Upload folder using huggingface_hub

mistral7b_kto_w2_delta0.01/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_w2_delta0.01
+  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_w2_delta0.01
+
+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_w2_delta0.01/adapter_config.json

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

mistral7b_kto_w2_delta0.01/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc532f3a9b2ccc58878b589488df8086ef4250cc231c88aba808dbb76b5294c1
+size 42002584

mistral7b_kto_w2_delta0.01/all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 1.359869228333843,
+    "train_runtime": 4299.7361,
+    "train_samples_per_second": 23.257,
+    "train_steps_per_second": 0.046
+}

mistral7b_kto_w2_delta0.01/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_w2_delta0.01/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_w2_delta0.01/checkpoint-196/adapter_config.json

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

mistral7b_kto_w2_delta0.01/checkpoint-196/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc532f3a9b2ccc58878b589488df8086ef4250cc231c88aba808dbb76b5294c1
+size 42002584

mistral7b_kto_w2_delta0.01/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_w2_delta0.01/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 sha256:779c93b9ef581111af46f25107bbc846dbc7efba4fc3330c6646a0f70d2a5705
+size 31467333

mistral7b_kto_w2_delta0.01/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 sha256:54845b390b610a05d465f7be357b40b0f638cade1acb414872a5f5e08b53a283
+size 31467333

mistral7b_kto_w2_delta0.01/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 sha256:1f5731e01ba75fdcdcc89b870fbc8b0b2ff2b0bd23fb9065431fdd5f62d6cacd
+size 31467333

mistral7b_kto_w2_delta0.01/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 sha256:3725d43dace61fe2434980b1d08a2a437d1dbd37c80f0637e8d29e23f70c69ff
+size 31467397

mistral7b_kto_w2_delta0.01/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 sha256:80a8bb45c28fc36d719d16d558d3a2d236e2e627012f2570e1a96022c7590666
+size 31467397

mistral7b_kto_w2_delta0.01/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:edb86df20859eb8a3c9f4d6fad4158ee1a925d0f89df2b091c68ae8737481d9e
+size 31467397

mistral7b_kto_w2_delta0.01/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:b89f2f782f2d2a6cd0871cfd698a8baad34179c1307f595f933a56d06251613d
+size 31467397

mistral7b_kto_w2_delta0.01/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 sha256:7b84b0e63bb09e2569efc034248f9908399a8a6cd39f255a426c6a7f67f6de21
+size 31467397

mistral7b_kto_w2_delta0.01/checkpoint-196/global_step196/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d264421cb65ff59b80635aaf644b74c9eb54ef0e25d0b629bd49ccc9d4be72f
+size 42288409

mistral7b_kto_w2_delta0.01/checkpoint-196/latest

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

mistral7b_kto_w2_delta0.01/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_w2_delta0.01/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_w2_delta0.01/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_w2_delta0.01/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_w2_delta0.01/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_w2_delta0.01/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_w2_delta0.01/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_w2_delta0.01/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_w2_delta0.01/checkpoint-196/scheduler.pt

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

mistral7b_kto_w2_delta0.01/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_w2_delta0.01/checkpoint-196/tokenizer.model

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

mistral7b_kto_w2_delta0.01/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_w2_delta0.01/checkpoint-196/trainer_state.json

@@ -0,0 +1,965 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 196,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.025591810620601407,
+      "grad_norm": 0.5667362213134766,
+      "kl": 0.008433527313172817,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59717241.2631579,
+      "logits/rejected": -50415911.61904762,
+      "logps/chosen": -439.02847450657896,
+      "logps/rejected": -198.40762183779762,
+      "loss": 1.3863,
+      "loss/base_kto": 0.4998113214969635,
+      "metrics/advantage_var": 0.04507648944854736,
+      "regularization/grad_norm_sq": 4328.5595703125,
+      "regularization/w2": 0.9065223932266235,
+      "rewards/chosen": 0.0017119320599656355,
+      "rewards/margins": 0.00153388447107229,
+      "rewards/rejected": 0.00017804758889334544,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.6054168343544006,
+      "kl": 0.019305892288684845,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59293257.14285714,
+      "logits/rejected": -50017699.469879515,
+      "logps/chosen": -429.92294034090907,
+      "logps/rejected": -192.84436182228916,
+      "loss": 1.3645,
+      "loss/base_kto": 0.50011146068573,
+      "metrics/advantage_var": 0.06692608445882797,
+      "regularization/grad_norm_sq": 3843.479736328125,
+      "regularization/w2": 0.8646374940872192,
+      "rewards/chosen": 0.0015573305162516508,
+      "rewards/margins": -0.0008008119947025051,
+      "rewards/rejected": 0.002358142510954156,
+      "stability/repetition_rate_mean": 0.3629409670829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.27001953125,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.6099650859832764,
+      "kl": 0.013482200913131237,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56229807.32121212,
+      "logits/rejected": -51916106.32258064,
+      "logps/chosen": -404.1315814393939,
+      "logps/rejected": -200.67066532258065,
+      "loss": 1.3548,
+      "loss/base_kto": 0.5002730488777161,
+      "metrics/advantage_var": 0.053939878940582275,
+      "regularization/grad_norm_sq": 4172.4462890625,
+      "regularization/w2": 0.8843773007392883,
+      "rewards/chosen": -7.632036100734364e-05,
+      "rewards/margins": -0.0024158480448806742,
+      "rewards/rejected": 0.0023395276838733305,
+      "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": 0.619081437587738,
+      "kl": 0.018938787281513214,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51664345.6,
+      "logits/rejected": -48065900.8,
+      "logps/chosen": -343.694580078125,
+      "logps/rejected": -225.647802734375,
+      "loss": 1.3654,
+      "loss/base_kto": 0.5002180337905884,
+      "metrics/advantage_var": 0.05758875235915184,
+      "regularization/grad_norm_sq": 3665.234375,
+      "regularization/w2": 0.843424916267395,
+      "rewards/chosen": 0.00021410472691059114,
+      "rewards/margins": -0.0013427346944808961,
+      "rewards/rejected": 0.0015568394213914872,
+      "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": 4.4140625,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.5774287581443787,
+      "kl": 0.02769310213625431,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56488998.84137931,
+      "logits/rejected": -52022798.62857143,
+      "logps/chosen": -378.4621767241379,
+      "logps/rejected": -202.77145089285713,
+      "loss": 1.3771,
+      "loss/base_kto": 0.4999386966228485,
+      "metrics/advantage_var": 0.053915154188871384,
+      "regularization/grad_norm_sq": 4001.77783203125,
+      "regularization/w2": 0.860791027545929,
+      "rewards/chosen": 0.0029272856383488095,
+      "rewards/margins": 0.0007881318172210544,
+      "rewards/rejected": 0.002139153821127755,
+      "stability/repetition_rate_mean": 0.17215879261493683,
+      "stability/response_length_mean": 43.375,
+      "stability/response_length_std": 28.54538917541504,
+      "stability/response_length_var": 814.8392944335938,
+      "stability/token_entropy_mean": 4.8125,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.603484034538269,
+      "kl": 0.018374785780906677,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57729649.39597315,
+      "logits/rejected": -51505858.61988304,
+      "logps/chosen": -468.588716442953,
+      "logps/rejected": -198.24766995614036,
+      "loss": 1.3694,
+      "loss/base_kto": 0.5000385642051697,
+      "metrics/advantage_var": 0.07535337656736374,
+      "regularization/grad_norm_sq": 4544.02490234375,
+      "regularization/w2": 0.917149007320404,
+      "rewards/chosen": -0.0007460268351855694,
+      "rewards/margins": -0.0007687252250420083,
+      "rewards/rejected": 2.2698389856438888e-05,
+      "stability/repetition_rate_mean": 0.1427641361951828,
+      "stability/response_length_mean": 43.0,
+      "stability/response_length_std": 29.602123260498047,
+      "stability/response_length_var": 876.2857055664062,
+      "stability/token_entropy_mean": 4.61328125,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6760314106941223,
+      "kl": 0.02894224040210247,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57635956.658227846,
+      "logits/rejected": -53888758.518518515,
+      "logps/chosen": -385.90286293512656,
+      "logps/rejected": -195.11945408950618,
+      "loss": 1.3758,
+      "loss/base_kto": 0.49967771768569946,
+      "metrics/advantage_var": 0.08824540674686432,
+      "regularization/grad_norm_sq": 3807.059326171875,
+      "regularization/w2": 0.8619928359985352,
+      "rewards/chosen": 0.003366763833202893,
+      "rewards/margins": 0.002664287083385363,
+      "rewards/rejected": 0.0007024767498175303,
+      "stability/repetition_rate_mean": 0.3166852593421936,
+      "stability/response_length_mean": 56.875,
+      "stability/response_length_std": 20.152544021606445,
+      "stability/response_length_var": 406.125,
+      "stability/token_entropy_mean": 5.419921875,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.5855346322059631,
+      "kl": 0.02706434763967991,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56634796.8,
+      "logits/rejected": -50219603.2,
+      "logps/chosen": -421.87265625,
+      "logps/rejected": -219.273095703125,
+      "loss": 1.3686,
+      "loss/base_kto": 0.5001084208488464,
+      "metrics/advantage_var": 0.08759542554616928,
+      "regularization/grad_norm_sq": 4305.833984375,
+      "regularization/w2": 0.9105228781700134,
+      "rewards/chosen": 0.0030031204223632812,
+      "rewards/margins": -0.0006122220307588576,
+      "rewards/rejected": 0.003615342453122139,
+      "stability/repetition_rate_mean": 0.13671875,
+      "stability/response_length_mean": 33.125,
+      "stability/response_length_std": 33.01704025268555,
+      "stability/response_length_var": 1090.125,
+      "stability/token_entropy_mean": 4.8935546875,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.6397272944450378,
+      "kl": 0.03937375545501709,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52699170.36241611,
+      "logits/rejected": -50773297.403508775,
+      "logps/chosen": -352.2831900167785,
+      "logps/rejected": -243.85115131578948,
+      "loss": 1.379,
+      "loss/base_kto": 0.4997560679912567,
+      "metrics/advantage_var": 0.07129540294408798,
+      "regularization/grad_norm_sq": 3761.641357421875,
+      "regularization/w2": 0.8400061726570129,
+      "rewards/chosen": 0.004526910365827932,
+      "rewards/margins": 0.0022242151040232504,
+      "rewards/rejected": 0.0023026952618046812,
+      "stability/repetition_rate_mean": 0.0994017943739891,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.87324333190918,
+      "stability/response_length_var": 892.4107055664062,
+      "stability/token_entropy_mean": 4.7578125,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.5777538418769836,
+      "kl": 0.047937434166669846,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54044414.281879194,
+      "logits/rejected": -48299597.847953215,
+      "logps/chosen": -375.1258389261745,
+      "logps/rejected": -180.18202896564327,
+      "loss": 1.3394,
+      "loss/base_kto": 0.49980711936950684,
+      "metrics/advantage_var": 0.05609378218650818,
+      "regularization/grad_norm_sq": 3838.28369140625,
+      "regularization/w2": 0.8510723114013672,
+      "rewards/chosen": 0.007330068805873794,
+      "rewards/margins": 0.0014725509611215326,
+      "rewards/rejected": 0.005857517844752262,
+      "stability/repetition_rate_mean": 0.2952169179916382,
+      "stability/response_length_mean": 53.375,
+      "stability/response_length_std": 21.480472564697266,
+      "stability/response_length_var": 461.41070556640625,
+      "stability/token_entropy_mean": 4.1162109375,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.6621295213699341,
+      "kl": 0.0575476698577404,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56949887.204968944,
+      "logits/rejected": -49557458.91823899,
+      "logps/chosen": -344.48966420807454,
+      "logps/rejected": -230.39740566037736,
+      "loss": 1.357,
+      "loss/base_kto": 0.5003090500831604,
+      "metrics/advantage_var": 0.06319057196378708,
+      "regularization/grad_norm_sq": 3955.544921875,
+      "regularization/w2": 0.8724905252456665,
+      "rewards/chosen": 0.0022328466362094286,
+      "rewards/margins": -0.002831306810132032,
+      "rewards/rejected": 0.00506415344634146,
+      "stability/repetition_rate_mean": 0.11589926481246948,
+      "stability/response_length_mean": 22.375,
+      "stability/response_length_std": 30.236862182617188,
+      "stability/response_length_var": 914.2678833007812,
+      "stability/token_entropy_mean": 4.302734375,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.5823009014129639,
+      "kl": 0.05734771117568016,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52677849.21212121,
+      "logits/rejected": -47824090.01290323,
+      "logps/chosen": -359.6180871212121,
+      "logps/rejected": -178.15743447580644,
+      "loss": 1.3664,
+      "loss/base_kto": 0.4995133578777313,
+      "metrics/advantage_var": 0.057131972163915634,
+      "regularization/grad_norm_sq": 3162.57421875,
+      "regularization/w2": 0.7870045900344849,
+      "rewards/chosen": 0.010271583903919567,
+      "rewards/margins": 0.003323828090320934,
+      "rewards/rejected": 0.006947755813598633,
+      "stability/repetition_rate_mean": 0.17885693907737732,
+      "stability/response_length_mean": 44.625,
+      "stability/response_length_std": 27.8820858001709,
+      "stability/response_length_var": 777.4107055664062,
+      "stability/token_entropy_mean": 4.607421875,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.5745700597763062,
+      "kl": 0.07911502569913864,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57565778.58064516,
+      "logits/rejected": -47265906.81212121,
+      "logps/chosen": -381.8476814516129,
+      "logps/rejected": -183.48361742424242,
+      "loss": 1.3773,
+      "loss/base_kto": 0.49997368454933167,
+      "metrics/advantage_var": 0.06365744024515152,
+      "regularization/grad_norm_sq": 3887.826904296875,
+      "regularization/w2": 0.8648955225944519,
+      "rewards/chosen": 0.009909362177694997,
+      "rewards/margins": 3.680861240957546e-05,
+      "rewards/rejected": 0.009872553565285422,
+      "stability/repetition_rate_mean": 0.1857219785451889,
+      "stability/response_length_mean": 20.625,
+      "stability/response_length_std": 27.176342010498047,
+      "stability/response_length_var": 738.5535888671875,
+      "stability/token_entropy_mean": 5.525390625,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5528952479362488,
+      "kl": 0.09082157909870148,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52773428.8516129,
+      "logits/rejected": -49024763.345454544,
+      "logps/chosen": -379.00569556451615,
+      "logps/rejected": -220.34614109848485,
+      "loss": 1.403,
+      "loss/base_kto": 0.49938470125198364,
+      "metrics/advantage_var": 0.07325268536806107,
+      "regularization/grad_norm_sq": 4501.79833984375,
+      "regularization/w2": 0.9129446148872375,
+      "rewards/chosen": 0.01227622724348499,
+      "rewards/margins": 0.004672638272260291,
+      "rewards/rejected": 0.007603588971224698,
+      "stability/repetition_rate_mean": 0.162109375,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.77734375,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.609993040561676,
+      "kl": 0.08937091380357742,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49499239.783783786,
+      "logits/rejected": -47880882.60465116,
+      "logps/chosen": -314.12774493243245,
+      "logps/rejected": -228.82871547965115,
+      "loss": 1.3809,
+      "loss/base_kto": 0.4988512098789215,
+      "metrics/advantage_var": 0.08098717033863068,
+      "regularization/grad_norm_sq": 3738.951171875,
+      "regularization/w2": 0.8484571576118469,
+      "rewards/chosen": 0.014305324167818637,
+      "rewards/margins": 0.008692953588076915,
+      "rewards/rejected": 0.005612370579741722,
+      "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": 2.15625,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.603708028793335,
+      "kl": 0.11913253366947174,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58623972.87417219,
+      "logits/rejected": -53456156.78106509,
+      "logps/chosen": -403.85572744205297,
+      "logps/rejected": -200.97545303254438,
+      "loss": 1.3598,
+      "loss/base_kto": 0.49891233444213867,
+      "metrics/advantage_var": 0.07670630514621735,
+      "regularization/grad_norm_sq": 3606.688232421875,
+      "regularization/w2": 0.8389098048210144,
+      "rewards/chosen": 0.01743706330558322,
+      "rewards/margins": 0.009403561979603481,
+      "rewards/rejected": 0.00803350132597974,
+      "stability/repetition_rate_mean": 0.19088542461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.2265625,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.6118994951248169,
+      "kl": 0.12905550003051758,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51481953.44516129,
+      "logits/rejected": -47134605.18787879,
+      "logps/chosen": -358.1473790322581,
+      "logps/rejected": -230.32398200757575,
+      "loss": 1.3679,
+      "loss/base_kto": 0.49895092844963074,
+      "metrics/advantage_var": 0.09141772240400314,
+      "regularization/grad_norm_sq": 3847.29296875,
+      "regularization/w2": 0.8647522330284119,
+      "rewards/chosen": 0.018512418193201863,
+      "rewards/margins": 0.008430653047235363,
+      "rewards/rejected": 0.0100817651459665,
+      "stability/repetition_rate_mean": 0.1218377947807312,
+      "stability/response_length_mean": 32.375,
+      "stability/response_length_std": 33.547142028808594,
+      "stability/response_length_var": 1125.4107666015625,
+      "stability/token_entropy_mean": 4.869140625,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.638875424861908,
+      "kl": 0.16841307282447815,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58414395.07692308,
+      "logits/rejected": -50576390.24390244,
+      "logps/chosen": -380.533203125,
+      "logps/rejected": -187.13657583841464,
+      "loss": 1.3594,
+      "loss/base_kto": 0.49779611825942993,
+      "metrics/advantage_var": 0.11126144975423813,
+      "regularization/grad_norm_sq": 3978.860595703125,
+      "regularization/w2": 0.8712137341499329,
+      "rewards/chosen": 0.025618803806793995,
+      "rewards/margins": 0.016595058194393664,
+      "rewards/rejected": 0.009023745612400333,
+      "stability/repetition_rate_mean": 0.2843276560306549,
+      "stability/response_length_mean": 26.875,
+      "stability/response_length_std": 27.544445037841797,
+      "stability/response_length_var": 758.6964111328125,
+      "stability/token_entropy_mean": 4.3203125,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.6166263222694397,
+      "kl": 0.1684073954820633,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50395103.594936706,
+      "logits/rejected": -48990799.01234568,
+      "logps/chosen": -314.7726463607595,
+      "logps/rejected": -207.06107735339506,
+      "loss": 1.3568,
+      "loss/base_kto": 0.4983139932155609,
+      "metrics/advantage_var": 0.11646153777837753,
+      "regularization/grad_norm_sq": 3889.73095703125,
+      "regularization/w2": 0.8536549806594849,
+      "rewards/chosen": 0.022073296051991136,
+      "rewards/margins": 0.013807725693997339,
+      "rewards/rejected": 0.008265570357993798,
+      "stability/repetition_rate_mean": 0.19374233484268188,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 31.1264705657959,
+      "stability/response_length_var": 968.8571166992188,
+      "stability/token_entropy_mean": 3.724609375,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.6097723841667175,
+      "kl": 0.19923549890518188,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54706169.08108108,
+      "logits/rejected": -50907100.27906977,
+      "logps/chosen": -359.5902396537162,
+      "logps/rejected": -216.37990552325581,
+      "loss": 1.3626,
+      "loss/base_kto": 0.4989898204803467,
+      "metrics/advantage_var": 0.1130441352725029,
+      "regularization/grad_norm_sq": 3821.28564453125,
+      "regularization/w2": 0.8549612164497375,
+      "rewards/chosen": 0.027993318196889515,
+      "rewards/margins": 0.008093704561404954,
+      "rewards/rejected": 0.01989961363548456,
+      "stability/repetition_rate_mean": 0.23982319235801697,
+      "stability/response_length_mean": 38.0,
+      "stability/response_length_std": 30.10457992553711,
+      "stability/response_length_var": 906.2857055664062,
+      "stability/token_entropy_mean": 4.146484375,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6825862526893616,
+      "kl": 0.21407940983772278,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59382716.144578315,
+      "logits/rejected": -53816409.766233765,
+      "logps/chosen": -399.5208960843373,
+      "logps/rejected": -209.84342025162337,
+      "loss": 1.3765,
+      "loss/base_kto": 0.4972083270549774,
+      "metrics/advantage_var": 0.20788688957691193,
+      "regularization/grad_norm_sq": 4485.09521484375,
+      "regularization/w2": 0.9070795178413391,
+      "rewards/chosen": 0.03815392413771296,
+      "rewards/margins": 0.022114001497791844,
+      "rewards/rejected": 0.016039922639921114,
+      "stability/repetition_rate_mean": 0.1741071343421936,
+      "stability/response_length_mean": 28.0,
+      "stability/response_length_std": 28.665559768676758,
+      "stability/response_length_var": 821.7142944335938,
+      "stability/token_entropy_mean": 3.37109375,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.6473393440246582,
+      "kl": 0.2727767527103424,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57382239.61445783,
+      "logits/rejected": -53606935.27272727,
+      "logps/chosen": -378.179499246988,
+      "logps/rejected": -205.8876572646104,
+      "loss": 1.3516,
+      "loss/base_kto": 0.49749407172203064,
+      "metrics/advantage_var": 0.22285981476306915,
+      "regularization/grad_norm_sq": 3648.669189453125,
+      "regularization/w2": 0.8446584939956665,
+      "rewards/chosen": 0.03837782790861934,
+      "rewards/margins": 0.01832955173811169,
+      "rewards/rejected": 0.020048276170507653,
+      "stability/repetition_rate_mean": 0.3225615620613098,
+      "stability/response_length_mean": 23.625,
+      "stability/response_length_std": 26.338117599487305,
+      "stability/response_length_var": 693.6964111328125,
+      "stability/token_entropy_mean": 4.37109375,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.6492971181869507,
+      "kl": 0.31304872035980225,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60607565.7721519,
+      "logits/rejected": -53416308.938271604,
+      "logps/chosen": -364.08957179588606,
+      "logps/rejected": -200.99373070987653,
+      "loss": 1.3496,
+      "loss/base_kto": 0.4962143898010254,
+      "metrics/advantage_var": 0.24412627518177032,
+      "regularization/grad_norm_sq": 3345.042724609375,
+      "regularization/w2": 0.8016390204429626,
+      "rewards/chosen": 0.050295352935791016,
+      "rewards/margins": 0.030011934998594683,
+      "rewards/rejected": 0.020283417937196332,
+      "stability/repetition_rate_mean": 0.2233072966337204,
+      "stability/response_length_mean": 49.25,
+      "stability/response_length_std": 27.316923141479492,
+      "stability/response_length_var": 746.2142944335938,
+      "stability/token_entropy_mean": 4.751953125,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.6205304861068726,
+      "kl": 0.3582621216773987,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57740241.150326796,
+      "logits/rejected": -54358218.34730539,
+      "logps/chosen": -383.6830065359477,
+      "logps/rejected": -211.94082148203591,
+      "loss": 1.3724,
+      "loss/base_kto": 0.49640893936157227,
+      "metrics/advantage_var": 0.4239996075630188,
+      "regularization/grad_norm_sq": 4084.145263671875,
+      "regularization/w2": 0.876289963722229,
+      "rewards/chosen": 0.05619120753668492,
+      "rewards/margins": 0.028636307190282997,
+      "rewards/rejected": 0.02755490034640192,
+      "stability/repetition_rate_mean": 0.16154514253139496,
+      "stability/response_length_mean": 27.625,
+      "stability/response_length_std": 30.486238479614258,
+      "stability/response_length_var": 929.4107055664062,
+      "stability/token_entropy_mean": 4.673828125,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.6123977899551392,
+      "kl": 0.39068296551704407,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60689534.24657534,
+      "logits/rejected": -55647396.78160919,
+      "logps/chosen": -416.7543343321918,
+      "logps/rejected": -216.09976652298852,
+      "loss": 1.3496,
+      "loss/base_kto": 0.4957490861415863,
+      "metrics/advantage_var": 0.5870842933654785,
+      "regularization/grad_norm_sq": 3718.946044921875,
+      "regularization/w2": 0.8455937504768372,
+      "rewards/chosen": 0.06249639432724208,
+      "rewards/margins": 0.035753407078152694,
+      "rewards/rejected": 0.026742987249089384,
+      "stability/repetition_rate_mean": 0.16215278208255768,
+      "stability/response_length_mean": 15.875,
+      "stability/response_length_std": 24.584766387939453,
+      "stability/response_length_var": 604.4107055664062,
+      "stability/token_entropy_mean": 3.48828125,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.592140793800354,
+      "kl": 0.4606175422668457,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57394050.477419354,
+      "logits/rejected": -53828483.878787875,
+      "logps/chosen": -391.94445564516127,
+      "logps/rejected": -183.64746685606062,
+      "loss": 1.3539,
+      "loss/base_kto": 0.4944417178630829,
+      "metrics/advantage_var": 0.848583996295929,
+      "regularization/grad_norm_sq": 3812.23291015625,
+      "regularization/w2": 0.8568418622016907,
+      "rewards/chosen": 0.07134679978893649,
+      "rewards/margins": 0.044202971714799116,
+      "rewards/rejected": 0.027143828074137368,
+      "stability/repetition_rate_mean": 0.26218748092651367,
+      "stability/response_length_mean": 43.25,
+      "stability/response_length_std": 29.620214462280273,
+      "stability/response_length_var": 877.3571166992188,
+      "stability/token_entropy_mean": 3.6240234375,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5895723700523376,
+      "kl": 0.5093870759010315,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59838715.288343556,
+      "logits/rejected": -54993019.92356688,
+      "logps/chosen": -400.97141008435585,
+      "logps/rejected": -247.50482683121018,
+      "loss": 1.3412,
+      "loss/base_kto": 0.4939149022102356,
+      "metrics/advantage_var": 1.0497407913208008,
+      "regularization/grad_norm_sq": 4162.27978515625,
+      "regularization/w2": 0.889818012714386,
+      "rewards/chosen": 0.08447715548649887,
+      "rewards/margins": 0.04788554667364432,
+      "rewards/rejected": 0.036591608812854545,
+      "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": 4.080078125,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.7277541160583496,
+      "kl": 0.4901605546474457,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57087714.52121212,
+      "logits/rejected": -50078577.96129032,
+      "logps/chosen": -366.3832623106061,
+      "logps/rejected": -189.90645161290323,
+      "loss": 1.3743,
+      "loss/base_kto": 0.4920271039009094,
+      "metrics/advantage_var": 1.8927977085113525,
+      "regularization/grad_norm_sq": 5975.46728515625,
+      "regularization/w2": 0.9738343358039856,
+      "rewards/chosen": 0.08064722003358783,
+      "rewards/margins": 0.06784514984660368,
+      "rewards/rejected": 0.012802070186984154,
+      "stability/repetition_rate_mean": 0.3125,
+      "stability/response_length_mean": 56.25,
+      "stability/response_length_std": 21.920310974121094,
+      "stability/response_length_var": 480.5,
+      "stability/token_entropy_mean": 4.9853515625,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.6773489713668823,
+      "kl": 0.5782578587532043,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51145351.71084338,
+      "logits/rejected": -51668925.50649351,
+      "logps/chosen": -329.48458678463857,
+      "logps/rejected": -199.14678875811688,
+      "loss": 1.3404,
+      "loss/base_kto": 0.49439021944999695,
+      "metrics/advantage_var": 0.7935466766357422,
+      "regularization/grad_norm_sq": 3477.833251953125,
+      "regularization/w2": 0.8221335411071777,
+      "rewards/chosen": 0.07872245972415051,
+      "rewards/margins": 0.044609513944573034,
+      "rewards/rejected": 0.03411294577957748,
+      "stability/repetition_rate_mean": 0.27182701230049133,
+      "stability/response_length_mean": 35.125,
+      "stability/response_length_std": 31.37076759338379,
+      "stability/response_length_var": 984.125,
+      "stability/token_entropy_mean": 5.19921875,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.6416272521018982,
+      "kl": 0.6353618502616882,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55416663.34117647,
+      "logits/rejected": -50182574.08,
+      "logps/chosen": -396.80445772058823,
+      "logps/rejected": -207.3536328125,
+      "loss": 1.3548,
+      "loss/base_kto": 0.490011066198349,
+      "metrics/advantage_var": 1.2986860275268555,
+      "regularization/grad_norm_sq": 4135.36962890625,
+      "regularization/w2": 0.900852382183075,
+      "rewards/chosen": 0.1129976609173943,
+      "rewards/margins": 0.07576675284142587,
+      "rewards/rejected": 0.037230908075968426,
+      "stability/repetition_rate_mean": 0.20136336982250214,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.193626403808594,
+      "stability/response_length_var": 852.2678833007812,
+      "stability/token_entropy_mean": 4.65625,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.6257898211479187,
+      "kl": 0.6446139216423035,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54021403.105882354,
+      "logits/rejected": -49364510.72,
+      "logps/chosen": -360.11148897058825,
+      "logps/rejected": -191.02584635416667,
+      "loss": 1.3426,
+      "loss/base_kto": 0.49283477663993835,
+      "metrics/advantage_var": 2.1838507652282715,
+      "regularization/grad_norm_sq": 4965.29248046875,
+      "regularization/w2": 0.9337390065193176,
+      "rewards/chosen": 0.0877797295065487,
+      "rewards/margins": 0.059804205239987834,
+      "rewards/rejected": 0.027975524266560872,
+      "stability/repetition_rate_mean": 0.3111416697502136,
+      "stability/response_length_mean": 56.5,
+      "stability/response_length_std": 10.928338050842285,
+      "stability/response_length_var": 119.42857360839844,
+      "stability/token_entropy_mean": 4.533203125,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.6220570206642151,
+      "kl": 0.7235702872276306,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55247811.047619045,
+      "logits/rejected": -48958012.631578945,
+      "logps/chosen": -415.5559895833333,
+      "logps/rejected": -186.48792146381578,
+      "loss": 1.357,
+      "loss/base_kto": 0.48817262053489685,
+      "metrics/advantage_var": 1.9768089056015015,
+      "regularization/grad_norm_sq": 3970.04931640625,
+      "regularization/w2": 0.8730607032775879,
+      "rewards/chosen": 0.13393676848638625,
+      "rewards/margins": 0.0929941403537167,
+      "rewards/rejected": 0.04094262813266955,
+      "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": 4.48828125,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.7207323908805847,
+      "kl": 0.7536167502403259,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54424352.36781609,
+      "logits/rejected": -49791210.95890411,
+      "logps/chosen": -369.7562185704023,
+      "logps/rejected": -212.1244381421233,
+      "loss": 1.3489,
+      "loss/base_kto": 0.4913412034511566,
+      "metrics/advantage_var": 1.7532039880752563,
+      "regularization/grad_norm_sq": 3547.708984375,
+      "regularization/w2": 0.8307213187217712,
+      "rewards/chosen": 0.11605137792126886,
+      "rewards/margins": 0.06928389490241987,
+      "rewards/rejected": 0.046767483018849,
+      "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": 4.88671875,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.6222322583198547,
+      "kl": 0.920440137386322,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59139877.86982249,
+      "logits/rejected": -52581216.63576159,
+      "logps/chosen": -439.6777921597633,
+      "logps/rejected": -190.6081332781457,
+      "loss": 1.3222,
+      "loss/base_kto": 0.48742732405662537,
+      "metrics/advantage_var": 2.3735947608947754,
+      "regularization/grad_norm_sq": 4127.0400390625,
+      "regularization/w2": 0.8687969446182251,
+      "rewards/chosen": 0.16640804223054964,
+      "rewards/margins": 0.10143800048816165,
+      "rewards/rejected": 0.06497004174238799,
+      "stability/repetition_rate_mean": 0.23209978640079498,
+      "stability/response_length_mean": 36.25,
+      "stability/response_length_std": 30.76988410949707,
+      "stability/response_length_var": 946.7857055664062,
+      "stability/token_entropy_mean": 3.67041015625,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.7240987420082092,
+      "kl": 0.9491209983825684,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56657488.192771085,
+      "logits/rejected": -48037266.28571428,
+      "logps/chosen": -354.8600809487952,
+      "logps/rejected": -194.32765574269482,
+      "loss": 1.3334,
+      "loss/base_kto": 0.4863254725933075,
+      "metrics/advantage_var": 4.277890682220459,
+      "regularization/grad_norm_sq": 4247.8359375,
+      "regularization/w2": 0.8748798370361328,
+      "rewards/chosen": 0.15770882296274943,
+      "rewards/margins": 0.10972130923479342,
+      "rewards/rejected": 0.047987513727956004,
+      "stability/repetition_rate_mean": 0.18672296404838562,
+      "stability/response_length_mean": 29.75,
+      "stability/response_length_std": 26.015106201171875,
+      "stability/response_length_var": 676.7857055664062,
+      "stability/token_entropy_mean": 4.822265625,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.6430587768554688,
+      "kl": 1.032440423965454,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52647628.17177914,
+      "logits/rejected": -47983459.46496815,
+      "logps/chosen": -374.4316765720859,
+      "logps/rejected": -210.53918690286625,
+      "loss": 1.321,
+      "loss/base_kto": 0.48797211050987244,
+      "metrics/advantage_var": 4.151429653167725,
+      "regularization/grad_norm_sq": 3608.227294921875,
+      "regularization/w2": 0.8262709975242615,
+      "rewards/chosen": 0.17141536408406824,
+      "rewards/margins": 0.09551231590283814,
+      "rewards/rejected": 0.0759030481812301,
+      "stability/repetition_rate_mean": 0.2868303656578064,
+      "stability/response_length_mean": 35.25,
+      "stability/response_length_std": 31.129911422729492,
+      "stability/response_length_var": 969.0714111328125,
+      "stability/token_entropy_mean": 3.130859375,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.6488748788833618,
+      "kl": 1.0751243829727173,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54614894.61728395,
+      "logits/rejected": -49520795.5443038,
+      "logps/chosen": -310.31404320987656,
+      "logps/rejected": -196.53752966772151,
+      "loss": 1.3781,
+      "loss/base_kto": 0.4869016706943512,
+      "metrics/advantage_var": 7.557289123535156,
+      "regularization/grad_norm_sq": 7128.36279296875,
+      "regularization/w2": 0.9516963958740234,
+      "rewards/chosen": 0.16873404420452354,
+      "rewards/margins": 0.1185720879727182,
+      "rewards/rejected": 0.05016195623180534,
+      "stability/repetition_rate_mean": 0.2109375,
+      "stability/response_length_mean": 32.875,
+      "stability/response_length_std": 33.30996322631836,
+      "stability/response_length_var": 1109.5535888671875,
+      "stability/token_entropy_mean": 3.71875,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.6966060996055603,
+      "kl": 1.1208827495574951,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53095589.556886226,
+      "logits/rejected": -49707021.385620914,
+      "logps/chosen": -390.324967252994,
+      "logps/rejected": -210.44881025326796,
+      "loss": 1.3626,
+      "loss/base_kto": 0.4836917519569397,
+      "metrics/advantage_var": 7.099040985107422,
+      "regularization/grad_norm_sq": 5476.40771484375,
+      "regularization/w2": 0.9177618026733398,
+      "rewards/chosen": 0.20803529202581167,
+      "rewards/margins": 0.1330566630569308,
+      "rewards/rejected": 0.07497862896888084,
+      "stability/repetition_rate_mean": 0.2964409589767456,
+      "stability/response_length_mean": 20.375,
+      "stability/response_length_std": 27.645395278930664,
+      "stability/response_length_var": 764.2678833007812,
+      "stability/token_entropy_mean": 4.3505859375,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.6411679983139038,
+      "kl": 1.1453860998153687,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60050742.5032258,
+      "logits/rejected": -53805540.07272727,
+      "logps/chosen": -390.8772681451613,
+      "logps/rejected": -197.0868252840909,
+      "loss": 1.3508,
+      "loss/base_kto": 0.48641711473464966,
+      "metrics/advantage_var": 4.3691325187683105,
+      "regularization/grad_norm_sq": 4039.490966796875,
+      "regularization/w2": 0.8582512736320496,
+      "rewards/chosen": 0.19076536855389994,
+      "rewards/margins": 0.1075151782814242,
+      "rewards/rejected": 0.08325019027247574,
+      "stability/repetition_rate_mean": 0.3895089328289032,
+      "stability/response_length_mean": 49.0,
+      "stability/response_length_std": 27.82085609436035,
+      "stability/response_length_var": 774.0,
+      "stability/token_entropy_mean": 5.19921875,
+      "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_w2_delta0.01/checkpoint-196/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9125ddbb4efe23bb499afbabc466a81ea7d184e9e0ac5bd7dc5acdf9cbac0b42
+size 8081

mistral7b_kto_w2_delta0.01/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_w2_delta0.01/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_w2_delta0.01/stability_metrics.jsonl

@@ -0,0 +1,38 @@
+{"stability/response_length_mean": 46.125, "stability/response_length_std": 27.231481552124023, "stability/response_length_var": 741.5535888671875, "stability/token_entropy_mean": 3.27001953125, "stability/repetition_rate_mean": 0.3629409670829773, "step": 10}
+{"stability/response_length_mean": 29.375, "stability/response_length_std": 29.92341423034668, "stability/response_length_var": 895.4107055664062, "stability/token_entropy_mean": 3.7001953125, "stability/repetition_rate_mean": 0.2929687201976776, "step": 15}
+{"stability/response_length_mean": 44.0, "stability/response_length_std": 28.0611572265625, "stability/response_length_var": 787.4285888671875, "stability/token_entropy_mean": 4.4140625, "stability/repetition_rate_mean": 0.2155257910490036, "step": 20}
+{"stability/response_length_mean": 43.375, "stability/response_length_std": 28.54538917541504, "stability/response_length_var": 814.8392944335938, "stability/token_entropy_mean": 4.8125, "stability/repetition_rate_mean": 0.17215879261493683, "step": 25}
+{"stability/response_length_mean": 43.0, "stability/response_length_std": 29.602123260498047, "stability/response_length_var": 876.2857055664062, "stability/token_entropy_mean": 4.61328125, "stability/repetition_rate_mean": 0.1427641361951828, "step": 30}
+{"stability/response_length_mean": 56.875, "stability/response_length_std": 20.152544021606445, "stability/response_length_var": 406.125, "stability/token_entropy_mean": 5.419921875, "stability/repetition_rate_mean": 0.3166852593421936, "step": 35}
+{"stability/response_length_mean": 33.125, "stability/response_length_std": 33.01704025268555, "stability/response_length_var": 1090.125, "stability/token_entropy_mean": 4.8935546875, "stability/repetition_rate_mean": 0.13671875, "step": 40}
+{"stability/response_length_mean": 28.875, "stability/response_length_std": 29.87324333190918, "stability/response_length_var": 892.4107055664062, "stability/token_entropy_mean": 4.7578125, "stability/repetition_rate_mean": 0.0994017943739891, "step": 45}
+{"stability/response_length_mean": 53.375, "stability/response_length_std": 21.480472564697266, "stability/response_length_var": 461.41070556640625, "stability/token_entropy_mean": 4.1162109375, "stability/repetition_rate_mean": 0.2952169179916382, "step": 50}
+{"stability/response_length_mean": 22.375, "stability/response_length_std": 30.236862182617188, "stability/response_length_var": 914.2678833007812, "stability/token_entropy_mean": 4.302734375, "stability/repetition_rate_mean": 0.11589926481246948, "step": 55}
+{"stability/response_length_mean": 44.625, "stability/response_length_std": 27.8820858001709, "stability/response_length_var": 777.4107055664062, "stability/token_entropy_mean": 4.607421875, "stability/repetition_rate_mean": 0.17885693907737732, "step": 60}
+{"stability/response_length_mean": 20.625, "stability/response_length_std": 27.176342010498047, "stability/response_length_var": 738.5535888671875, "stability/token_entropy_mean": 5.525390625, "stability/repetition_rate_mean": 0.1857219785451889, "step": 65}
+{"stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.77734375, "stability/repetition_rate_mean": 0.162109375, "step": 70}
+{"stability/response_length_mean": 22.75, "stability/response_length_std": 31.07479476928711, "stability/response_length_var": 965.6428833007812, "stability/token_entropy_mean": 2.15625, "stability/repetition_rate_mean": 0.08308292925357819, "step": 75}
+{"stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.2265625, "stability/repetition_rate_mean": 0.19088542461395264, "step": 80}
+{"stability/response_length_mean": 32.375, "stability/response_length_std": 33.547142028808594, "stability/response_length_var": 1125.4107666015625, "stability/token_entropy_mean": 4.869140625, "stability/repetition_rate_mean": 0.1218377947807312, "step": 85}
+{"stability/response_length_mean": 26.875, "stability/response_length_std": 27.544445037841797, "stability/response_length_var": 758.6964111328125, "stability/token_entropy_mean": 4.3203125, "stability/repetition_rate_mean": 0.2843276560306549, "step": 90}
+{"stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.724609375, "stability/repetition_rate_mean": 0.19374233484268188, "step": 95}
+{"stability/response_length_mean": 38.0, "stability/response_length_std": 30.10457992553711, "stability/response_length_var": 906.2857055664062, "stability/token_entropy_mean": 4.146484375, "stability/repetition_rate_mean": 0.23982319235801697, "step": 100}
+{"stability/response_length_mean": 28.0, "stability/response_length_std": 28.665559768676758, "stability/response_length_var": 821.7142944335938, "stability/token_entropy_mean": 3.37109375, "stability/repetition_rate_mean": 0.1741071343421936, "step": 105}
+{"stability/response_length_mean": 23.625, "stability/response_length_std": 26.338117599487305, "stability/response_length_var": 693.6964111328125, "stability/token_entropy_mean": 4.37109375, "stability/repetition_rate_mean": 0.3225615620613098, "step": 110}
+{"stability/response_length_mean": 49.25, "stability/response_length_std": 27.316923141479492, "stability/response_length_var": 746.2142944335938, "stability/token_entropy_mean": 4.751953125, "stability/repetition_rate_mean": 0.2233072966337204, "step": 115}
+{"stability/response_length_mean": 27.625, "stability/response_length_std": 30.486238479614258, "stability/response_length_var": 929.4107055664062, "stability/token_entropy_mean": 4.673828125, "stability/repetition_rate_mean": 0.16154514253139496, "step": 120}
+{"stability/response_length_mean": 15.875, "stability/response_length_std": 24.584766387939453, "stability/response_length_var": 604.4107055664062, "stability/token_entropy_mean": 3.48828125, "stability/repetition_rate_mean": 0.16215278208255768, "step": 125}
+{"stability/response_length_mean": 43.25, "stability/response_length_std": 29.620214462280273, "stability/response_length_var": 877.3571166992188, "stability/token_entropy_mean": 3.6240234375, "stability/repetition_rate_mean": 0.26218748092651367, "step": 130}
+{"stability/response_length_mean": 29.25, "stability/response_length_std": 29.936838150024414, "stability/response_length_var": 896.2142944335938, "stability/token_entropy_mean": 4.080078125, "stability/repetition_rate_mean": 0.2230902761220932, "step": 135}
+{"stability/response_length_mean": 56.25, "stability/response_length_std": 21.920310974121094, "stability/response_length_var": 480.5, "stability/token_entropy_mean": 4.9853515625, "stability/repetition_rate_mean": 0.3125, "step": 140}
+{"stability/response_length_mean": 35.125, "stability/response_length_std": 31.37076759338379, "stability/response_length_var": 984.125, "stability/token_entropy_mean": 5.19921875, "stability/repetition_rate_mean": 0.27182701230049133, "step": 145}
+{"stability/response_length_mean": 29.375, "stability/response_length_std": 29.193626403808594, "stability/response_length_var": 852.2678833007812, "stability/token_entropy_mean": 4.65625, "stability/repetition_rate_mean": 0.20136336982250214, "step": 150}
+{"stability/response_length_mean": 56.5, "stability/response_length_std": 10.928338050842285, "stability/response_length_var": 119.42857360839844, "stability/token_entropy_mean": 4.533203125, "stability/repetition_rate_mean": 0.3111416697502136, "step": 155}
+{"stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.48828125, "stability/repetition_rate_mean": 0.30851560831069946, "step": 160}
+{"stability/response_length_mean": 22.5, "stability/response_length_std": 28.33977508544922, "stability/response_length_var": 803.1428833007812, "stability/token_entropy_mean": 4.88671875, "stability/repetition_rate_mean": 0.2411106526851654, "step": 165}
+{"stability/response_length_mean": 36.25, "stability/response_length_std": 30.76988410949707, "stability/response_length_var": 946.7857055664062, "stability/token_entropy_mean": 3.67041015625, "stability/repetition_rate_mean": 0.23209978640079498, "step": 170}
+{"stability/response_length_mean": 29.75, "stability/response_length_std": 26.015106201171875, "stability/response_length_var": 676.7857055664062, "stability/token_entropy_mean": 4.822265625, "stability/repetition_rate_mean": 0.18672296404838562, "step": 175}
+{"stability/response_length_mean": 35.25, "stability/response_length_std": 31.129911422729492, "stability/response_length_var": 969.0714111328125, "stability/token_entropy_mean": 3.130859375, "stability/repetition_rate_mean": 0.2868303656578064, "step": 180}
+{"stability/response_length_mean": 32.875, "stability/response_length_std": 33.30996322631836, "stability/response_length_var": 1109.5535888671875, "stability/token_entropy_mean": 3.71875, "stability/repetition_rate_mean": 0.2109375, "step": 185}
+{"stability/response_length_mean": 20.375, "stability/response_length_std": 27.645395278930664, "stability/response_length_var": 764.2678833007812, "stability/token_entropy_mean": 4.3505859375, "stability/repetition_rate_mean": 0.2964409589767456, "step": 190}
+{"stability/response_length_mean": 49.0, "stability/response_length_std": 27.82085609436035, "stability/response_length_var": 774.0, "stability/token_entropy_mean": 5.19921875, "stability/repetition_rate_mean": 0.3895089328289032, "step": 195}

mistral7b_kto_w2_delta0.01/tokenizer.model

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

mistral7b_kto_w2_delta0.01/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_w2_delta0.01/train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 1.359869228333843,
+    "train_runtime": 4299.7361,
+    "train_samples_per_second": 23.257,
+    "train_steps_per_second": 0.046
+}

mistral7b_kto_w2_delta0.01/trainer_log.jsonl

@@ -0,0 +1,40 @@
+{"current_steps": 5, "total_steps": 196, "loss": 1.3863, "lr": 4e-08, "epoch": 0.025591810620601407, "percentage": 2.55, "elapsed_time": "0:01:47", "remaining_time": "1:08:27", "rewards/chosen": 0.0017119320599656355, "logps/chosen": -439.02847450657896, "logits/chosen": -59717241.2631579, "rewards/rejected": 0.00017804758889334544, "logps/rejected": -198.40762183779762, "logits/rejected": -50415911.61904762, "rewards/margins": 0.00153388447107229, "regularization/w2": 0.9065223932266235, "regularization/grad_norm_sq": 4328.5595703125, "kl": 0.008433527313172817}
+{"current_steps": 10, "total_steps": 196, "loss": 1.3645, "lr": 9e-08, "epoch": 0.05118362124120281, "percentage": 5.1, "elapsed_time": "0:03:37", "remaining_time": "1:07:28", "rewards/chosen": 0.0015573305162516508, "logps/chosen": -429.92294034090907, "logits/chosen": -59293257.14285714, "rewards/rejected": 0.002358142510954156, "logps/rejected": -192.84436182228916, "logits/rejected": -50017699.469879515, "rewards/margins": -0.0008008119947025051, "regularization/w2": 0.8646374940872192, "regularization/grad_norm_sq": 3843.479736328125, "kl": 0.019305892288684845, "stability/response_length_mean": 46.125, "stability/response_length_std": 27.231481552124023, "stability/response_length_var": 741.5535888671875, "stability/token_entropy_mean": 3.27001953125, "stability/repetition_rate_mean": 0.3629409670829773}
+{"current_steps": 15, "total_steps": 196, "loss": 1.3548, "lr": 1.4e-07, "epoch": 0.07677543186180422, "percentage": 7.65, "elapsed_time": "0:05:26", "remaining_time": "1:05:36", "rewards/chosen": -7.632036100734364e-05, "logps/chosen": -404.1315814393939, "logits/chosen": -56229807.32121212, "rewards/rejected": 0.0023395276838733305, "logps/rejected": -200.67066532258065, "logits/rejected": -51916106.32258064, "rewards/margins": -0.0024158480448806742, "regularization/w2": 0.8843773007392883, "regularization/grad_norm_sq": 4172.4462890625, "kl": 0.013482200913131237, "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": 1.3654, "lr": 1.8999999999999998e-07, "epoch": 0.10236724248240563, "percentage": 10.2, "elapsed_time": "0:07:17", "remaining_time": "1:04:08", "rewards/chosen": 0.00021410472691059114, "logps/chosen": -343.694580078125, "logits/chosen": -51664345.6, "rewards/rejected": 0.0015568394213914872, "logps/rejected": -225.647802734375, "logits/rejected": -48065900.8, "rewards/margins": -0.0013427346944808961, "regularization/w2": 0.843424916267395, "regularization/grad_norm_sq": 3665.234375, "kl": 0.018938787281513214, "stability/response_length_mean": 44.0, "stability/response_length_std": 28.0611572265625, "stability/response_length_var": 787.4285888671875, "stability/token_entropy_mean": 4.4140625, "stability/repetition_rate_mean": 0.2155257910490036}
+{"current_steps": 25, "total_steps": 196, "loss": 1.3771, "lr": 2.4e-07, "epoch": 0.12795905310300704, "percentage": 12.76, "elapsed_time": "0:09:06", "remaining_time": "1:02:15", "rewards/chosen": 0.0029272856383488095, "logps/chosen": -378.4621767241379, "logits/chosen": -56488998.84137931, "rewards/rejected": 0.002139153821127755, "logps/rejected": -202.77145089285713, "logits/rejected": -52022798.62857143, "rewards/margins": 0.0007881318172210544, "regularization/w2": 0.860791027545929, "regularization/grad_norm_sq": 4001.77783203125, "kl": 0.02769310213625431, "stability/response_length_mean": 43.375, "stability/response_length_std": 28.54538917541504, "stability/response_length_var": 814.8392944335938, "stability/token_entropy_mean": 4.8125, "stability/repetition_rate_mean": 0.17215879261493683}
+{"current_steps": 30, "total_steps": 196, "loss": 1.3694, "lr": 2.9e-07, "epoch": 0.15355086372360843, "percentage": 15.31, "elapsed_time": "0:10:55", "remaining_time": "1:00:26", "rewards/chosen": -0.0007460268351855694, "logps/chosen": -468.588716442953, "logits/chosen": -57729649.39597315, "rewards/rejected": 2.2698389856438888e-05, "logps/rejected": -198.24766995614036, "logits/rejected": -51505858.61988304, "rewards/margins": -0.0007687252250420083, "regularization/w2": 0.917149007320404, "regularization/grad_norm_sq": 4544.02490234375, "kl": 0.018374785780906677, "stability/response_length_mean": 43.0, "stability/response_length_std": 29.602123260498047, "stability/response_length_var": 876.2857055664062, "stability/token_entropy_mean": 4.61328125, "stability/repetition_rate_mean": 0.1427641361951828}
+{"current_steps": 35, "total_steps": 196, "loss": 1.3758, "lr": 3.4000000000000003e-07, "epoch": 0.17914267434420986, "percentage": 17.86, "elapsed_time": "0:12:45", "remaining_time": "0:58:39", "rewards/chosen": 0.003366763833202893, "logps/chosen": -385.90286293512656, "logits/chosen": -57635956.658227846, "rewards/rejected": 0.0007024767498175303, "logps/rejected": -195.11945408950618, "logits/rejected": -53888758.518518515, "rewards/margins": 0.002664287083385363, "regularization/w2": 0.8619928359985352, "regularization/grad_norm_sq": 3807.059326171875, "kl": 0.02894224040210247, "stability/response_length_mean": 56.875, "stability/response_length_std": 20.152544021606445, "stability/response_length_var": 406.125, "stability/token_entropy_mean": 5.419921875, "stability/repetition_rate_mean": 0.3166852593421936}
+{"current_steps": 40, "total_steps": 196, "loss": 1.3686, "lr": 3.8999999999999997e-07, "epoch": 0.20473448496481125, "percentage": 20.41, "elapsed_time": "0:14:35", "remaining_time": "0:56:54", "rewards/chosen": 0.0030031204223632812, "logps/chosen": -421.87265625, "logits/chosen": -56634796.8, "rewards/rejected": 0.003615342453122139, "logps/rejected": -219.273095703125, "logits/rejected": -50219603.2, "rewards/margins": -0.0006122220307588576, "regularization/w2": 0.9105228781700134, "regularization/grad_norm_sq": 4305.833984375, "kl": 0.02706434763967991, "stability/response_length_mean": 33.125, "stability/response_length_std": 33.01704025268555, "stability/response_length_var": 1090.125, "stability/token_entropy_mean": 4.8935546875, "stability/repetition_rate_mean": 0.13671875}
+{"current_steps": 45, "total_steps": 196, "loss": 1.379, "lr": 4.3999999999999997e-07, "epoch": 0.23032629558541268, "percentage": 22.96, "elapsed_time": "0:16:25", "remaining_time": "0:55:07", "rewards/chosen": 0.004526910365827932, "logps/chosen": -352.2831900167785, "logits/chosen": -52699170.36241611, "rewards/rejected": 0.0023026952618046812, "logps/rejected": -243.85115131578948, "logits/rejected": -50773297.403508775, "rewards/margins": 0.0022242151040232504, "regularization/w2": 0.8400061726570129, "regularization/grad_norm_sq": 3761.641357421875, "kl": 0.03937375545501709, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.87324333190918, "stability/response_length_var": 892.4107055664062, "stability/token_entropy_mean": 4.7578125, "stability/repetition_rate_mean": 0.0994017943739891}
+{"current_steps": 50, "total_steps": 196, "loss": 1.3394, "lr": 4.9e-07, "epoch": 0.2559181062060141, "percentage": 25.51, "elapsed_time": "0:18:14", "remaining_time": "0:53:17", "rewards/chosen": 0.007330068805873794, "logps/chosen": -375.1258389261745, "logits/chosen": -54044414.281879194, "rewards/rejected": 0.005857517844752262, "logps/rejected": -180.18202896564327, "logits/rejected": -48299597.847953215, "rewards/margins": 0.0014725509611215326, "regularization/w2": 0.8510723114013672, "regularization/grad_norm_sq": 3838.28369140625, "kl": 0.047937434166669846, "stability/response_length_mean": 53.375, "stability/response_length_std": 21.480472564697266, "stability/response_length_var": 461.41070556640625, "stability/token_entropy_mean": 4.1162109375, "stability/repetition_rate_mean": 0.2952169179916382}
+{"current_steps": 55, "total_steps": 196, "loss": 1.357, "lr": 5.4e-07, "epoch": 0.28150991682661547, "percentage": 28.06, "elapsed_time": "0:20:03", "remaining_time": "0:51:26", "rewards/chosen": 0.0022328466362094286, "logps/chosen": -344.48966420807454, "logits/chosen": -56949887.204968944, "rewards/rejected": 0.00506415344634146, "logps/rejected": -230.39740566037736, "logits/rejected": -49557458.91823899, "rewards/margins": -0.002831306810132032, "regularization/w2": 0.8724905252456665, "regularization/grad_norm_sq": 3955.544921875, "kl": 0.0575476698577404, "stability/response_length_mean": 22.375, "stability/response_length_std": 30.236862182617188, "stability/response_length_var": 914.2678833007812, "stability/token_entropy_mean": 4.302734375, "stability/repetition_rate_mean": 0.11589926481246948}
+{"current_steps": 60, "total_steps": 196, "loss": 1.3664, "lr": 5.9e-07, "epoch": 0.30710172744721687, "percentage": 30.61, "elapsed_time": "0:21:54", "remaining_time": "0:49:40", "rewards/chosen": 0.010271583903919567, "logps/chosen": -359.6180871212121, "logits/chosen": -52677849.21212121, "rewards/rejected": 0.006947755813598633, "logps/rejected": -178.15743447580644, "logits/rejected": -47824090.01290323, "rewards/margins": 0.003323828090320934, "regularization/w2": 0.7870045900344849, "regularization/grad_norm_sq": 3162.57421875, "kl": 0.05734771117568016, "stability/response_length_mean": 44.625, "stability/response_length_std": 27.8820858001709, "stability/response_length_var": 777.4107055664062, "stability/token_entropy_mean": 4.607421875, "stability/repetition_rate_mean": 0.17885693907737732}
+{"current_steps": 65, "total_steps": 196, "loss": 1.3773, "lr": 6.4e-07, "epoch": 0.3326935380678183, "percentage": 33.16, "elapsed_time": "0:23:45", "remaining_time": "0:47:52", "rewards/chosen": 0.009909362177694997, "logps/chosen": -381.8476814516129, "logits/chosen": -57565778.58064516, "rewards/rejected": 0.009872553565285422, "logps/rejected": -183.48361742424242, "logits/rejected": -47265906.81212121, "rewards/margins": 3.680861240957546e-05, "regularization/w2": 0.8648955225944519, "regularization/grad_norm_sq": 3887.826904296875, "kl": 0.07911502569913864, "stability/response_length_mean": 20.625, "stability/response_length_std": 27.176342010498047, "stability/response_length_var": 738.5535888671875, "stability/token_entropy_mean": 5.525390625, "stability/repetition_rate_mean": 0.1857219785451889}
+{"current_steps": 70, "total_steps": 196, "loss": 1.403, "lr": 6.9e-07, "epoch": 0.3582853486884197, "percentage": 35.71, "elapsed_time": "0:25:34", "remaining_time": "0:46:01", "rewards/chosen": 0.01227622724348499, "logps/chosen": -379.00569556451615, "logits/chosen": -52773428.8516129, "rewards/rejected": 0.007603588971224698, "logps/rejected": -220.34614109848485, "logits/rejected": -49024763.345454544, "rewards/margins": 0.004672638272260291, "regularization/w2": 0.9129446148872375, "regularization/grad_norm_sq": 4501.79833984375, "kl": 0.09082157909870148, "stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.77734375, "stability/repetition_rate_mean": 0.162109375}
+{"current_steps": 75, "total_steps": 196, "loss": 1.3809, "lr": 7.4e-07, "epoch": 0.3838771593090211, "percentage": 38.27, "elapsed_time": "0:27:24", "remaining_time": "0:44:13", "rewards/chosen": 0.014305324167818637, "logps/chosen": -314.12774493243245, "logits/chosen": -49499239.783783786, "rewards/rejected": 0.005612370579741722, "logps/rejected": -228.82871547965115, "logits/rejected": -47880882.60465116, "rewards/margins": 0.008692953588076915, "regularization/w2": 0.8484571576118469, "regularization/grad_norm_sq": 3738.951171875, "kl": 0.08937091380357742, "stability/response_length_mean": 22.75, "stability/response_length_std": 31.07479476928711, "stability/response_length_var": 965.6428833007812, "stability/token_entropy_mean": 2.15625, "stability/repetition_rate_mean": 0.08308292925357819}
+{"current_steps": 80, "total_steps": 196, "loss": 1.3598, "lr": 7.9e-07, "epoch": 0.4094689699296225, "percentage": 40.82, "elapsed_time": "0:29:15", "remaining_time": "0:42:25", "rewards/chosen": 0.01743706330558322, "logps/chosen": -403.85572744205297, "logits/chosen": -58623972.87417219, "rewards/rejected": 0.00803350132597974, "logps/rejected": -200.97545303254438, "logits/rejected": -53456156.78106509, "rewards/margins": 0.009403561979603481, "regularization/w2": 0.8389098048210144, "regularization/grad_norm_sq": 3606.688232421875, "kl": 0.11913253366947174, "stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.2265625, "stability/repetition_rate_mean": 0.19088542461395264}
+{"current_steps": 85, "total_steps": 196, "loss": 1.3679, "lr": 8.399999999999999e-07, "epoch": 0.4350607805502239, "percentage": 43.37, "elapsed_time": "0:31:04", "remaining_time": "0:40:35", "rewards/chosen": 0.018512418193201863, "logps/chosen": -358.1473790322581, "logits/chosen": -51481953.44516129, "rewards/rejected": 0.0100817651459665, "logps/rejected": -230.32398200757575, "logits/rejected": -47134605.18787879, "rewards/margins": 0.008430653047235363, "regularization/w2": 0.8647522330284119, "regularization/grad_norm_sq": 3847.29296875, "kl": 0.12905550003051758, "stability/response_length_mean": 32.375, "stability/response_length_std": 33.547142028808594, "stability/response_length_var": 1125.4107666015625, "stability/token_entropy_mean": 4.869140625, "stability/repetition_rate_mean": 0.1218377947807312}
+{"current_steps": 90, "total_steps": 196, "loss": 1.3594, "lr": 8.9e-07, "epoch": 0.46065259117082535, "percentage": 45.92, "elapsed_time": "0:32:55", "remaining_time": "0:38:47", "rewards/chosen": 0.025618803806793995, "logps/chosen": -380.533203125, "logits/chosen": -58414395.07692308, "rewards/rejected": 0.009023745612400333, "logps/rejected": -187.13657583841464, "logits/rejected": -50576390.24390244, "rewards/margins": 0.016595058194393664, "regularization/w2": 0.8712137341499329, "regularization/grad_norm_sq": 3978.860595703125, "kl": 0.16841307282447815, "stability/response_length_mean": 26.875, "stability/response_length_std": 27.544445037841797, "stability/response_length_var": 758.6964111328125, "stability/token_entropy_mean": 4.3203125, "stability/repetition_rate_mean": 0.2843276560306549}
+{"current_steps": 95, "total_steps": 196, "loss": 1.3568, "lr": 9.399999999999999e-07, "epoch": 0.48624440179142675, "percentage": 48.47, "elapsed_time": "0:34:45", "remaining_time": "0:36:57", "rewards/chosen": 0.022073296051991136, "logps/chosen": -314.7726463607595, "logits/chosen": -50395103.594936706, "rewards/rejected": 0.008265570357993798, "logps/rejected": -207.06107735339506, "logits/rejected": -48990799.01234568, "rewards/margins": 0.013807725693997339, "regularization/w2": 0.8536549806594849, "regularization/grad_norm_sq": 3889.73095703125, "kl": 0.1684073954820633, "stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.724609375, "stability/repetition_rate_mean": 0.19374233484268188}
+{"current_steps": 100, "total_steps": 196, "loss": 1.3626, "lr": 9.9e-07, "epoch": 0.5118362124120281, "percentage": 51.02, "elapsed_time": "0:36:35", "remaining_time": "0:35:07", "rewards/chosen": 0.027993318196889515, "logps/chosen": -359.5902396537162, "logits/chosen": -54706169.08108108, "rewards/rejected": 0.01989961363548456, "logps/rejected": -216.37990552325581, "logits/rejected": -50907100.27906977, "rewards/margins": 0.008093704561404954, "regularization/w2": 0.8549612164497375, "regularization/grad_norm_sq": 3821.28564453125, "kl": 0.19923549890518188, "stability/response_length_mean": 38.0, "stability/response_length_std": 30.10457992553711, "stability/response_length_var": 906.2857055664062, "stability/token_entropy_mean": 4.146484375, "stability/repetition_rate_mean": 0.23982319235801697}
+{"current_steps": 105, "total_steps": 196, "loss": 1.3765, "lr": 9.999815554351224e-07, "epoch": 0.5374280230326296, "percentage": 53.57, "elapsed_time": "0:38:25", "remaining_time": "0:33:18", "rewards/chosen": 0.03815392413771296, "logps/chosen": -399.5208960843373, "logits/chosen": -59382716.144578315, "rewards/rejected": 0.016039922639921114, "logps/rejected": -209.84342025162337, "logits/rejected": -53816409.766233765, "rewards/margins": 0.022114001497791844, "regularization/w2": 0.9070795178413391, "regularization/grad_norm_sq": 4485.09521484375, "kl": 0.21407940983772278, "stability/response_length_mean": 28.0, "stability/response_length_std": 28.665559768676758, "stability/response_length_var": 821.7142944335938, "stability/token_entropy_mean": 3.37109375, "stability/repetition_rate_mean": 0.1741071343421936}
+{"current_steps": 110, "total_steps": 196, "loss": 1.3516, "lr": 9.999066267225444e-07, "epoch": 0.5630198336532309, "percentage": 56.12, "elapsed_time": "0:40:17", "remaining_time": "0:31:29", "rewards/chosen": 0.03837782790861934, "logps/chosen": -378.179499246988, "logits/chosen": -57382239.61445783, "rewards/rejected": 0.020048276170507653, "logps/rejected": -205.8876572646104, "logits/rejected": -53606935.27272727, "rewards/margins": 0.01832955173811169, "regularization/w2": 0.8446584939956665, "regularization/grad_norm_sq": 3648.669189453125, "kl": 0.2727767527103424, "stability/response_length_mean": 23.625, "stability/response_length_std": 26.338117599487305, "stability/response_length_var": 693.6964111328125, "stability/token_entropy_mean": 4.37109375, "stability/repetition_rate_mean": 0.3225615620613098}
+{"current_steps": 115, "total_steps": 196, "loss": 1.3496, "lr": 9.997740697079592e-07, "epoch": 0.5886116442738324, "percentage": 58.67, "elapsed_time": "0:42:08", "remaining_time": "0:29:41", "rewards/chosen": 0.050295352935791016, "logps/chosen": -364.08957179588606, "logits/chosen": -60607565.7721519, "rewards/rejected": 0.020283417937196332, "logps/rejected": -200.99373070987653, "logits/rejected": -53416308.938271604, "rewards/margins": 0.030011934998594683, "regularization/w2": 0.8016390204429626, "regularization/grad_norm_sq": 3345.042724609375, "kl": 0.31304872035980225, "stability/response_length_mean": 49.25, "stability/response_length_std": 27.316923141479492, "stability/response_length_var": 746.2142944335938, "stability/token_entropy_mean": 4.751953125, "stability/repetition_rate_mean": 0.2233072966337204}
+{"current_steps": 120, "total_steps": 196, "loss": 1.3724, "lr": 9.995838996722914e-07, "epoch": 0.6142034548944337, "percentage": 61.22, "elapsed_time": "0:43:57", "remaining_time": "0:27:50", "rewards/chosen": 0.05619120753668492, "logps/chosen": -383.6830065359477, "logits/chosen": -57740241.150326796, "rewards/rejected": 0.02755490034640192, "logps/rejected": -211.94082148203591, "logits/rejected": -54358218.34730539, "rewards/margins": 0.028636307190282997, "regularization/w2": 0.876289963722229, "regularization/grad_norm_sq": 4084.145263671875, "kl": 0.3582621216773987, "stability/response_length_mean": 27.625, "stability/response_length_std": 30.486238479614258, "stability/response_length_var": 929.4107055664062, "stability/token_entropy_mean": 4.673828125, "stability/repetition_rate_mean": 0.16154514253139496}
+{"current_steps": 125, "total_steps": 196, "loss": 1.3496, "lr": 9.993361385379874e-07, "epoch": 0.6397952655150352, "percentage": 63.78, "elapsed_time": "0:45:48", "remaining_time": "0:26:01", "rewards/chosen": 0.06249639432724208, "logps/chosen": -416.7543343321918, "logits/chosen": -60689534.24657534, "rewards/rejected": 0.026742987249089384, "logps/rejected": -216.09976652298852, "logits/rejected": -55647396.78160919, "rewards/margins": 0.035753407078152694, "regularization/w2": 0.8455937504768372, "regularization/grad_norm_sq": 3718.946044921875, "kl": 0.39068296551704407, "stability/response_length_mean": 15.875, "stability/response_length_std": 24.584766387939453, "stability/response_length_var": 604.4107055664062, "stability/token_entropy_mean": 3.48828125, "stability/repetition_rate_mean": 0.16215278208255768}
+{"current_steps": 130, "total_steps": 196, "loss": 1.3539, "lr": 9.99030814866488e-07, "epoch": 0.6653870761356366, "percentage": 66.33, "elapsed_time": "0:47:37", "remaining_time": "0:24:10", "rewards/chosen": 0.07134679978893649, "logps/chosen": -391.94445564516127, "logits/chosen": -57394050.477419354, "rewards/rejected": 0.027143828074137368, "logps/rejected": -183.64746685606062, "logits/rejected": -53828483.878787875, "rewards/margins": 0.044202971714799116, "regularization/w2": 0.8568418622016907, "regularization/grad_norm_sq": 3812.23291015625, "kl": 0.4606175422668457, "stability/response_length_mean": 43.25, "stability/response_length_std": 29.620214462280273, "stability/response_length_var": 877.3571166992188, "stability/token_entropy_mean": 3.6240234375, "stability/repetition_rate_mean": 0.26218748092651367}
+{"current_steps": 135, "total_steps": 196, "loss": 1.3412, "lr": 9.986679638549358e-07, "epoch": 0.690978886756238, "percentage": 68.88, "elapsed_time": "0:49:27", "remaining_time": "0:22:20", "rewards/chosen": 0.08447715548649887, "logps/chosen": -400.97141008435585, "logits/chosen": -59838715.288343556, "rewards/rejected": 0.036591608812854545, "logps/rejected": -247.50482683121018, "logits/rejected": -54993019.92356688, "rewards/margins": 0.04788554667364432, "regularization/w2": 0.889818012714386, "regularization/grad_norm_sq": 4162.27978515625, "kl": 0.5093870759010315, "stability/response_length_mean": 29.25, "stability/response_length_std": 29.936838150024414, "stability/response_length_var": 896.2142944335938, "stability/token_entropy_mean": 4.080078125, "stability/repetition_rate_mean": 0.2230902761220932}
+{"current_steps": 140, "total_steps": 196, "loss": 1.3743, "lr": 9.982476273321175e-07, "epoch": 0.7165706973768394, "percentage": 71.43, "elapsed_time": "0:51:17", "remaining_time": "0:20:30", "rewards/chosen": 0.08064722003358783, "logps/chosen": -366.3832623106061, "logits/chosen": -57087714.52121212, "rewards/rejected": 0.012802070186984154, "logps/rejected": -189.90645161290323, "logits/rejected": -50078577.96129032, "rewards/margins": 0.06784514984660368, "regularization/w2": 0.9738343358039856, "regularization/grad_norm_sq": 5975.46728515625, "kl": 0.4901605546474457, "stability/response_length_mean": 56.25, "stability/response_length_std": 21.920310974121094, "stability/response_length_var": 480.5, "stability/token_entropy_mean": 4.9853515625, "stability/repetition_rate_mean": 0.3125}
+{"current_steps": 145, "total_steps": 196, "loss": 1.3404, "lr": 9.977698537536417e-07, "epoch": 0.7421625079974408, "percentage": 73.98, "elapsed_time": "0:53:06", "remaining_time": "0:18:40", "rewards/chosen": 0.07872245972415051, "logps/chosen": -329.48458678463857, "logits/chosen": -51145351.71084338, "rewards/rejected": 0.03411294577957748, "logps/rejected": -199.14678875811688, "logits/rejected": -51668925.50649351, "rewards/margins": 0.044609513944573034, "regularization/w2": 0.8221335411071777, "regularization/grad_norm_sq": 3477.833251953125, "kl": 0.5782578587532043, "stability/response_length_mean": 35.125, "stability/response_length_std": 31.37076759338379, "stability/response_length_var": 984.125, "stability/token_entropy_mean": 5.19921875, "stability/repetition_rate_mean": 0.27182701230049133}
+{"current_steps": 150, "total_steps": 196, "loss": 1.3548, "lr": 9.972346981963546e-07, "epoch": 0.7677543186180422, "percentage": 76.53, "elapsed_time": "0:54:56", "remaining_time": "0:16:50", "rewards/chosen": 0.1129976609173943, "logps/chosen": -396.80445772058823, "logits/chosen": -55416663.34117647, "rewards/rejected": 0.037230908075968426, "logps/rejected": -207.3536328125, "logits/rejected": -50182574.08, "rewards/margins": 0.07576675284142587, "regularization/w2": 0.900852382183075, "regularization/grad_norm_sq": 4135.36962890625, "kl": 0.6353618502616882, "stability/response_length_mean": 29.375, "stability/response_length_std": 29.193626403808594, "stability/response_length_var": 852.2678833007812, "stability/token_entropy_mean": 4.65625, "stability/repetition_rate_mean": 0.20136336982250214}
+{"current_steps": 155, "total_steps": 196, "loss": 1.3426, "lr": 9.966422223519885e-07, "epoch": 0.7933461292386437, "percentage": 79.08, "elapsed_time": "0:56:46", "remaining_time": "0:15:01", "rewards/chosen": 0.0877797295065487, "logps/chosen": -360.11148897058825, "logits/chosen": -54021403.105882354, "rewards/rejected": 0.027975524266560872, "logps/rejected": -191.02584635416667, "logits/rejected": -49364510.72, "rewards/margins": 0.059804205239987834, "regularization/w2": 0.9337390065193176, "regularization/grad_norm_sq": 4965.29248046875, "kl": 0.6446139216423035, "stability/response_length_mean": 56.5, "stability/response_length_std": 10.928338050842285, "stability/response_length_var": 119.42857360839844, "stability/token_entropy_mean": 4.533203125, "stability/repetition_rate_mean": 0.3111416697502136}
+{"current_steps": 160, "total_steps": 196, "loss": 1.357, "lr": 9.959924945200523e-07, "epoch": 0.818937939859245, "percentage": 81.63, "elapsed_time": "0:58:37", "remaining_time": "0:13:11", "rewards/chosen": 0.13393676848638625, "logps/chosen": -415.5559895833333, "logits/chosen": -55247811.047619045, "rewards/rejected": 0.04094262813266955, "logps/rejected": -186.48792146381578, "logits/rejected": -48958012.631578945, "rewards/margins": 0.0929941403537167, "regularization/w2": 0.8730607032775879, "regularization/grad_norm_sq": 3970.04931640625, "kl": 0.7235702872276306, "stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.48828125, "stability/repetition_rate_mean": 0.30851560831069946}
+{"current_steps": 165, "total_steps": 196, "loss": 1.3489, "lr": 9.952855895999568e-07, "epoch": 0.8445297504798465, "percentage": 84.18, "elapsed_time": "1:00:26", "remaining_time": "0:11:21", "rewards/chosen": 0.11605137792126886, "logps/chosen": -369.7562185704023, "logits/chosen": -54424352.36781609, "rewards/rejected": 0.046767483018849, "logps/rejected": -212.1244381421233, "logits/rejected": -49791210.95890411, "rewards/margins": 0.06928389490241987, "regularization/w2": 0.8307213187217712, "regularization/grad_norm_sq": 3547.708984375, "kl": 0.7536167502403259, "stability/response_length_mean": 22.5, "stability/response_length_std": 28.33977508544922, "stability/response_length_var": 803.1428833007812, "stability/token_entropy_mean": 4.88671875, "stability/repetition_rate_mean": 0.2411106526851654}
+{"current_steps": 170, "total_steps": 196, "loss": 1.3222, "lr": 9.9452158908238e-07, "epoch": 0.8701215611004478, "percentage": 86.73, "elapsed_time": "1:02:17", "remaining_time": "0:09:31", "rewards/chosen": 0.16640804223054964, "logps/chosen": -439.6777921597633, "logits/chosen": -59139877.86982249, "rewards/rejected": 0.06497004174238799, "logps/rejected": -190.6081332781457, "logits/rejected": -52581216.63576159, "rewards/margins": 0.10143800048816165, "regularization/w2": 0.8687969446182251, "regularization/grad_norm_sq": 4127.0400390625, "kl": 0.920440137386322, "stability/response_length_mean": 36.25, "stability/response_length_std": 30.76988410949707, "stability/response_length_var": 946.7857055664062, "stability/token_entropy_mean": 3.67041015625, "stability/repetition_rate_mean": 0.23209978640079498}
+{"current_steps": 175, "total_steps": 196, "loss": 1.3334, "lr": 9.937005810398745e-07, "epoch": 0.8957133717210493, "percentage": 89.29, "elapsed_time": "1:04:07", "remaining_time": "0:07:41", "rewards/chosen": 0.15770882296274943, "logps/chosen": -354.8600809487952, "logits/chosen": -56657488.192771085, "rewards/rejected": 0.047987513727956004, "logps/rejected": -194.32765574269482, "logits/rejected": -48037266.28571428, "rewards/margins": 0.10972130923479342, "regularization/w2": 0.8748798370361328, "regularization/grad_norm_sq": 4247.8359375, "kl": 0.9491209983825684, "stability/response_length_mean": 29.75, "stability/response_length_std": 26.015106201171875, "stability/response_length_var": 676.7857055664062, "stability/token_entropy_mean": 4.822265625, "stability/repetition_rate_mean": 0.18672296404838562}
+{"current_steps": 180, "total_steps": 196, "loss": 1.321, "lr": 9.928226601167138e-07, "epoch": 0.9213051823416507, "percentage": 91.84, "elapsed_time": "1:05:57", "remaining_time": "0:05:51", "rewards/chosen": 0.17141536408406824, "logps/chosen": -374.4316765720859, "logits/chosen": -52647628.17177914, "rewards/rejected": 0.0759030481812301, "logps/rejected": -210.53918690286625, "logits/rejected": -47983459.46496815, "rewards/margins": 0.09551231590283814, "regularization/w2": 0.8262709975242615, "regularization/grad_norm_sq": 3608.227294921875, "kl": 1.032440423965454, "stability/response_length_mean": 35.25, "stability/response_length_std": 31.129911422729492, "stability/response_length_var": 969.0714111328125, "stability/token_entropy_mean": 3.130859375, "stability/repetition_rate_mean": 0.2868303656578064}
+{"current_steps": 185, "total_steps": 196, "loss": 1.3781, "lr": 9.918879275179817e-07, "epoch": 0.946896992962252, "percentage": 94.39, "elapsed_time": "1:07:46", "remaining_time": "0:04:01", "rewards/chosen": 0.16873404420452354, "logps/chosen": -310.31404320987656, "logits/chosen": -54614894.61728395, "rewards/rejected": 0.05016195623180534, "logps/rejected": -196.53752966772151, "logits/rejected": -49520795.5443038, "rewards/margins": 0.1185720879727182, "regularization/w2": 0.9516963958740234, "regularization/grad_norm_sq": 7128.36279296875, "kl": 1.0751243829727173, "stability/response_length_mean": 32.875, "stability/response_length_std": 33.30996322631836, "stability/response_length_var": 1109.5535888671875, "stability/token_entropy_mean": 3.71875, "stability/repetition_rate_mean": 0.2109375}
+{"current_steps": 190, "total_steps": 196, "loss": 1.3626, "lr": 9.90896490997906e-07, "epoch": 0.9724888035828535, "percentage": 96.94, "elapsed_time": "1:09:35", "remaining_time": "0:02:11", "rewards/chosen": 0.20803529202581167, "logps/chosen": -390.324967252994, "logits/chosen": -53095589.556886226, "rewards/rejected": 0.07497862896888084, "logps/rejected": -210.44881025326796, "logits/rejected": -49707021.385620914, "rewards/margins": 0.1330566630569308, "regularization/w2": 0.9177618026733398, "regularization/grad_norm_sq": 5476.40771484375, "kl": 1.1208827495574951, "stability/response_length_mean": 20.375, "stability/response_length_std": 27.645395278930664, "stability/response_length_var": 764.2678833007812, "stability/token_entropy_mean": 4.3505859375, "stability/repetition_rate_mean": 0.2964409589767456}
+{"current_steps": 195, "total_steps": 196, "loss": 1.3508, "lr": 9.898484648474362e-07, "epoch": 0.9980806142034548, "percentage": 99.49, "elapsed_time": "1:11:24", "remaining_time": "0:00:21", "rewards/chosen": 0.19076536855389994, "logps/chosen": -390.8772681451613, "logits/chosen": -60050742.5032258, "rewards/rejected": 0.08325019027247574, "logps/rejected": -197.0868252840909, "logits/rejected": -53805540.07272727, "rewards/margins": 0.1075151782814242, "regularization/w2": 0.8582512736320496, "regularization/grad_norm_sq": 4039.490966796875, "kl": 1.1453860998153687, "stability/response_length_mean": 49.0, "stability/response_length_std": 27.82085609436035, "stability/response_length_var": 774.0, "stability/token_entropy_mean": 5.19921875, "stability/repetition_rate_mean": 0.3895089328289032}
+{"current_steps": 196, "total_steps": 196, "epoch": 1.0, "percentage": 100.0, "elapsed_time": "1:11:39", "remaining_time": "0:00:00"}

mistral7b_kto_w2_delta0.01/trainer_state.json

@@ -0,0 +1,974 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 196,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.025591810620601407,
+      "grad_norm": 0.5667362213134766,
+      "kl": 0.008433527313172817,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59717241.2631579,
+      "logits/rejected": -50415911.61904762,
+      "logps/chosen": -439.02847450657896,
+      "logps/rejected": -198.40762183779762,
+      "loss": 1.3863,
+      "loss/base_kto": 0.4998113214969635,
+      "metrics/advantage_var": 0.04507648944854736,
+      "regularization/grad_norm_sq": 4328.5595703125,
+      "regularization/w2": 0.9065223932266235,
+      "rewards/chosen": 0.0017119320599656355,
+      "rewards/margins": 0.00153388447107229,
+      "rewards/rejected": 0.00017804758889334544,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.6054168343544006,
+      "kl": 0.019305892288684845,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59293257.14285714,
+      "logits/rejected": -50017699.469879515,
+      "logps/chosen": -429.92294034090907,
+      "logps/rejected": -192.84436182228916,
+      "loss": 1.3645,
+      "loss/base_kto": 0.50011146068573,
+      "metrics/advantage_var": 0.06692608445882797,
+      "regularization/grad_norm_sq": 3843.479736328125,
+      "regularization/w2": 0.8646374940872192,
+      "rewards/chosen": 0.0015573305162516508,
+      "rewards/margins": -0.0008008119947025051,
+      "rewards/rejected": 0.002358142510954156,
+      "stability/repetition_rate_mean": 0.3629409670829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.27001953125,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.6099650859832764,
+      "kl": 0.013482200913131237,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56229807.32121212,
+      "logits/rejected": -51916106.32258064,
+      "logps/chosen": -404.1315814393939,
+      "logps/rejected": -200.67066532258065,
+      "loss": 1.3548,
+      "loss/base_kto": 0.5002730488777161,
+      "metrics/advantage_var": 0.053939878940582275,
+      "regularization/grad_norm_sq": 4172.4462890625,
+      "regularization/w2": 0.8843773007392883,
+      "rewards/chosen": -7.632036100734364e-05,
+      "rewards/margins": -0.0024158480448806742,
+      "rewards/rejected": 0.0023395276838733305,
+      "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": 0.619081437587738,
+      "kl": 0.018938787281513214,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51664345.6,
+      "logits/rejected": -48065900.8,
+      "logps/chosen": -343.694580078125,
+      "logps/rejected": -225.647802734375,
+      "loss": 1.3654,
+      "loss/base_kto": 0.5002180337905884,
+      "metrics/advantage_var": 0.05758875235915184,
+      "regularization/grad_norm_sq": 3665.234375,
+      "regularization/w2": 0.843424916267395,
+      "rewards/chosen": 0.00021410472691059114,
+      "rewards/margins": -0.0013427346944808961,
+      "rewards/rejected": 0.0015568394213914872,
+      "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": 4.4140625,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.5774287581443787,
+      "kl": 0.02769310213625431,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56488998.84137931,
+      "logits/rejected": -52022798.62857143,
+      "logps/chosen": -378.4621767241379,
+      "logps/rejected": -202.77145089285713,
+      "loss": 1.3771,
+      "loss/base_kto": 0.4999386966228485,
+      "metrics/advantage_var": 0.053915154188871384,
+      "regularization/grad_norm_sq": 4001.77783203125,
+      "regularization/w2": 0.860791027545929,
+      "rewards/chosen": 0.0029272856383488095,
+      "rewards/margins": 0.0007881318172210544,
+      "rewards/rejected": 0.002139153821127755,
+      "stability/repetition_rate_mean": 0.17215879261493683,
+      "stability/response_length_mean": 43.375,
+      "stability/response_length_std": 28.54538917541504,
+      "stability/response_length_var": 814.8392944335938,
+      "stability/token_entropy_mean": 4.8125,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.603484034538269,
+      "kl": 0.018374785780906677,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57729649.39597315,
+      "logits/rejected": -51505858.61988304,
+      "logps/chosen": -468.588716442953,
+      "logps/rejected": -198.24766995614036,
+      "loss": 1.3694,
+      "loss/base_kto": 0.5000385642051697,
+      "metrics/advantage_var": 0.07535337656736374,
+      "regularization/grad_norm_sq": 4544.02490234375,
+      "regularization/w2": 0.917149007320404,
+      "rewards/chosen": -0.0007460268351855694,
+      "rewards/margins": -0.0007687252250420083,
+      "rewards/rejected": 2.2698389856438888e-05,
+      "stability/repetition_rate_mean": 0.1427641361951828,
+      "stability/response_length_mean": 43.0,
+      "stability/response_length_std": 29.602123260498047,
+      "stability/response_length_var": 876.2857055664062,
+      "stability/token_entropy_mean": 4.61328125,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6760314106941223,
+      "kl": 0.02894224040210247,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57635956.658227846,
+      "logits/rejected": -53888758.518518515,
+      "logps/chosen": -385.90286293512656,
+      "logps/rejected": -195.11945408950618,
+      "loss": 1.3758,
+      "loss/base_kto": 0.49967771768569946,
+      "metrics/advantage_var": 0.08824540674686432,
+      "regularization/grad_norm_sq": 3807.059326171875,
+      "regularization/w2": 0.8619928359985352,
+      "rewards/chosen": 0.003366763833202893,
+      "rewards/margins": 0.002664287083385363,
+      "rewards/rejected": 0.0007024767498175303,
+      "stability/repetition_rate_mean": 0.3166852593421936,
+      "stability/response_length_mean": 56.875,
+      "stability/response_length_std": 20.152544021606445,
+      "stability/response_length_var": 406.125,
+      "stability/token_entropy_mean": 5.419921875,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.5855346322059631,
+      "kl": 0.02706434763967991,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56634796.8,
+      "logits/rejected": -50219603.2,
+      "logps/chosen": -421.87265625,
+      "logps/rejected": -219.273095703125,
+      "loss": 1.3686,
+      "loss/base_kto": 0.5001084208488464,
+      "metrics/advantage_var": 0.08759542554616928,
+      "regularization/grad_norm_sq": 4305.833984375,
+      "regularization/w2": 0.9105228781700134,
+      "rewards/chosen": 0.0030031204223632812,
+      "rewards/margins": -0.0006122220307588576,
+      "rewards/rejected": 0.003615342453122139,
+      "stability/repetition_rate_mean": 0.13671875,
+      "stability/response_length_mean": 33.125,
+      "stability/response_length_std": 33.01704025268555,
+      "stability/response_length_var": 1090.125,
+      "stability/token_entropy_mean": 4.8935546875,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.6397272944450378,
+      "kl": 0.03937375545501709,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52699170.36241611,
+      "logits/rejected": -50773297.403508775,
+      "logps/chosen": -352.2831900167785,
+      "logps/rejected": -243.85115131578948,
+      "loss": 1.379,
+      "loss/base_kto": 0.4997560679912567,
+      "metrics/advantage_var": 0.07129540294408798,
+      "regularization/grad_norm_sq": 3761.641357421875,
+      "regularization/w2": 0.8400061726570129,
+      "rewards/chosen": 0.004526910365827932,
+      "rewards/margins": 0.0022242151040232504,
+      "rewards/rejected": 0.0023026952618046812,
+      "stability/repetition_rate_mean": 0.0994017943739891,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.87324333190918,
+      "stability/response_length_var": 892.4107055664062,
+      "stability/token_entropy_mean": 4.7578125,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.5777538418769836,
+      "kl": 0.047937434166669846,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54044414.281879194,
+      "logits/rejected": -48299597.847953215,
+      "logps/chosen": -375.1258389261745,
+      "logps/rejected": -180.18202896564327,
+      "loss": 1.3394,
+      "loss/base_kto": 0.49980711936950684,
+      "metrics/advantage_var": 0.05609378218650818,
+      "regularization/grad_norm_sq": 3838.28369140625,
+      "regularization/w2": 0.8510723114013672,
+      "rewards/chosen": 0.007330068805873794,
+      "rewards/margins": 0.0014725509611215326,
+      "rewards/rejected": 0.005857517844752262,
+      "stability/repetition_rate_mean": 0.2952169179916382,
+      "stability/response_length_mean": 53.375,
+      "stability/response_length_std": 21.480472564697266,
+      "stability/response_length_var": 461.41070556640625,
+      "stability/token_entropy_mean": 4.1162109375,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.6621295213699341,
+      "kl": 0.0575476698577404,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56949887.204968944,
+      "logits/rejected": -49557458.91823899,
+      "logps/chosen": -344.48966420807454,
+      "logps/rejected": -230.39740566037736,
+      "loss": 1.357,
+      "loss/base_kto": 0.5003090500831604,
+      "metrics/advantage_var": 0.06319057196378708,
+      "regularization/grad_norm_sq": 3955.544921875,
+      "regularization/w2": 0.8724905252456665,
+      "rewards/chosen": 0.0022328466362094286,
+      "rewards/margins": -0.002831306810132032,
+      "rewards/rejected": 0.00506415344634146,
+      "stability/repetition_rate_mean": 0.11589926481246948,
+      "stability/response_length_mean": 22.375,
+      "stability/response_length_std": 30.236862182617188,
+      "stability/response_length_var": 914.2678833007812,
+      "stability/token_entropy_mean": 4.302734375,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.5823009014129639,
+      "kl": 0.05734771117568016,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52677849.21212121,
+      "logits/rejected": -47824090.01290323,
+      "logps/chosen": -359.6180871212121,
+      "logps/rejected": -178.15743447580644,
+      "loss": 1.3664,
+      "loss/base_kto": 0.4995133578777313,
+      "metrics/advantage_var": 0.057131972163915634,
+      "regularization/grad_norm_sq": 3162.57421875,
+      "regularization/w2": 0.7870045900344849,
+      "rewards/chosen": 0.010271583903919567,
+      "rewards/margins": 0.003323828090320934,
+      "rewards/rejected": 0.006947755813598633,
+      "stability/repetition_rate_mean": 0.17885693907737732,
+      "stability/response_length_mean": 44.625,
+      "stability/response_length_std": 27.8820858001709,
+      "stability/response_length_var": 777.4107055664062,
+      "stability/token_entropy_mean": 4.607421875,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.5745700597763062,
+      "kl": 0.07911502569913864,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57565778.58064516,
+      "logits/rejected": -47265906.81212121,
+      "logps/chosen": -381.8476814516129,
+      "logps/rejected": -183.48361742424242,
+      "loss": 1.3773,
+      "loss/base_kto": 0.49997368454933167,
+      "metrics/advantage_var": 0.06365744024515152,
+      "regularization/grad_norm_sq": 3887.826904296875,
+      "regularization/w2": 0.8648955225944519,
+      "rewards/chosen": 0.009909362177694997,
+      "rewards/margins": 3.680861240957546e-05,
+      "rewards/rejected": 0.009872553565285422,
+      "stability/repetition_rate_mean": 0.1857219785451889,
+      "stability/response_length_mean": 20.625,
+      "stability/response_length_std": 27.176342010498047,
+      "stability/response_length_var": 738.5535888671875,
+      "stability/token_entropy_mean": 5.525390625,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5528952479362488,
+      "kl": 0.09082157909870148,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52773428.8516129,
+      "logits/rejected": -49024763.345454544,
+      "logps/chosen": -379.00569556451615,
+      "logps/rejected": -220.34614109848485,
+      "loss": 1.403,
+      "loss/base_kto": 0.49938470125198364,
+      "metrics/advantage_var": 0.07325268536806107,
+      "regularization/grad_norm_sq": 4501.79833984375,
+      "regularization/w2": 0.9129446148872375,
+      "rewards/chosen": 0.01227622724348499,
+      "rewards/margins": 0.004672638272260291,
+      "rewards/rejected": 0.007603588971224698,
+      "stability/repetition_rate_mean": 0.162109375,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.77734375,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.609993040561676,
+      "kl": 0.08937091380357742,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49499239.783783786,
+      "logits/rejected": -47880882.60465116,
+      "logps/chosen": -314.12774493243245,
+      "logps/rejected": -228.82871547965115,
+      "loss": 1.3809,
+      "loss/base_kto": 0.4988512098789215,
+      "metrics/advantage_var": 0.08098717033863068,
+      "regularization/grad_norm_sq": 3738.951171875,
+      "regularization/w2": 0.8484571576118469,
+      "rewards/chosen": 0.014305324167818637,
+      "rewards/margins": 0.008692953588076915,
+      "rewards/rejected": 0.005612370579741722,
+      "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": 2.15625,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.603708028793335,
+      "kl": 0.11913253366947174,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58623972.87417219,
+      "logits/rejected": -53456156.78106509,
+      "logps/chosen": -403.85572744205297,
+      "logps/rejected": -200.97545303254438,
+      "loss": 1.3598,
+      "loss/base_kto": 0.49891233444213867,
+      "metrics/advantage_var": 0.07670630514621735,
+      "regularization/grad_norm_sq": 3606.688232421875,
+      "regularization/w2": 0.8389098048210144,
+      "rewards/chosen": 0.01743706330558322,
+      "rewards/margins": 0.009403561979603481,
+      "rewards/rejected": 0.00803350132597974,
+      "stability/repetition_rate_mean": 0.19088542461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.2265625,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.6118994951248169,
+      "kl": 0.12905550003051758,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51481953.44516129,
+      "logits/rejected": -47134605.18787879,
+      "logps/chosen": -358.1473790322581,
+      "logps/rejected": -230.32398200757575,
+      "loss": 1.3679,
+      "loss/base_kto": 0.49895092844963074,
+      "metrics/advantage_var": 0.09141772240400314,
+      "regularization/grad_norm_sq": 3847.29296875,
+      "regularization/w2": 0.8647522330284119,
+      "rewards/chosen": 0.018512418193201863,
+      "rewards/margins": 0.008430653047235363,
+      "rewards/rejected": 0.0100817651459665,
+      "stability/repetition_rate_mean": 0.1218377947807312,
+      "stability/response_length_mean": 32.375,
+      "stability/response_length_std": 33.547142028808594,
+      "stability/response_length_var": 1125.4107666015625,
+      "stability/token_entropy_mean": 4.869140625,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.638875424861908,
+      "kl": 0.16841307282447815,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58414395.07692308,
+      "logits/rejected": -50576390.24390244,
+      "logps/chosen": -380.533203125,
+      "logps/rejected": -187.13657583841464,
+      "loss": 1.3594,
+      "loss/base_kto": 0.49779611825942993,
+      "metrics/advantage_var": 0.11126144975423813,
+      "regularization/grad_norm_sq": 3978.860595703125,
+      "regularization/w2": 0.8712137341499329,
+      "rewards/chosen": 0.025618803806793995,
+      "rewards/margins": 0.016595058194393664,
+      "rewards/rejected": 0.009023745612400333,
+      "stability/repetition_rate_mean": 0.2843276560306549,
+      "stability/response_length_mean": 26.875,
+      "stability/response_length_std": 27.544445037841797,
+      "stability/response_length_var": 758.6964111328125,
+      "stability/token_entropy_mean": 4.3203125,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.6166263222694397,
+      "kl": 0.1684073954820633,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50395103.594936706,
+      "logits/rejected": -48990799.01234568,
+      "logps/chosen": -314.7726463607595,
+      "logps/rejected": -207.06107735339506,
+      "loss": 1.3568,
+      "loss/base_kto": 0.4983139932155609,
+      "metrics/advantage_var": 0.11646153777837753,
+      "regularization/grad_norm_sq": 3889.73095703125,
+      "regularization/w2": 0.8536549806594849,
+      "rewards/chosen": 0.022073296051991136,
+      "rewards/margins": 0.013807725693997339,
+      "rewards/rejected": 0.008265570357993798,
+      "stability/repetition_rate_mean": 0.19374233484268188,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 31.1264705657959,
+      "stability/response_length_var": 968.8571166992188,
+      "stability/token_entropy_mean": 3.724609375,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.6097723841667175,
+      "kl": 0.19923549890518188,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54706169.08108108,
+      "logits/rejected": -50907100.27906977,
+      "logps/chosen": -359.5902396537162,
+      "logps/rejected": -216.37990552325581,
+      "loss": 1.3626,
+      "loss/base_kto": 0.4989898204803467,
+      "metrics/advantage_var": 0.1130441352725029,
+      "regularization/grad_norm_sq": 3821.28564453125,
+      "regularization/w2": 0.8549612164497375,
+      "rewards/chosen": 0.027993318196889515,
+      "rewards/margins": 0.008093704561404954,
+      "rewards/rejected": 0.01989961363548456,
+      "stability/repetition_rate_mean": 0.23982319235801697,
+      "stability/response_length_mean": 38.0,
+      "stability/response_length_std": 30.10457992553711,
+      "stability/response_length_var": 906.2857055664062,
+      "stability/token_entropy_mean": 4.146484375,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6825862526893616,
+      "kl": 0.21407940983772278,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59382716.144578315,
+      "logits/rejected": -53816409.766233765,
+      "logps/chosen": -399.5208960843373,
+      "logps/rejected": -209.84342025162337,
+      "loss": 1.3765,
+      "loss/base_kto": 0.4972083270549774,
+      "metrics/advantage_var": 0.20788688957691193,
+      "regularization/grad_norm_sq": 4485.09521484375,
+      "regularization/w2": 0.9070795178413391,
+      "rewards/chosen": 0.03815392413771296,
+      "rewards/margins": 0.022114001497791844,
+      "rewards/rejected": 0.016039922639921114,
+      "stability/repetition_rate_mean": 0.1741071343421936,
+      "stability/response_length_mean": 28.0,
+      "stability/response_length_std": 28.665559768676758,
+      "stability/response_length_var": 821.7142944335938,
+      "stability/token_entropy_mean": 3.37109375,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.6473393440246582,
+      "kl": 0.2727767527103424,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57382239.61445783,
+      "logits/rejected": -53606935.27272727,
+      "logps/chosen": -378.179499246988,
+      "logps/rejected": -205.8876572646104,
+      "loss": 1.3516,
+      "loss/base_kto": 0.49749407172203064,
+      "metrics/advantage_var": 0.22285981476306915,
+      "regularization/grad_norm_sq": 3648.669189453125,
+      "regularization/w2": 0.8446584939956665,
+      "rewards/chosen": 0.03837782790861934,
+      "rewards/margins": 0.01832955173811169,
+      "rewards/rejected": 0.020048276170507653,
+      "stability/repetition_rate_mean": 0.3225615620613098,
+      "stability/response_length_mean": 23.625,
+      "stability/response_length_std": 26.338117599487305,
+      "stability/response_length_var": 693.6964111328125,
+      "stability/token_entropy_mean": 4.37109375,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.6492971181869507,
+      "kl": 0.31304872035980225,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60607565.7721519,
+      "logits/rejected": -53416308.938271604,
+      "logps/chosen": -364.08957179588606,
+      "logps/rejected": -200.99373070987653,
+      "loss": 1.3496,
+      "loss/base_kto": 0.4962143898010254,
+      "metrics/advantage_var": 0.24412627518177032,
+      "regularization/grad_norm_sq": 3345.042724609375,
+      "regularization/w2": 0.8016390204429626,
+      "rewards/chosen": 0.050295352935791016,
+      "rewards/margins": 0.030011934998594683,
+      "rewards/rejected": 0.020283417937196332,
+      "stability/repetition_rate_mean": 0.2233072966337204,
+      "stability/response_length_mean": 49.25,
+      "stability/response_length_std": 27.316923141479492,
+      "stability/response_length_var": 746.2142944335938,
+      "stability/token_entropy_mean": 4.751953125,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.6205304861068726,
+      "kl": 0.3582621216773987,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57740241.150326796,
+      "logits/rejected": -54358218.34730539,
+      "logps/chosen": -383.6830065359477,
+      "logps/rejected": -211.94082148203591,
+      "loss": 1.3724,
+      "loss/base_kto": 0.49640893936157227,
+      "metrics/advantage_var": 0.4239996075630188,
+      "regularization/grad_norm_sq": 4084.145263671875,
+      "regularization/w2": 0.876289963722229,
+      "rewards/chosen": 0.05619120753668492,
+      "rewards/margins": 0.028636307190282997,
+      "rewards/rejected": 0.02755490034640192,
+      "stability/repetition_rate_mean": 0.16154514253139496,
+      "stability/response_length_mean": 27.625,
+      "stability/response_length_std": 30.486238479614258,
+      "stability/response_length_var": 929.4107055664062,
+      "stability/token_entropy_mean": 4.673828125,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.6123977899551392,
+      "kl": 0.39068296551704407,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60689534.24657534,
+      "logits/rejected": -55647396.78160919,
+      "logps/chosen": -416.7543343321918,
+      "logps/rejected": -216.09976652298852,
+      "loss": 1.3496,
+      "loss/base_kto": 0.4957490861415863,
+      "metrics/advantage_var": 0.5870842933654785,
+      "regularization/grad_norm_sq": 3718.946044921875,
+      "regularization/w2": 0.8455937504768372,
+      "rewards/chosen": 0.06249639432724208,
+      "rewards/margins": 0.035753407078152694,
+      "rewards/rejected": 0.026742987249089384,
+      "stability/repetition_rate_mean": 0.16215278208255768,
+      "stability/response_length_mean": 15.875,
+      "stability/response_length_std": 24.584766387939453,
+      "stability/response_length_var": 604.4107055664062,
+      "stability/token_entropy_mean": 3.48828125,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.592140793800354,
+      "kl": 0.4606175422668457,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57394050.477419354,
+      "logits/rejected": -53828483.878787875,
+      "logps/chosen": -391.94445564516127,
+      "logps/rejected": -183.64746685606062,
+      "loss": 1.3539,
+      "loss/base_kto": 0.4944417178630829,
+      "metrics/advantage_var": 0.848583996295929,
+      "regularization/grad_norm_sq": 3812.23291015625,
+      "regularization/w2": 0.8568418622016907,
+      "rewards/chosen": 0.07134679978893649,
+      "rewards/margins": 0.044202971714799116,
+      "rewards/rejected": 0.027143828074137368,
+      "stability/repetition_rate_mean": 0.26218748092651367,
+      "stability/response_length_mean": 43.25,
+      "stability/response_length_std": 29.620214462280273,
+      "stability/response_length_var": 877.3571166992188,
+      "stability/token_entropy_mean": 3.6240234375,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5895723700523376,
+      "kl": 0.5093870759010315,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59838715.288343556,
+      "logits/rejected": -54993019.92356688,
+      "logps/chosen": -400.97141008435585,
+      "logps/rejected": -247.50482683121018,
+      "loss": 1.3412,
+      "loss/base_kto": 0.4939149022102356,
+      "metrics/advantage_var": 1.0497407913208008,
+      "regularization/grad_norm_sq": 4162.27978515625,
+      "regularization/w2": 0.889818012714386,
+      "rewards/chosen": 0.08447715548649887,
+      "rewards/margins": 0.04788554667364432,
+      "rewards/rejected": 0.036591608812854545,
+      "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": 4.080078125,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.7277541160583496,
+      "kl": 0.4901605546474457,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57087714.52121212,
+      "logits/rejected": -50078577.96129032,
+      "logps/chosen": -366.3832623106061,
+      "logps/rejected": -189.90645161290323,
+      "loss": 1.3743,
+      "loss/base_kto": 0.4920271039009094,
+      "metrics/advantage_var": 1.8927977085113525,
+      "regularization/grad_norm_sq": 5975.46728515625,
+      "regularization/w2": 0.9738343358039856,
+      "rewards/chosen": 0.08064722003358783,
+      "rewards/margins": 0.06784514984660368,
+      "rewards/rejected": 0.012802070186984154,
+      "stability/repetition_rate_mean": 0.3125,
+      "stability/response_length_mean": 56.25,
+      "stability/response_length_std": 21.920310974121094,
+      "stability/response_length_var": 480.5,
+      "stability/token_entropy_mean": 4.9853515625,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.6773489713668823,
+      "kl": 0.5782578587532043,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51145351.71084338,
+      "logits/rejected": -51668925.50649351,
+      "logps/chosen": -329.48458678463857,
+      "logps/rejected": -199.14678875811688,
+      "loss": 1.3404,
+      "loss/base_kto": 0.49439021944999695,
+      "metrics/advantage_var": 0.7935466766357422,
+      "regularization/grad_norm_sq": 3477.833251953125,
+      "regularization/w2": 0.8221335411071777,
+      "rewards/chosen": 0.07872245972415051,
+      "rewards/margins": 0.044609513944573034,
+      "rewards/rejected": 0.03411294577957748,
+      "stability/repetition_rate_mean": 0.27182701230049133,
+      "stability/response_length_mean": 35.125,
+      "stability/response_length_std": 31.37076759338379,
+      "stability/response_length_var": 984.125,
+      "stability/token_entropy_mean": 5.19921875,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.6416272521018982,
+      "kl": 0.6353618502616882,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55416663.34117647,
+      "logits/rejected": -50182574.08,
+      "logps/chosen": -396.80445772058823,
+      "logps/rejected": -207.3536328125,
+      "loss": 1.3548,
+      "loss/base_kto": 0.490011066198349,
+      "metrics/advantage_var": 1.2986860275268555,
+      "regularization/grad_norm_sq": 4135.36962890625,
+      "regularization/w2": 0.900852382183075,
+      "rewards/chosen": 0.1129976609173943,
+      "rewards/margins": 0.07576675284142587,
+      "rewards/rejected": 0.037230908075968426,
+      "stability/repetition_rate_mean": 0.20136336982250214,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.193626403808594,
+      "stability/response_length_var": 852.2678833007812,
+      "stability/token_entropy_mean": 4.65625,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.6257898211479187,
+      "kl": 0.6446139216423035,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54021403.105882354,
+      "logits/rejected": -49364510.72,
+      "logps/chosen": -360.11148897058825,
+      "logps/rejected": -191.02584635416667,
+      "loss": 1.3426,
+      "loss/base_kto": 0.49283477663993835,
+      "metrics/advantage_var": 2.1838507652282715,
+      "regularization/grad_norm_sq": 4965.29248046875,
+      "regularization/w2": 0.9337390065193176,
+      "rewards/chosen": 0.0877797295065487,
+      "rewards/margins": 0.059804205239987834,
+      "rewards/rejected": 0.027975524266560872,
+      "stability/repetition_rate_mean": 0.3111416697502136,
+      "stability/response_length_mean": 56.5,
+      "stability/response_length_std": 10.928338050842285,
+      "stability/response_length_var": 119.42857360839844,
+      "stability/token_entropy_mean": 4.533203125,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.6220570206642151,
+      "kl": 0.7235702872276306,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55247811.047619045,
+      "logits/rejected": -48958012.631578945,
+      "logps/chosen": -415.5559895833333,
+      "logps/rejected": -186.48792146381578,
+      "loss": 1.357,
+      "loss/base_kto": 0.48817262053489685,
+      "metrics/advantage_var": 1.9768089056015015,
+      "regularization/grad_norm_sq": 3970.04931640625,
+      "regularization/w2": 0.8730607032775879,
+      "rewards/chosen": 0.13393676848638625,
+      "rewards/margins": 0.0929941403537167,
+      "rewards/rejected": 0.04094262813266955,
+      "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": 4.48828125,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.7207323908805847,
+      "kl": 0.7536167502403259,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54424352.36781609,
+      "logits/rejected": -49791210.95890411,
+      "logps/chosen": -369.7562185704023,
+      "logps/rejected": -212.1244381421233,
+      "loss": 1.3489,
+      "loss/base_kto": 0.4913412034511566,
+      "metrics/advantage_var": 1.7532039880752563,
+      "regularization/grad_norm_sq": 3547.708984375,
+      "regularization/w2": 0.8307213187217712,
+      "rewards/chosen": 0.11605137792126886,
+      "rewards/margins": 0.06928389490241987,
+      "rewards/rejected": 0.046767483018849,
+      "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": 4.88671875,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.6222322583198547,
+      "kl": 0.920440137386322,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59139877.86982249,
+      "logits/rejected": -52581216.63576159,
+      "logps/chosen": -439.6777921597633,
+      "logps/rejected": -190.6081332781457,
+      "loss": 1.3222,
+      "loss/base_kto": 0.48742732405662537,
+      "metrics/advantage_var": 2.3735947608947754,
+      "regularization/grad_norm_sq": 4127.0400390625,
+      "regularization/w2": 0.8687969446182251,
+      "rewards/chosen": 0.16640804223054964,
+      "rewards/margins": 0.10143800048816165,
+      "rewards/rejected": 0.06497004174238799,
+      "stability/repetition_rate_mean": 0.23209978640079498,
+      "stability/response_length_mean": 36.25,
+      "stability/response_length_std": 30.76988410949707,
+      "stability/response_length_var": 946.7857055664062,
+      "stability/token_entropy_mean": 3.67041015625,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.7240987420082092,
+      "kl": 0.9491209983825684,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56657488.192771085,
+      "logits/rejected": -48037266.28571428,
+      "logps/chosen": -354.8600809487952,
+      "logps/rejected": -194.32765574269482,
+      "loss": 1.3334,
+      "loss/base_kto": 0.4863254725933075,
+      "metrics/advantage_var": 4.277890682220459,
+      "regularization/grad_norm_sq": 4247.8359375,
+      "regularization/w2": 0.8748798370361328,
+      "rewards/chosen": 0.15770882296274943,
+      "rewards/margins": 0.10972130923479342,
+      "rewards/rejected": 0.047987513727956004,
+      "stability/repetition_rate_mean": 0.18672296404838562,
+      "stability/response_length_mean": 29.75,
+      "stability/response_length_std": 26.015106201171875,
+      "stability/response_length_var": 676.7857055664062,
+      "stability/token_entropy_mean": 4.822265625,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.6430587768554688,
+      "kl": 1.032440423965454,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52647628.17177914,
+      "logits/rejected": -47983459.46496815,
+      "logps/chosen": -374.4316765720859,
+      "logps/rejected": -210.53918690286625,
+      "loss": 1.321,
+      "loss/base_kto": 0.48797211050987244,
+      "metrics/advantage_var": 4.151429653167725,
+      "regularization/grad_norm_sq": 3608.227294921875,
+      "regularization/w2": 0.8262709975242615,
+      "rewards/chosen": 0.17141536408406824,
+      "rewards/margins": 0.09551231590283814,
+      "rewards/rejected": 0.0759030481812301,
+      "stability/repetition_rate_mean": 0.2868303656578064,
+      "stability/response_length_mean": 35.25,
+      "stability/response_length_std": 31.129911422729492,
+      "stability/response_length_var": 969.0714111328125,
+      "stability/token_entropy_mean": 3.130859375,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.6488748788833618,
+      "kl": 1.0751243829727173,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54614894.61728395,
+      "logits/rejected": -49520795.5443038,
+      "logps/chosen": -310.31404320987656,
+      "logps/rejected": -196.53752966772151,
+      "loss": 1.3781,
+      "loss/base_kto": 0.4869016706943512,
+      "metrics/advantage_var": 7.557289123535156,
+      "regularization/grad_norm_sq": 7128.36279296875,
+      "regularization/w2": 0.9516963958740234,
+      "rewards/chosen": 0.16873404420452354,
+      "rewards/margins": 0.1185720879727182,
+      "rewards/rejected": 0.05016195623180534,
+      "stability/repetition_rate_mean": 0.2109375,
+      "stability/response_length_mean": 32.875,
+      "stability/response_length_std": 33.30996322631836,
+      "stability/response_length_var": 1109.5535888671875,
+      "stability/token_entropy_mean": 3.71875,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.6966060996055603,
+      "kl": 1.1208827495574951,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53095589.556886226,
+      "logits/rejected": -49707021.385620914,
+      "logps/chosen": -390.324967252994,
+      "logps/rejected": -210.44881025326796,
+      "loss": 1.3626,
+      "loss/base_kto": 0.4836917519569397,
+      "metrics/advantage_var": 7.099040985107422,
+      "regularization/grad_norm_sq": 5476.40771484375,
+      "regularization/w2": 0.9177618026733398,
+      "rewards/chosen": 0.20803529202581167,
+      "rewards/margins": 0.1330566630569308,
+      "rewards/rejected": 0.07497862896888084,
+      "stability/repetition_rate_mean": 0.2964409589767456,
+      "stability/response_length_mean": 20.375,
+      "stability/response_length_std": 27.645395278930664,
+      "stability/response_length_var": 764.2678833007812,
+      "stability/token_entropy_mean": 4.3505859375,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.6411679983139038,
+      "kl": 1.1453860998153687,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60050742.5032258,
+      "logits/rejected": -53805540.07272727,
+      "logps/chosen": -390.8772681451613,
+      "logps/rejected": -197.0868252840909,
+      "loss": 1.3508,
+      "loss/base_kto": 0.48641711473464966,
+      "metrics/advantage_var": 4.3691325187683105,
+      "regularization/grad_norm_sq": 4039.490966796875,
+      "regularization/w2": 0.8582512736320496,
+      "rewards/chosen": 0.19076536855389994,
+      "rewards/margins": 0.1075151782814242,
+      "rewards/rejected": 0.08325019027247574,
+      "stability/repetition_rate_mean": 0.3895089328289032,
+      "stability/response_length_mean": 49.0,
+      "stability/response_length_std": 27.82085609436035,
+      "stability/response_length_var": 774.0,
+      "stability/token_entropy_mean": 5.19921875,
+      "step": 195
+    },
+    {
+      "epoch": 1.0,
+      "step": 196,
+      "total_flos": 3.1861866257511875e+18,
+      "train_loss": 1.359869228333843,
+      "train_runtime": 4299.7361,
+      "train_samples_per_second": 23.257,
+      "train_steps_per_second": 0.046
+    }
+  ],
+  "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_w2_delta0.01/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9125ddbb4efe23bb499afbabc466a81ea7d184e9e0ac5bd7dc5acdf9cbac0b42
+size 8081