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.349609375,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.6502929925918579,
+      "kl": 0.266899973154068,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57387927.13253012,
+      "logits/rejected": -53615536.20779221,
+      "logps/chosen": -378.2028661521084,
+      "logps/rejected": -205.8787413758117,
+      "loss": 0.4975,
+      "loss/base_kto": 0.49790263175964355,
+      "rewards/chosen": 0.03603898760784103,
+      "rewards/margins": 0.015099021306975758,
+      "rewards/rejected": 0.02093996630086527,
+      "stability/repetition_rate_mean": 0.31048768758773804,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.911476135253906,
+      "stability/response_length_var": 894.6964111328125,
+      "stability/token_entropy_mean": 4.357421875,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.6486251950263977,
+      "kl": 0.32395443320274353,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60606632.50632911,
+      "logits/rejected": -53405677.03703704,
+      "logps/chosen": -364.0806467563291,
+      "logps/rejected": -200.9707995756173,
+      "loss": 0.4965,
+      "loss/base_kto": 0.4963931143283844,
+      "rewards/chosen": 0.05118938035602811,
+      "rewards/margins": 0.0286137355568074,
+      "rewards/rejected": 0.02257564479922071,
+      "stability/repetition_rate_mean": 0.2749515473842621,
+      "stability/response_length_mean": 38.75,
+      "stability/response_length_std": 29.936838150024414,
+      "stability/response_length_var": 896.2142944335938,
+      "stability/token_entropy_mean": 4.966796875,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.6199327111244202,
+      "kl": 0.3587552011013031,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57752502.37908497,
+      "logits/rejected": -54366698.53892215,
+      "logps/chosen": -383.6620710784314,
+      "logps/rejected": -211.94790886976048,
+      "loss": 0.4959,
+      "loss/base_kto": 0.4960261285305023,
+      "rewards/chosen": 0.05828505871342678,
+      "rewards/margins": 0.03143982054748571,
+      "rewards/rejected": 0.026845238165941067,
+      "stability/repetition_rate_mean": 0.1723366528749466,
+      "stability/response_length_mean": 27.875,
+      "stability/response_length_std": 30.31943130493164,
+      "stability/response_length_var": 919.2678833007812,
+      "stability/token_entropy_mean": 4.640625,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.6118870377540588,
+      "kl": 0.3919747769832611,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60696351.56164384,
+      "logits/rejected": -55669807.08045977,
+      "logps/chosen": -416.76867508561645,
+      "logps/rejected": -216.09132543103448,
+      "loss": 0.4961,
+      "loss/base_kto": 0.4960012435913086,
+      "rewards/chosen": 0.061065817532474045,
+      "rewards/margins": 0.033479519105111495,
+      "rewards/rejected": 0.02758629842736255,
+      "stability/repetition_rate_mean": 0.15937499701976776,
+      "stability/response_length_mean": 15.875,
+      "stability/response_length_std": 24.584766387939453,
+      "stability/response_length_var": 604.4107055664062,
+      "stability/token_entropy_mean": 3.49609375,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.5947321653366089,
+      "kl": 0.45393142104148865,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57384028.49032258,
+      "logits/rejected": -53841212.50909091,
+      "logps/chosen": -391.96595262096776,
+      "logps/rejected": -183.6536221590909,
+      "loss": 0.4954,
+      "loss/base_kto": 0.49458155035972595,
+      "rewards/chosen": 0.06919794390278478,
+      "rewards/margins": 0.042669391818526664,
+      "rewards/rejected": 0.026528552084258108,
+      "stability/repetition_rate_mean": 0.25242185592651367,
+      "stability/response_length_mean": 43.25,
+      "stability/response_length_std": 29.620214462280273,
+      "stability/response_length_var": 877.3571166992188,
+      "stability/token_entropy_mean": 3.7138671875,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5941182971000671,
+      "kl": 0.501304566860199,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59851254.57668712,
+      "logits/rejected": -55000540.12738854,
+      "logps/chosen": -400.9979629984663,
+      "logps/rejected": -247.54722332802547,
+      "loss": 0.4945,
+      "loss/base_kto": 0.4937683045864105,
+      "rewards/chosen": 0.08181942898802962,
+      "rewards/margins": 0.049467831888482444,
+      "rewards/rejected": 0.03235159709954717,
+      "stability/repetition_rate_mean": 0.2230902761220932,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.936838150024414,
+      "stability/response_length_var": 896.2142944335938,
+      "stability/token_entropy_mean": 4.072265625,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.7290944457054138,
+      "kl": 0.4828576147556305,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57085976.82424243,
+      "logits/rejected": -50076896.61935484,
+      "logps/chosen": -366.3819128787879,
+      "logps/rejected": -189.91817036290323,
+      "loss": 0.4937,
+      "loss/base_kto": 0.4917580187320709,
+      "rewards/chosen": 0.08078298857717803,
+      "rewards/margins": 0.06915265418450736,
+      "rewards/rejected": 0.011630334392670662,
+      "stability/repetition_rate_mean": 0.2265625,
+      "stability/response_length_mean": 56.25,
+      "stability/response_length_std": 21.920310974121094,
+      "stability/response_length_var": 480.5,
+      "stability/token_entropy_mean": 5.150390625,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.6742141246795654,
+      "kl": 0.5725060105323792,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51148522.409638554,
+      "logits/rejected": -51662778.18181818,
+      "logps/chosen": -329.4726327183735,
+      "logps/rejected": -199.14629413555195,
+      "loss": 0.4932,
+      "loss/base_kto": 0.49420061707496643,
+      "rewards/chosen": 0.07991740789758153,
+      "rewards/margins": 0.04575571016309556,
+      "rewards/rejected": 0.03416169773448597,
+      "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": 5.2734375,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.6416594386100769,
+      "kl": 0.638058602809906,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55423500.04705882,
+      "logits/rejected": -50200978.77333333,
+      "logps/chosen": -396.80946691176473,
+      "logps/rejected": -207.36325520833333,
+      "loss": 0.493,
+      "loss/base_kto": 0.49001631140708923,
+      "rewards/chosen": 0.11249845168169807,
+      "rewards/margins": 0.0762304780997482,
+      "rewards/rejected": 0.03626797358194987,
+      "stability/repetition_rate_mean": 0.23066024482250214,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.193626403808594,
+      "stability/response_length_var": 852.2678833007812,
+      "stability/token_entropy_mean": 4.6015625,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.6336647272109985,
+      "kl": 0.6403493881225586,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54021613.92941176,
+      "logits/rejected": -49351734.61333334,
+      "logps/chosen": -360.1190487132353,
+      "logps/rejected": -191.01639322916665,
+      "loss": 0.4925,
+      "loss/base_kto": 0.4930504858493805,
+      "rewards/chosen": 0.0870250758002786,
+      "rewards/margins": 0.058103303161321895,
+      "rewards/rejected": 0.028921772638956705,
+      "stability/repetition_rate_mean": 0.3113190531730652,
+      "stability/response_length_mean": 61.375,
+      "stability/response_length_std": 7.424621105194092,
+      "stability/response_length_var": 55.125,
+      "stability/token_entropy_mean": 4.529296875,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.6168467402458191,
+      "kl": 0.7161628603935242,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55242325.333333336,
+      "logits/rejected": -48961121.684210524,
+      "logps/chosen": -415.55738467261904,
+      "logps/rejected": -186.4566072162829,
+      "loss": 0.4911,
+      "loss/base_kto": 0.4885805547237396,
+      "rewards/chosen": 0.13379353568667457,
+      "rewards/margins": 0.08971994144277165,
+      "rewards/rejected": 0.04407359424390291,
+      "stability/repetition_rate_mean": 0.30851560831069946,
+      "stability/response_length_mean": 32.5,
+      "stability/response_length_std": 30.528675079345703,
+      "stability/response_length_var": 932.0,
+      "stability/token_entropy_mean": 4.474609375,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.7232491374015808,
+      "kl": 0.7498390078544617,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54396980.96551724,
+      "logits/rejected": -49794756.38356164,
+      "logps/chosen": -369.75480423850576,
+      "logps/rejected": -212.1316486515411,
+      "loss": 0.4894,
+      "loss/base_kto": 0.49128732085227966,
+      "rewards/chosen": 0.11619376040053093,
+      "rewards/margins": 0.07014617738226721,
+      "rewards/rejected": 0.04604758301826373,
+      "stability/repetition_rate_mean": 0.2411106526851654,
+      "stability/response_length_mean": 22.5,
+      "stability/response_length_std": 28.33977508544922,
+      "stability/response_length_var": 803.1428833007812,
+      "stability/token_entropy_mean": 4.86328125,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.6223028302192688,
+      "kl": 0.9178180694580078,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59141307.834319524,
+      "logits/rejected": -52591744.84768212,
+      "logps/chosen": -439.69304733727813,
+      "logps/rejected": -190.61025455298014,
+      "loss": 0.4881,
+      "loss/base_kto": 0.48755979537963867,
+      "rewards/chosen": 0.16488214075212648,
+      "rewards/margins": 0.10012487574960424,
+      "rewards/rejected": 0.06475726500252224,
+      "stability/repetition_rate_mean": 0.3434279263019562,
+      "stability/response_length_mean": 44.125,
+      "stability/response_length_std": 28.43256378173828,
+      "stability/response_length_var": 808.4107055664062,
+      "stability/token_entropy_mean": 3.70166015625,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.7227226495742798,
+      "kl": 0.937430202960968,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56671336.86746988,
+      "logits/rejected": -48043027.948051944,
+      "logps/chosen": -354.8946018448795,
+      "logps/rejected": -194.3057528409091,
+      "loss": 0.4875,
+      "loss/base_kto": 0.48688632249832153,
+      "rewards/chosen": 0.154259325510048,
+      "rewards/margins": 0.10408077600360047,
+      "rewards/rejected": 0.05017854950644753,
+      "stability/repetition_rate_mean": 0.18430830538272858,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 25.773740768432617,
+      "stability/response_length_var": 664.2857055664062,
+      "stability/token_entropy_mean": 4.8564453125,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.6434662938117981,
+      "kl": 1.026900291442871,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52659602.06134969,
+      "logits/rejected": -47986926.06369427,
+      "logps/chosen": -374.42537384969324,
+      "logps/rejected": -210.55354299363057,
+      "loss": 0.4865,
+      "loss/base_kto": 0.4877110421657562,
+      "rewards/chosen": 0.17204565504577263,
+      "rewards/margins": 0.09757829369276705,
+      "rewards/rejected": 0.07446736135300558,
+      "stability/repetition_rate_mean": 0.16679687798023224,
+      "stability/response_length_mean": 34.75,
+      "stability/response_length_std": 31.54928207397461,
+      "stability/response_length_var": 995.3571166992188,
+      "stability/token_entropy_mean": 3.541015625,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.6501672863960266,
+      "kl": 1.079397439956665,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54617631.60493827,
+      "logits/rejected": -49526434.025316454,
+      "logps/chosen": -310.3054108796296,
+      "logps/rejected": -196.56382268591773,
+      "loss": 0.4857,
+      "loss/base_kto": 0.4866374433040619,
+      "rewards/chosen": 0.16959618933406878,
+      "rewards/margins": 0.12206393268857642,
+      "rewards/rejected": 0.04753225664549236,
+      "stability/repetition_rate_mean": 0.205078125,
+      "stability/response_length_mean": 32.875,
+      "stability/response_length_std": 33.30996322631836,
+      "stability/response_length_var": 1109.5535888671875,
+      "stability/token_entropy_mean": 3.732421875,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.6943851113319397,
+      "kl": 1.1196650266647339,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53094718.8502994,
+      "logits/rejected": -49691517.49019608,
+      "logps/chosen": -390.31998502994014,
+      "logps/rejected": -210.46075878267973,
+      "loss": 0.484,
+      "loss/base_kto": 0.48342210054397583,
+      "rewards/chosen": 0.2085319564727966,
+      "rewards/margins": 0.13474922263035827,
+      "rewards/rejected": 0.07378273384243834,
+      "stability/repetition_rate_mean": 0.18919041752815247,
+      "stability/response_length_mean": 18.625,
+      "stability/response_length_std": 26.52189064025879,
+      "stability/response_length_var": 703.4107055664062,
+      "stability/token_entropy_mean": 4.5703125,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.6349285244941711,
+      "kl": 1.1486577987670898,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60025334.090322584,
+      "logits/rejected": -53795740.7030303,
+      "logps/chosen": -390.8627268145161,
+      "logps/rejected": -197.078125,
+      "loss": 0.4843,
+      "loss/base_kto": 0.4863618016242981,
+      "rewards/chosen": 0.1922200725924584,
+      "rewards/margins": 0.10809968629540585,
+      "rewards/rejected": 0.08412038629705255,
+      "stability/repetition_rate_mean": 0.341796875,
+      "stability/response_length_mean": 48.5,
+      "stability/response_length_std": 28.705400466918945,
+      "stability/response_length_var": 824.0,
+      "stability/token_entropy_mean": 5.171875,
+      "step": 195
+    },
+    {
+      "epoch": 1.0,
+      "step": 196,
+      "total_flos": 3.1861866257511875e+18,
+      "train_loss": 0.4956757848968311,
+      "train_runtime": 3145.5892,
+      "train_samples_per_second": 31.791,
+      "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_simple/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cca02ae9ad4d71243fc18608eed865018fb06c94becd454c7ce13581852dd3e2
+size 8017