Upload folder using huggingface_hub

mistral7b_kto_w1_gamma0.001/README.md

@@ -0,0 +1,64 @@
+---
+library_name: peft
+license: other
+base_model: mistralai/Mistral-7B-v0.1
+tags:
+- base_model:adapter:mistralai/Mistral-7B-v0.1
+- llama-factory
+- lora
+- transformers
+pipeline_tag: text-generation
+model-index:
+- name: mistral7b_kto_w1_gamma0.001
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+# mistral7b_kto_w1_gamma0.001
+
+This model is a fine-tuned version of [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) on the shp_kto_train dataset.
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 1e-06
+- train_batch_size: 8
+- eval_batch_size: 8
+- seed: 42
+- distributed_type: multi-GPU
+- num_devices: 8
+- gradient_accumulation_steps: 8
+- total_train_batch_size: 512
+- total_eval_batch_size: 64
+- optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
+- lr_scheduler_type: cosine
+- lr_scheduler_warmup_steps: 100
+- num_epochs: 1.0
+
+### Training results
+
+
+
+### Framework versions
+
+- PEFT 0.17.1
+- Transformers 4.57.1
+- Pytorch 2.10.0+cu128
+- Datasets 4.0.0
+- Tokenizers 0.22.2

mistral7b_kto_w1_gamma0.001/adapter_config.json

@@ -0,0 +1,42 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "mistralai/Mistral-7B-v0.1",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "gate_proj",
+    "q_proj",
+    "v_proj",
+    "k_proj",
+    "down_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_w1_gamma0.001/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5aebf32544f03ec9b8149bd05bfe7363f21b5fd81dbdb84dd607ea355b6e9f1e
+size 42002584

mistral7b_kto_w1_gamma0.001/all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.6270372308030421,
+    "train_runtime": 3152.8646,
+    "train_samples_per_second": 31.717,
+    "train_steps_per_second": 0.062
+}

mistral7b_kto_w1_gamma0.001/chat_template.jinja

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/README.md

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/adapter_config.json

@@ -0,0 +1,42 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "mistralai/Mistral-7B-v0.1",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "gate_proj",
+    "q_proj",
+    "v_proj",
+    "k_proj",
+    "down_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_w1_gamma0.001/checkpoint-196/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5aebf32544f03ec9b8149bd05bfe7363f21b5fd81dbdb84dd607ea355b6e9f1e
+size 42002584

mistral7b_kto_w1_gamma0.001/checkpoint-196/chat_template.jinja

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d4dd3b8c8a0af44e054cc94dbf3d18a33495e25dc4fa9e65fab9662c8b1ca7c
+size 31467333

mistral7b_kto_w1_gamma0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:640d306a5561a38626031b725e93590d33417f4c7fdf2cdd9a9e1805d4500952
+size 31467333

mistral7b_kto_w1_gamma0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c864223180fa39c59961182fef0781b8c8e24f65e8148d3168ff228de24b3af
+size 31467333

mistral7b_kto_w1_gamma0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7d35b74daa77b16e9f428f167b88f9213a0efccc898afdcec89b8ab70f0071c
+size 31467397

mistral7b_kto_w1_gamma0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:894a4941c192c2e77e7e8337a4a56546db9dd95e4a9d66f7c3ad874a47e83ffa
+size 31467397

mistral7b_kto_w1_gamma0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b76158b7f867559bdf1b3ff685814754c3b9ba21124d56b09c6e1a8e083fd90d
+size 31467397

mistral7b_kto_w1_gamma0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02956437aac73c5725637d4e2c7539bd0aea7f85134d991d811f5fd26e19b940
+size 31467397

mistral7b_kto_w1_gamma0.001/checkpoint-196/global_step196/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5670e98b4ffd3a6e431a3a1812a0d7df76f22cfe859e746b583eb39794909372
+size 31467397

mistral7b_kto_w1_gamma0.001/checkpoint-196/global_step196/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c560779105ef06032f5ad437c6353654b2b7111eac159a39c9769abd93ede6fc
+size 42288409

mistral7b_kto_w1_gamma0.001/checkpoint-196/latest

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/rng_state_0.pth

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/rng_state_1.pth

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/rng_state_2.pth

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/rng_state_3.pth

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/rng_state_4.pth

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/rng_state_5.pth

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/rng_state_6.pth

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/rng_state_7.pth

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/scheduler.pt

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/special_tokens_map.json

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/tokenizer.model

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/tokenizer_config.json

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

mistral7b_kto_w1_gamma0.001/checkpoint-196/trainer_state.json

@@ -0,0 +1,965 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 196,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.025591810620601407,
+      "grad_norm": 0.5678192377090454,
+      "kl": 0.009494942612946033,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59717214.315789476,
+      "logits/rejected": -50418761.14285714,
+      "logps/chosen": -439.0025699013158,
+      "logps/rejected": -198.39857700892858,
+      "loss": 0.5319,
+      "loss/base_kto": 0.49962836503982544,
+      "metrics/advantage_var": 0.04574162885546684,
+      "regularization/lipschitz_norm": 36.127017974853516,
+      "regularization/w1": 0.03612701967358589,
+      "rewards/chosen": 0.004304667836741397,
+      "rewards/margins": 0.0032234100442855877,
+      "rewards/rejected": 0.0010812577924558095,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.6058618426322937,
+      "kl": 0.017178865149617195,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59305279.16883117,
+      "logits/rejected": -50020561.73493976,
+      "logps/chosen": -429.8796672077922,
+      "logps/rejected": -192.88354198042168,
+      "loss": 0.5474,
+      "loss/base_kto": 0.4991108477115631,
+      "metrics/advantage_var": 0.07220327109098434,
+      "regularization/lipschitz_norm": 51.88991165161133,
+      "regularization/w1": 0.051889922469854355,
+      "rewards/chosen": 0.00588285071509225,
+      "rewards/margins": 0.007441361728100276,
+      "rewards/rejected": -0.0015585110130080257,
+      "stability/repetition_rate_mean": 0.3605554699897766,
+      "stability/response_length_mean": 44.25,
+      "stability/response_length_std": 28.13614845275879,
+      "stability/response_length_var": 791.6428833007812,
+      "stability/token_entropy_mean": 3.3544921875,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.6088977456092834,
+      "kl": 0.013535099104046822,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56235889.26060606,
+      "logits/rejected": -51918445.006451614,
+      "logps/chosen": -404.12386363636364,
+      "logps/rejected": -200.66649445564516,
+      "loss": 0.5528,
+      "loss/base_kto": 0.5002279281616211,
+      "metrics/advantage_var": 0.057587385177612305,
+      "regularization/lipschitz_norm": 47.16032409667969,
+      "regularization/w1": 0.047160327434539795,
+      "rewards/chosen": 0.0006950111100167939,
+      "rewards/margins": -0.0020621884311981795,
+      "rewards/rejected": 0.0027571995412149736,
+      "stability/repetition_rate_mean": 0.2651909589767456,
+      "stability/response_length_mean": 27.125,
+      "stability/response_length_std": 30.79163932800293,
+      "stability/response_length_var": 948.125,
+      "stability/token_entropy_mean": 3.8203125,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.6186808347702026,
+      "kl": 0.014954889193177223,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51655552.0,
+      "logits/rejected": -48047257.6,
+      "logps/chosen": -343.684423828125,
+      "logps/rejected": -225.644677734375,
+      "loss": 0.5531,
+      "loss/base_kto": 0.500113308429718,
+      "metrics/advantage_var": 0.0725272074341774,
+      "regularization/lipschitz_norm": 51.32513427734375,
+      "regularization/w1": 0.05132513865828514,
+      "rewards/chosen": 0.0012291071005165577,
+      "rewards/margins": -0.0006385699845850468,
+      "rewards/rejected": 0.0018676770851016045,
+      "stability/repetition_rate_mean": 0.2174789160490036,
+      "stability/response_length_mean": 44.0,
+      "stability/response_length_std": 28.0611572265625,
+      "stability/response_length_var": 787.4285888671875,
+      "stability/token_entropy_mean": 4.462890625,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.5771578550338745,
+      "kl": 0.019569864496588707,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56482780.68965517,
+      "logits/rejected": -52020662.85714286,
+      "logps/chosen": -378.46988146551723,
+      "logps/rejected": -202.78185267857143,
+      "loss": 0.5539,
+      "loss/base_kto": 0.49986138939857483,
+      "metrics/advantage_var": 0.05885173752903938,
+      "regularization/lipschitz_norm": 46.89044189453125,
+      "regularization/w1": 0.04689044505357742,
+      "rewards/chosen": 0.002159434351427802,
+      "rewards/margins": 0.0010595623583629214,
+      "rewards/rejected": 0.0010998719930648804,
+      "stability/repetition_rate_mean": 0.28109976649284363,
+      "stability/response_length_mean": 50.25,
+      "stability/response_length_std": 25.549671173095703,
+      "stability/response_length_var": 652.7857055664062,
+      "stability/token_entropy_mean": 4.5078125,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.603465735912323,
+      "kl": 0.018669696524739265,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57732102.87248322,
+      "logits/rejected": -51518679.578947365,
+      "logps/chosen": -468.6152999161074,
+      "logps/rejected": -198.2326845760234,
+      "loss": 0.5527,
+      "loss/base_kto": 0.5005181431770325,
+      "metrics/advantage_var": 0.07389438152313232,
+      "regularization/lipschitz_norm": 52.765899658203125,
+      "regularization/w1": 0.05276590213179588,
+      "rewards/chosen": -0.003406525858296644,
+      "rewards/margins": -0.004925517465271106,
+      "rewards/rejected": 0.0015189916069744624,
+      "stability/repetition_rate_mean": 0.158203125,
+      "stability/response_length_mean": 48.375,
+      "stability/response_length_std": 28.93311309814453,
+      "stability/response_length_var": 837.125,
+      "stability/token_entropy_mean": 4.595703125,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6781691908836365,
+      "kl": 0.03436283394694328,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57640635.94936709,
+      "logits/rejected": -53904560.98765432,
+      "logps/chosen": -385.9036293512658,
+      "logps/rejected": -195.09850019290124,
+      "loss": 0.5528,
+      "loss/base_kto": 0.4999621510505676,
+      "metrics/advantage_var": 0.05386241897940636,
+      "regularization/lipschitz_norm": 46.16049575805664,
+      "regularization/w1": 0.0461605004966259,
+      "rewards/chosen": 0.003288782095607323,
+      "rewards/margins": 0.0004902812913537566,
+      "rewards/rejected": 0.0027985008042535663,
+      "stability/repetition_rate_mean": 0.3537946343421936,
+      "stability/response_length_mean": 56.875,
+      "stability/response_length_std": 20.152544021606445,
+      "stability/response_length_var": 406.125,
+      "stability/token_entropy_mean": 5.416015625,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.5843126177787781,
+      "kl": 0.027318641543388367,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56632812.8,
+      "logits/rejected": -50224707.2,
+      "logps/chosen": -421.8916015625,
+      "logps/rejected": -219.2817138671875,
+      "loss": 0.5509,
+      "loss/base_kto": 0.5002434849739075,
+      "metrics/advantage_var": 0.07726090401411057,
+      "regularization/lipschitz_norm": 52.84709548950195,
+      "regularization/w1": 0.05284709855914116,
+      "rewards/chosen": 0.0011051620356738567,
+      "rewards/margins": -0.0016500151716172696,
+      "rewards/rejected": 0.0027551772072911263,
+      "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": 0.6399968862533569,
+      "kl": 0.04743483290076256,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52704039.51677852,
+      "logits/rejected": -50776093.94152047,
+      "logps/chosen": -352.3003355704698,
+      "logps/rejected": -243.84438961988303,
+      "loss": 0.5532,
+      "loss/base_kto": 0.4999566078186035,
+      "metrics/advantage_var": 0.07218363136053085,
+      "regularization/lipschitz_norm": 56.39221954345703,
+      "regularization/w1": 0.05639222264289856,
+      "rewards/chosen": 0.002814074090663219,
+      "rewards/margins": -0.00016493607860582982,
+      "rewards/rejected": 0.0029790101692690487,
+      "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": 4.767578125,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.5783849954605103,
+      "kl": 0.051512785255908966,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54039232.4295302,
+      "logits/rejected": -48303834.57309941,
+      "logps/chosen": -375.13357277684565,
+      "logps/rejected": -180.2006807383041,
+      "loss": 0.5494,
+      "loss/base_kto": 0.4995388090610504,
+      "metrics/advantage_var": 0.04794999957084656,
+      "regularization/lipschitz_norm": 41.682228088378906,
+      "regularization/w1": 0.04168223217129707,
+      "rewards/chosen": 0.006553848717836726,
+      "rewards/margins": 0.0025606038216785853,
+      "rewards/rejected": 0.003993244896158141,
+      "stability/repetition_rate_mean": 0.3249044120311737,
+      "stability/response_length_mean": 55.375,
+      "stability/response_length_std": 16.361431121826172,
+      "stability/response_length_var": 267.6964416503906,
+      "stability/token_entropy_mean": 4.068359375,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.6621458530426025,
+      "kl": 0.05490356683731079,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56948316.22360248,
+      "logits/rejected": -49577713.50943396,
+      "logps/chosen": -344.4683618012422,
+      "logps/rejected": -230.42428262578616,
+      "loss": 0.5531,
+      "loss/base_kto": 0.4997487962245941,
+      "metrics/advantage_var": 0.05243971571326256,
+      "regularization/lipschitz_norm": 45.507625579833984,
+      "regularization/w1": 0.04550762102007866,
+      "rewards/chosen": 0.0043661816519980105,
+      "rewards/margins": 0.0019911858475652076,
+      "rewards/rejected": 0.002374995804432803,
+      "stability/repetition_rate_mean": 0.1015625,
+      "stability/response_length_mean": 24.5,
+      "stability/response_length_std": 32.710853576660156,
+      "stability/response_length_var": 1070.0,
+      "stability/token_entropy_mean": 4.6484375,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.5836600661277771,
+      "kl": 0.05875139310956001,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52669911.660606064,
+      "logits/rejected": -47820423.43225806,
+      "logps/chosen": -359.61126893939394,
+      "logps/rejected": -178.15209173387098,
+      "loss": 0.5496,
+      "loss/base_kto": 0.4995574951171875,
+      "metrics/advantage_var": 0.05988660454750061,
+      "regularization/lipschitz_norm": 46.30636215209961,
+      "regularization/w1": 0.046306367963552475,
+      "rewards/chosen": 0.010953712463378906,
+      "rewards/margins": 0.0034705385085075126,
+      "rewards/rejected": 0.007483173954871393,
+      "stability/repetition_rate_mean": 0.20815381407737732,
+      "stability/response_length_mean": 52.625,
+      "stability/response_length_std": 21.75800323486328,
+      "stability/response_length_var": 473.41070556640625,
+      "stability/token_entropy_mean": 5.314453125,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.5745318531990051,
+      "kl": 0.08697053045034409,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57561999.69032258,
+      "logits/rejected": -47265897.5030303,
+      "logps/chosen": -381.825,
+      "logps/rejected": -183.50355113636363,
+      "loss": 0.5544,
+      "loss/base_kto": 0.4995245635509491,
+      "metrics/advantage_var": 0.0773346945643425,
+      "regularization/lipschitz_norm": 53.16814041137695,
+      "regularization/w1": 0.053168147802352905,
+      "rewards/chosen": 0.012180716760696904,
+      "rewards/margins": 0.00430064315553518,
+      "rewards/rejected": 0.007880073605161724,
+      "stability/repetition_rate_mean": 0.20286151766777039,
+      "stability/response_length_mean": 17.875,
+      "stability/response_length_std": 23.485177993774414,
+      "stability/response_length_var": 551.5535888671875,
+      "stability/token_entropy_mean": 5.6875,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5525624752044678,
+      "kl": 0.09267856180667877,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52762746.21935484,
+      "logits/rejected": -49031695.515151516,
+      "logps/chosen": -379.0071572580645,
+      "logps/rejected": -220.3510890151515,
+      "loss": 0.5558,
+      "loss/base_kto": 0.4993644654750824,
+      "metrics/advantage_var": 0.07783074676990509,
+      "regularization/lipschitz_norm": 53.91382598876953,
+      "regularization/w1": 0.05391382798552513,
+      "rewards/chosen": 0.012132589278682585,
+      "rewards/margins": 0.005023109807996106,
+      "rewards/rejected": 0.007109479470686479,
+      "stability/repetition_rate_mean": 0.119140625,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.3349609375,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.6093087792396545,
+      "kl": 0.10173000395298004,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49496755.89189189,
+      "logits/rejected": -47891048.18604651,
+      "logps/chosen": -314.1371674408784,
+      "logps/rejected": -228.80718568313952,
+      "loss": 0.5574,
+      "loss/base_kto": 0.4992505609989166,
+      "metrics/advantage_var": 0.07528817653656006,
+      "regularization/lipschitz_norm": 53.78229904174805,
+      "regularization/w1": 0.05378229543566704,
+      "rewards/chosen": 0.013364916717683946,
+      "rewards/margins": 0.0056003187528426805,
+      "rewards/rejected": 0.007764597964841266,
+      "stability/repetition_rate_mean": 0.06745792925357819,
+      "stability/response_length_mean": 22.75,
+      "stability/response_length_std": 31.07479476928711,
+      "stability/response_length_var": 965.6428833007812,
+      "stability/token_entropy_mean": 2.16015625,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.6032807230949402,
+      "kl": 0.13186340034008026,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58616967.62913907,
+      "logits/rejected": -53431344.47337278,
+      "logps/chosen": -403.86936051324506,
+      "logps/rejected": -200.9628328402367,
+      "loss": 0.5561,
+      "loss/base_kto": 0.4992029666900635,
+      "metrics/advantage_var": 0.0667102262377739,
+      "regularization/lipschitz_norm": 47.66132354736328,
+      "regularization/w1": 0.047661323100328445,
+      "rewards/chosen": 0.01607713478290482,
+      "rewards/margins": 0.006778749535228491,
+      "rewards/rejected": 0.00929838524767633,
+      "stability/repetition_rate_mean": 0.17916667461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.25390625,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.6124517321586609,
+      "kl": 0.1272238940000534,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51464935.22580645,
+      "logits/rejected": -47132777.5030303,
+      "logps/chosen": -358.1264616935484,
+      "logps/rejected": -230.3289772727273,
+      "loss": 0.5594,
+      "loss/base_kto": 0.4986273944377899,
+      "metrics/advantage_var": 0.0937027856707573,
+      "regularization/lipschitz_norm": 55.11994552612305,
+      "regularization/w1": 0.055119942873716354,
+      "rewards/chosen": 0.020604954996416645,
+      "rewards/margins": 0.011022517827948516,
+      "rewards/rejected": 0.00958243716846813,
+      "stability/repetition_rate_mean": 0.11328125,
+      "stability/response_length_mean": 32.5,
+      "stability/response_length_std": 33.67915725708008,
+      "stability/response_length_var": 1134.2857666015625,
+      "stability/token_entropy_mean": 4.9453125,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.6401332020759583,
+      "kl": 0.1703931838274002,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58414500.102564104,
+      "logits/rejected": -50554680.19512195,
+      "logps/chosen": -380.5234124599359,
+      "logps/rejected": -187.12747713414635,
+      "loss": 0.5657,
+      "loss/base_kto": 0.4978005886077881,
+      "metrics/advantage_var": 0.12542474269866943,
+      "regularization/lipschitz_norm": 66.9811782836914,
+      "regularization/w1": 0.06698118895292282,
+      "rewards/chosen": 0.026595650575099848,
+      "rewards/margins": 0.016664286864556843,
+      "rewards/rejected": 0.009931363710543005,
+      "stability/repetition_rate_mean": 0.3812144994735718,
+      "stability/response_length_mean": 30.625,
+      "stability/response_length_std": 28.779644012451172,
+      "stability/response_length_var": 828.2678833007812,
+      "stability/token_entropy_mean": 3.736328125,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.6237964034080505,
+      "kl": 0.16741442680358887,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50386140.35443038,
+      "logits/rejected": -48979835.25925926,
+      "logps/chosen": -314.7296034414557,
+      "logps/rejected": -207.09116994598764,
+      "loss": 0.5682,
+      "loss/base_kto": 0.49736830592155457,
+      "metrics/advantage_var": 0.1331707090139389,
+      "regularization/lipschitz_norm": 66.74359130859375,
+      "regularization/w1": 0.06674359738826752,
+      "rewards/chosen": 0.026378269436993177,
+      "rewards/margins": 0.021123482745445267,
+      "rewards/rejected": 0.005254786691547912,
+      "stability/repetition_rate_mean": 0.20546108484268188,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 31.1264705657959,
+      "stability/response_length_var": 968.8571166992188,
+      "stability/token_entropy_mean": 3.689453125,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.6090478301048279,
+      "kl": 0.20850256085395813,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54716139.24324324,
+      "logits/rejected": -50905552.37209302,
+      "logps/chosen": -359.56579919763516,
+      "logps/rejected": -216.39162427325581,
+      "loss": 0.5714,
+      "loss/base_kto": 0.4984969198703766,
+      "metrics/advantage_var": 0.13034458458423615,
+      "regularization/lipschitz_norm": 67.18830871582031,
+      "regularization/w1": 0.0671883150935173,
+      "rewards/chosen": 0.030437920544598554,
+      "rewards/margins": 0.011710877204226068,
+      "rewards/rejected": 0.018727043340372486,
+      "stability/repetition_rate_mean": 0.24010416865348816,
+      "stability/response_length_mean": 40.75,
+      "stability/response_length_std": 31.093177795410156,
+      "stability/response_length_var": 966.7857055664062,
+      "stability/token_entropy_mean": 4.140625,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6770948171615601,
+      "kl": 0.2234809845685959,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59376942.26506024,
+      "logits/rejected": -53793147.01298701,
+      "logps/chosen": -399.5205666415663,
+      "logps/rejected": -209.8428622159091,
+      "loss": 0.5793,
+      "loss/base_kto": 0.4972439706325531,
+      "metrics/advantage_var": 0.22079260647296906,
+      "regularization/lipschitz_norm": 90.85485076904297,
+      "regularization/w1": 0.09085484594106674,
+      "rewards/chosen": 0.03818985353033227,
+      "rewards/margins": 0.022094057341581444,
+      "rewards/rejected": 0.016095796188750823,
+      "stability/repetition_rate_mean": 0.2014508843421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.40625,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.6547015309333801,
+      "kl": 0.26629599928855896,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57391949.10843374,
+      "logits/rejected": -53613863.896103896,
+      "logps/chosen": -378.19820689006025,
+      "logps/rejected": -205.89108157467533,
+      "loss": 0.5834,
+      "loss/base_kto": 0.4977465569972992,
+      "metrics/advantage_var": 0.21738283336162567,
+      "regularization/lipschitz_norm": 89.9118881225586,
+      "regularization/w1": 0.08991190046072006,
+      "rewards/chosen": 0.036506069711892,
+      "rewards/margins": 0.01679964862767379,
+      "rewards/rejected": 0.01970642108421821,
+      "stability/repetition_rate_mean": 0.31048768758773804,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.911476135253906,
+      "stability/response_length_var": 894.6964111328125,
+      "stability/token_entropy_mean": 4.3857421875,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.6483462452888489,
+      "kl": 0.32535991072654724,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60606263.088607594,
+      "logits/rejected": -53428704.39506173,
+      "logps/chosen": -364.06833465189874,
+      "logps/rejected": -200.9859905478395,
+      "loss": 0.5942,
+      "loss/base_kto": 0.49606046080589294,
+      "metrics/advantage_var": 0.26823315024375916,
+      "regularization/lipschitz_norm": 103.65421295166016,
+      "regularization/w1": 0.10365422070026398,
+      "rewards/chosen": 0.052421219741241844,
+      "rewards/margins": 0.031364458764897865,
+      "rewards/rejected": 0.02105676097634398,
+      "stability/repetition_rate_mean": 0.3151041865348816,
+      "stability/response_length_mean": 49.0,
+      "stability/response_length_std": 27.795169830322266,
+      "stability/response_length_var": 772.5714111328125,
+      "stability/token_entropy_mean": 4.830078125,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.6184374690055847,
+      "kl": 0.3553595244884491,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57748794.5620915,
+      "logits/rejected": -54355410.01197605,
+      "logps/chosen": -383.7023845996732,
+      "logps/rejected": -211.95403723802394,
+      "loss": 0.6004,
+      "loss/base_kto": 0.4964432418346405,
+      "metrics/advantage_var": 0.3870581090450287,
+      "regularization/lipschitz_norm": 109.87580871582031,
+      "regularization/w1": 0.10987582057714462,
+      "rewards/chosen": 0.05425496818193423,
+      "rewards/margins": 0.028023887709821817,
+      "rewards/rejected": 0.026231080472112416,
+      "stability/repetition_rate_mean": 0.1723366528749466,
+      "stability/response_length_mean": 27.875,
+      "stability/response_length_std": 30.31943130493164,
+      "stability/response_length_var": 919.2678833007812,
+      "stability/token_entropy_mean": 4.64453125,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.6122660636901855,
+      "kl": 0.39908552169799805,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60685662.68493151,
+      "logits/rejected": -55654294.06896552,
+      "logps/chosen": -416.8324058219178,
+      "logps/rejected": -216.1103627873563,
+      "loss": 0.6157,
+      "loss/base_kto": 0.4964618682861328,
+      "metrics/advantage_var": 0.5276581645011902,
+      "regularization/lipschitz_norm": 126.17388153076172,
+      "regularization/w1": 0.12617388367652893,
+      "rewards/chosen": 0.05468781353676156,
+      "rewards/margins": 0.029002990971923985,
+      "rewards/rejected": 0.025684822564837576,
+      "stability/repetition_rate_mean": 0.15937499701976776,
+      "stability/response_length_mean": 15.875,
+      "stability/response_length_std": 24.584766387939453,
+      "stability/response_length_var": 604.4107055664062,
+      "stability/token_entropy_mean": 3.49609375,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.5939815044403076,
+      "kl": 0.44654884934425354,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57391097.393548384,
+      "logits/rejected": -53830773.915151514,
+      "logps/chosen": -391.98465221774194,
+      "logps/rejected": -183.64661458333333,
+      "loss": 0.6243,
+      "loss/base_kto": 0.49498701095581055,
+      "metrics/advantage_var": 0.7337404489517212,
+      "regularization/lipschitz_norm": 144.0272216796875,
+      "regularization/w1": 0.1440272331237793,
+      "rewards/chosen": 0.06732863149335308,
+      "rewards/margins": 0.04009907965785947,
+      "rewards/rejected": 0.02722955183549361,
+      "stability/repetition_rate_mean": 0.22330255806446075,
+      "stability/response_length_mean": 38.0,
+      "stability/response_length_std": 29.13515281677246,
+      "stability/response_length_var": 848.8571166992188,
+      "stability/token_entropy_mean": 3.91796875,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5939186215400696,
+      "kl": 0.4990025460720062,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59841391.50920245,
+      "logits/rejected": -55011230.1656051,
+      "logps/chosen": -401.0247795245399,
+      "logps/rejected": -247.52191978503186,
+      "loss": 0.6335,
+      "loss/base_kto": 0.494366854429245,
+      "metrics/advantage_var": 0.9336435198783875,
+      "regularization/lipschitz_norm": 167.570068359375,
+      "regularization/w1": 0.167570099234581,
+      "rewards/chosen": 0.07913663197148797,
+      "rewards/margins": 0.044253570179806793,
+      "rewards/rejected": 0.03488306179168118,
+      "stability/repetition_rate_mean": 0.2270585298538208,
+      "stability/response_length_mean": 29.875,
+      "stability/response_length_std": 29.502723693847656,
+      "stability/response_length_var": 870.4107055664062,
+      "stability/token_entropy_mean": 3.9013671875,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.7301244139671326,
+      "kl": 0.4819028973579407,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57072553.11515152,
+      "logits/rejected": -50065289.08387097,
+      "logps/chosen": -366.36526988636365,
+      "logps/rejected": -189.91263860887096,
+      "loss": 0.6499,
+      "loss/base_kto": 0.4916805922985077,
+      "metrics/advantage_var": 1.7661739587783813,
+      "regularization/lipschitz_norm": 183.31504821777344,
+      "regularization/w1": 0.18331506848335266,
+      "rewards/chosen": 0.08244560126102332,
+      "rewards/margins": 0.0702617435278198,
+      "rewards/rejected": 0.012183857733203518,
+      "stability/repetition_rate_mean": 0.228515625,
+      "stability/response_length_mean": 56.25,
+      "stability/response_length_std": 21.920310974121094,
+      "stability/response_length_var": 480.5,
+      "stability/token_entropy_mean": 5.134765625,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.6743220090866089,
+      "kl": 0.5741755366325378,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51148250.98795181,
+      "logits/rejected": -51665840.20779221,
+      "logps/chosen": -329.439406061747,
+      "logps/rejected": -199.1564275568182,
+      "loss": 0.6604,
+      "loss/base_kto": 0.49365338683128357,
+      "metrics/advantage_var": 0.856822669506073,
+      "regularization/lipschitz_norm": 163.44825744628906,
+      "regularization/w1": 0.1634482592344284,
+      "rewards/chosen": 0.08324084224471126,
+      "rewards/margins": 0.050092330581051434,
+      "rewards/rejected": 0.03314851166365983,
+      "stability/repetition_rate_mean": 0.21016408503055573,
+      "stability/response_length_mean": 22.625,
+      "stability/response_length_std": 26.359804153442383,
+      "stability/response_length_var": 694.8392944335938,
+      "stability/token_entropy_mean": 5.6640625,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.6430969834327698,
+      "kl": 0.6443031430244446,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55426325.08235294,
+      "logits/rejected": -50193100.8,
+      "logps/chosen": -396.8122242647059,
+      "logps/rejected": -207.33588541666666,
+      "loss": 0.6752,
+      "loss/base_kto": 0.4903334677219391,
+      "metrics/advantage_var": 1.2321922779083252,
+      "regularization/lipschitz_norm": 185.36038208007812,
+      "regularization/w1": 0.1853603571653366,
+      "rewards/chosen": 0.11221859875847312,
+      "rewards/margins": 0.07321333511202943,
+      "rewards/rejected": 0.03900526364644368,
+      "stability/repetition_rate_mean": 0.23066024482250214,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.193626403808594,
+      "stability/response_length_var": 852.2678833007812,
+      "stability/token_entropy_mean": 4.58984375,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.6289855241775513,
+      "kl": 0.6507260799407959,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54025167.8117647,
+      "logits/rejected": -49363421.86666667,
+      "logps/chosen": -360.06833639705883,
+      "logps/rejected": -191.04104166666667,
+      "loss": 0.7031,
+      "loss/base_kto": 0.49213314056396484,
+      "metrics/advantage_var": 2.246683120727539,
+      "regularization/lipschitz_norm": 223.3385467529297,
+      "regularization/w1": 0.2233385592699051,
+      "rewards/chosen": 0.09209441016702091,
+      "rewards/margins": 0.06563791106728947,
+      "rewards/rejected": 0.026456499099731447,
+      "stability/repetition_rate_mean": 0.3115246295928955,
+      "stability/response_length_mean": 58.125,
+      "stability/response_length_std": 10.960155487060547,
+      "stability/response_length_var": 120.125,
+      "stability/token_entropy_mean": 4.5673828125,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.619614839553833,
+      "kl": 0.7176083326339722,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55248286.47619048,
+      "logits/rejected": -48961236.21052632,
+      "logps/chosen": -415.52762276785717,
+      "logps/rejected": -186.48067434210526,
+      "loss": 0.7229,
+      "loss/base_kto": 0.487854927778244,
+      "metrics/advantage_var": 2.0578365325927734,
+      "regularization/lipschitz_norm": 231.88279724121094,
+      "regularization/w1": 0.23188279569149017,
+      "rewards/chosen": 0.13677164486476354,
+      "rewards/margins": 0.09510390426879539,
+      "rewards/rejected": 0.041667740595968145,
+      "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": 4.484375,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.7249348163604736,
+      "kl": 0.770777702331543,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54404449.10344828,
+      "logits/rejected": -49778060.2739726,
+      "logps/chosen": -369.70171066810343,
+      "logps/rejected": -212.1182041952055,
+      "loss": 0.7436,
+      "loss/base_kto": 0.49073153734207153,
+      "metrics/advantage_var": 1.810006022453308,
+      "regularization/lipschitz_norm": 221.78726196289062,
+      "regularization/w1": 0.22178728878498077,
+      "rewards/chosen": 0.12150281051109577,
+      "rewards/margins": 0.07411109539603195,
+      "rewards/rejected": 0.047391715115063814,
+      "stability/repetition_rate_mean": 0.2411106526851654,
+      "stability/response_length_mean": 22.5,
+      "stability/response_length_std": 28.33977508544922,
+      "stability/response_length_var": 803.1428833007812,
+      "stability/token_entropy_mean": 4.87109375,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.621860682964325,
+      "kl": 0.9223539233207703,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59132721.988165684,
+      "logits/rejected": -52588998.357615896,
+      "logps/chosen": -439.69253883136093,
+      "logps/rejected": -190.58294960678808,
+      "loss": 0.7621,
+      "loss/base_kto": 0.48786044120788574,
+      "metrics/advantage_var": 2.32456111907959,
+      "regularization/lipschitz_norm": 273.0086975097656,
+      "regularization/w1": 0.27300870418548584,
+      "rewards/chosen": 0.16493017433663093,
+      "rewards/margins": 0.09744154819757748,
+      "rewards/rejected": 0.06748862613905345,
+      "stability/repetition_rate_mean": 0.3512404263019562,
+      "stability/response_length_mean": 44.125,
+      "stability/response_length_std": 28.43256378173828,
+      "stability/response_length_var": 808.4107055664062,
+      "stability/token_entropy_mean": 3.59228515625,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.7261084318161011,
+      "kl": 0.9431293606758118,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56663225.06024096,
+      "logits/rejected": -48029942.02597403,
+      "logps/chosen": -354.8965549698795,
+      "logps/rejected": -194.3030768060065,
+      "loss": 0.7733,
+      "loss/base_kto": 0.4871121346950531,
+      "metrics/advantage_var": 4.2855119705200195,
+      "regularization/lipschitz_norm": 322.96240234375,
+      "regularization/w1": 0.3229624629020691,
+      "rewards/chosen": 0.15406293754118033,
+      "rewards/margins": 0.1036179753988305,
+      "rewards/rejected": 0.050444962142349836,
+      "stability/repetition_rate_mean": 0.19270984828472137,
+      "stability/response_length_mean": 26.875,
+      "stability/response_length_std": 25.570281982421875,
+      "stability/response_length_var": 653.8392944335938,
+      "stability/token_entropy_mean": 4.849609375,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.6424126625061035,
+      "kl": 1.0352585315704346,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52662171.48466258,
+      "logits/rejected": -47984167.133757964,
+      "logps/chosen": -374.4056269171779,
+      "logps/rejected": -210.5522243232484,
+      "loss": 0.8001,
+      "loss/base_kto": 0.4874431788921356,
+      "metrics/advantage_var": 4.169354438781738,
+      "regularization/lipschitz_norm": 315.4919128417969,
+      "regularization/w1": 0.3154919147491455,
+      "rewards/chosen": 0.1740191524014151,
+      "rewards/margins": 0.0994211801875728,
+      "rewards/rejected": 0.0745979722138423,
+      "stability/repetition_rate_mean": 0.2008928656578064,
+      "stability/response_length_mean": 35.25,
+      "stability/response_length_std": 31.129911422729492,
+      "stability/response_length_var": 969.0714111328125,
+      "stability/token_entropy_mean": 3.46875,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.6501648426055908,
+      "kl": 1.078233003616333,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54614294.12345679,
+      "logits/rejected": -49507992.30379747,
+      "logps/chosen": -310.3076774691358,
+      "logps/rejected": -196.53772745253164,
+      "loss": 0.8358,
+      "loss/base_kto": 0.48679420351982117,
+      "metrics/advantage_var": 7.409839630126953,
+      "regularization/lipschitz_norm": 392.8832702636719,
+      "regularization/w1": 0.3928832709789276,
+      "rewards/chosen": 0.16936941500063296,
+      "rewards/margins": 0.11922688233813265,
+      "rewards/rejected": 0.05014253266250031,
+      "stability/repetition_rate_mean": 0.19140625,
+      "stability/response_length_mean": 32.875,
+      "stability/response_length_std": 33.30996322631836,
+      "stability/response_length_var": 1109.5535888671875,
+      "stability/token_entropy_mean": 3.705078125,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.6980246901512146,
+      "kl": 1.122229814529419,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53080229.556886226,
+      "logits/rejected": -49681528.47058824,
+      "logps/chosen": -390.28709767964074,
+      "logps/rejected": -210.45874183006535,
+      "loss": 0.8981,
+      "loss/base_kto": 0.483151912689209,
+      "metrics/advantage_var": 7.226253032684326,
+      "regularization/lipschitz_norm": 438.9554748535156,
+      "regularization/w1": 0.4389554560184479,
+      "rewards/chosen": 0.21182136764069517,
+      "rewards/margins": 0.13783560089950975,
+      "rewards/rejected": 0.07398576674118541,
+      "stability/repetition_rate_mean": 0.19138340651988983,
+      "stability/response_length_mean": 18.625,
+      "stability/response_length_std": 26.52189064025879,
+      "stability/response_length_var": 703.4107055664062,
+      "stability/token_entropy_mean": 4.580078125,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.6413615942001343,
+      "kl": 1.1505008935928345,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60036446.14193548,
+      "logits/rejected": -53804826.375757575,
+      "logps/chosen": -390.84732862903223,
+      "logps/rejected": -197.08839962121212,
+      "loss": 0.8839,
+      "loss/base_kto": 0.4860563278198242,
+      "metrics/advantage_var": 4.479440212249756,
+      "regularization/lipschitz_norm": 372.052001953125,
+      "regularization/w1": 0.37205204367637634,
+      "rewards/chosen": 0.19376148100822202,
+      "rewards/margins": 0.11066871057624107,
+      "rewards/rejected": 0.08309277043198095,
+      "stability/repetition_rate_mean": 0.3934151828289032,
+      "stability/response_length_mean": 49.0,
+      "stability/response_length_std": 27.82085609436035,
+      "stability/response_length_var": 774.0,
+      "stability/token_entropy_mean": 5.1875,
+      "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_w1_gamma0.001/checkpoint-196/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:137a09d8908cce7eb46d8c86b2908fab853f68db7a0ae41f0299b4a8db223316
+size 8081

mistral7b_kto_w1_gamma0.001/checkpoint-196/zero_to_fp32.py

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

mistral7b_kto_w1_gamma0.001/special_tokens_map.json

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

mistral7b_kto_w1_gamma0.001/stability_metrics.jsonl

@@ -0,0 +1,38 @@
+{"stability/response_length_mean": 44.25, "stability/response_length_std": 28.13614845275879, "stability/response_length_var": 791.6428833007812, "stability/token_entropy_mean": 3.3544921875, "stability/repetition_rate_mean": 0.3605554699897766, "step": 10}
+{"stability/response_length_mean": 27.125, "stability/response_length_std": 30.79163932800293, "stability/response_length_var": 948.125, "stability/token_entropy_mean": 3.8203125, "stability/repetition_rate_mean": 0.2651909589767456, "step": 15}
+{"stability/response_length_mean": 44.0, "stability/response_length_std": 28.0611572265625, "stability/response_length_var": 787.4285888671875, "stability/token_entropy_mean": 4.462890625, "stability/repetition_rate_mean": 0.2174789160490036, "step": 20}
+{"stability/response_length_mean": 50.25, "stability/response_length_std": 25.549671173095703, "stability/response_length_var": 652.7857055664062, "stability/token_entropy_mean": 4.5078125, "stability/repetition_rate_mean": 0.28109976649284363, "step": 25}
+{"stability/response_length_mean": 48.375, "stability/response_length_std": 28.93311309814453, "stability/response_length_var": 837.125, "stability/token_entropy_mean": 4.595703125, "stability/repetition_rate_mean": 0.158203125, "step": 30}
+{"stability/response_length_mean": 56.875, "stability/response_length_std": 20.152544021606445, "stability/response_length_var": 406.125, "stability/token_entropy_mean": 5.416015625, "stability/repetition_rate_mean": 0.3537946343421936, "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": 4.767578125, "stability/repetition_rate_mean": 0.0994017943739891, "step": 45}
+{"stability/response_length_mean": 55.375, "stability/response_length_std": 16.361431121826172, "stability/response_length_var": 267.6964416503906, "stability/token_entropy_mean": 4.068359375, "stability/repetition_rate_mean": 0.3249044120311737, "step": 50}
+{"stability/response_length_mean": 24.5, "stability/response_length_std": 32.710853576660156, "stability/response_length_var": 1070.0, "stability/token_entropy_mean": 4.6484375, "stability/repetition_rate_mean": 0.1015625, "step": 55}
+{"stability/response_length_mean": 52.625, "stability/response_length_std": 21.75800323486328, "stability/response_length_var": 473.41070556640625, "stability/token_entropy_mean": 5.314453125, "stability/repetition_rate_mean": 0.20815381407737732, "step": 60}
+{"stability/response_length_mean": 17.875, "stability/response_length_std": 23.485177993774414, "stability/response_length_var": 551.5535888671875, "stability/token_entropy_mean": 5.6875, "stability/repetition_rate_mean": 0.20286151766777039, "step": 65}
+{"stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.3349609375, "stability/repetition_rate_mean": 0.119140625, "step": 70}
+{"stability/response_length_mean": 22.75, "stability/response_length_std": 31.07479476928711, "stability/response_length_var": 965.6428833007812, "stability/token_entropy_mean": 2.16015625, "stability/repetition_rate_mean": 0.06745792925357819, "step": 75}
+{"stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.25390625, "stability/repetition_rate_mean": 0.17916667461395264, "step": 80}
+{"stability/response_length_mean": 32.5, "stability/response_length_std": 33.67915725708008, "stability/response_length_var": 1134.2857666015625, "stability/token_entropy_mean": 4.9453125, "stability/repetition_rate_mean": 0.11328125, "step": 85}
+{"stability/response_length_mean": 30.625, "stability/response_length_std": 28.779644012451172, "stability/response_length_var": 828.2678833007812, "stability/token_entropy_mean": 3.736328125, "stability/repetition_rate_mean": 0.3812144994735718, "step": 90}
+{"stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.689453125, "stability/repetition_rate_mean": 0.20546108484268188, "step": 95}
+{"stability/response_length_mean": 40.75, "stability/response_length_std": 31.093177795410156, "stability/response_length_var": 966.7857055664062, "stability/token_entropy_mean": 4.140625, "stability/repetition_rate_mean": 0.24010416865348816, "step": 100}
+{"stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.40625, "stability/repetition_rate_mean": 0.2014508843421936, "step": 105}
+{"stability/response_length_mean": 28.875, "stability/response_length_std": 29.911476135253906, "stability/response_length_var": 894.6964111328125, "stability/token_entropy_mean": 4.3857421875, "stability/repetition_rate_mean": 0.31048768758773804, "step": 110}
+{"stability/response_length_mean": 49.0, "stability/response_length_std": 27.795169830322266, "stability/response_length_var": 772.5714111328125, "stability/token_entropy_mean": 4.830078125, "stability/repetition_rate_mean": 0.3151041865348816, "step": 115}
+{"stability/response_length_mean": 27.875, "stability/response_length_std": 30.31943130493164, "stability/response_length_var": 919.2678833007812, "stability/token_entropy_mean": 4.64453125, "stability/repetition_rate_mean": 0.1723366528749466, "step": 120}
+{"stability/response_length_mean": 15.875, "stability/response_length_std": 24.584766387939453, "stability/response_length_var": 604.4107055664062, "stability/token_entropy_mean": 3.49609375, "stability/repetition_rate_mean": 0.15937499701976776, "step": 125}
+{"stability/response_length_mean": 38.0, "stability/response_length_std": 29.13515281677246, "stability/response_length_var": 848.8571166992188, "stability/token_entropy_mean": 3.91796875, "stability/repetition_rate_mean": 0.22330255806446075, "step": 130}
+{"stability/response_length_mean": 29.875, "stability/response_length_std": 29.502723693847656, "stability/response_length_var": 870.4107055664062, "stability/token_entropy_mean": 3.9013671875, "stability/repetition_rate_mean": 0.2270585298538208, "step": 135}
+{"stability/response_length_mean": 56.25, "stability/response_length_std": 21.920310974121094, "stability/response_length_var": 480.5, "stability/token_entropy_mean": 5.134765625, "stability/repetition_rate_mean": 0.228515625, "step": 140}
+{"stability/response_length_mean": 22.625, "stability/response_length_std": 26.359804153442383, "stability/response_length_var": 694.8392944335938, "stability/token_entropy_mean": 5.6640625, "stability/repetition_rate_mean": 0.21016408503055573, "step": 145}
+{"stability/response_length_mean": 29.375, "stability/response_length_std": 29.193626403808594, "stability/response_length_var": 852.2678833007812, "stability/token_entropy_mean": 4.58984375, "stability/repetition_rate_mean": 0.23066024482250214, "step": 150}
+{"stability/response_length_mean": 58.125, "stability/response_length_std": 10.960155487060547, "stability/response_length_var": 120.125, "stability/token_entropy_mean": 4.5673828125, "stability/repetition_rate_mean": 0.3115246295928955, "step": 155}
+{"stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.484375, "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": 4.87109375, "stability/repetition_rate_mean": 0.2411106526851654, "step": 165}
+{"stability/response_length_mean": 44.125, "stability/response_length_std": 28.43256378173828, "stability/response_length_var": 808.4107055664062, "stability/token_entropy_mean": 3.59228515625, "stability/repetition_rate_mean": 0.3512404263019562, "step": 170}
+{"stability/response_length_mean": 26.875, "stability/response_length_std": 25.570281982421875, "stability/response_length_var": 653.8392944335938, "stability/token_entropy_mean": 4.849609375, "stability/repetition_rate_mean": 0.19270984828472137, "step": 175}
+{"stability/response_length_mean": 35.25, "stability/response_length_std": 31.129911422729492, "stability/response_length_var": 969.0714111328125, "stability/token_entropy_mean": 3.46875, "stability/repetition_rate_mean": 0.2008928656578064, "step": 180}
+{"stability/response_length_mean": 32.875, "stability/response_length_std": 33.30996322631836, "stability/response_length_var": 1109.5535888671875, "stability/token_entropy_mean": 3.705078125, "stability/repetition_rate_mean": 0.19140625, "step": 185}
+{"stability/response_length_mean": 18.625, "stability/response_length_std": 26.52189064025879, "stability/response_length_var": 703.4107055664062, "stability/token_entropy_mean": 4.580078125, "stability/repetition_rate_mean": 0.19138340651988983, "step": 190}
+{"stability/response_length_mean": 49.0, "stability/response_length_std": 27.82085609436035, "stability/response_length_var": 774.0, "stability/token_entropy_mean": 5.1875, "stability/repetition_rate_mean": 0.3934151828289032, "step": 195}

mistral7b_kto_w1_gamma0.001/tokenizer.model

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

mistral7b_kto_w1_gamma0.001/tokenizer_config.json

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

mistral7b_kto_w1_gamma0.001/train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.6270372308030421,
+    "train_runtime": 3152.8646,
+    "train_samples_per_second": 31.717,
+    "train_steps_per_second": 0.062
+}

mistral7b_kto_w1_gamma0.001/trainer_log.jsonl

@@ -0,0 +1,40 @@
+{"current_steps": 5, "total_steps": 196, "loss": 0.5319, "lr": 4e-08, "epoch": 0.025591810620601407, "percentage": 2.55, "elapsed_time": "0:01:18", "remaining_time": "0:49:47", "rewards/chosen": 0.004304667836741397, "logps/chosen": -439.0025699013158, "logits/chosen": -59717214.315789476, "rewards/rejected": 0.0010812577924558095, "logps/rejected": -198.39857700892858, "logits/rejected": -50418761.14285714, "rewards/margins": 0.0032234100442855877, "regularization/w1": 0.03612701967358589, "regularization/lipschitz_norm": 36.127017974853516, "kl": 0.009494942612946033}
+{"current_steps": 10, "total_steps": 196, "loss": 0.5474, "lr": 9e-08, "epoch": 0.05118362124120281, "percentage": 5.1, "elapsed_time": "0:02:39", "remaining_time": "0:49:18", "rewards/chosen": 0.00588285071509225, "logps/chosen": -429.8796672077922, "logits/chosen": -59305279.16883117, "rewards/rejected": -0.0015585110130080257, "logps/rejected": -192.88354198042168, "logits/rejected": -50020561.73493976, "rewards/margins": 0.007441361728100276, "regularization/w1": 0.051889922469854355, "regularization/lipschitz_norm": 51.88991165161133, "kl": 0.017178865149617195, "stability/response_length_mean": 44.25, "stability/response_length_std": 28.13614845275879, "stability/response_length_var": 791.6428833007812, "stability/token_entropy_mean": 3.3544921875, "stability/repetition_rate_mean": 0.3605554699897766}
+{"current_steps": 15, "total_steps": 196, "loss": 0.5528, "lr": 1.4e-07, "epoch": 0.07677543186180422, "percentage": 7.65, "elapsed_time": "0:03:58", "remaining_time": "0:48:02", "rewards/chosen": 0.0006950111100167939, "logps/chosen": -404.12386363636364, "logits/chosen": -56235889.26060606, "rewards/rejected": 0.0027571995412149736, "logps/rejected": -200.66649445564516, "logits/rejected": -51918445.006451614, "rewards/margins": -0.0020621884311981795, "regularization/w1": 0.047160327434539795, "regularization/lipschitz_norm": 47.16032409667969, "kl": 0.013535099104046822, "stability/response_length_mean": 27.125, "stability/response_length_std": 30.79163932800293, "stability/response_length_var": 948.125, "stability/token_entropy_mean": 3.8203125, "stability/repetition_rate_mean": 0.2651909589767456}
+{"current_steps": 20, "total_steps": 196, "loss": 0.5531, "lr": 1.8999999999999998e-07, "epoch": 0.10236724248240563, "percentage": 10.2, "elapsed_time": "0:05:20", "remaining_time": "0:46:59", "rewards/chosen": 0.0012291071005165577, "logps/chosen": -343.684423828125, "logits/chosen": -51655552.0, "rewards/rejected": 0.0018676770851016045, "logps/rejected": -225.644677734375, "logits/rejected": -48047257.6, "rewards/margins": -0.0006385699845850468, "regularization/w1": 0.05132513865828514, "regularization/lipschitz_norm": 51.32513427734375, "kl": 0.014954889193177223, "stability/response_length_mean": 44.0, "stability/response_length_std": 28.0611572265625, "stability/response_length_var": 787.4285888671875, "stability/token_entropy_mean": 4.462890625, "stability/repetition_rate_mean": 0.2174789160490036}
+{"current_steps": 25, "total_steps": 196, "loss": 0.5539, "lr": 2.4e-07, "epoch": 0.12795905310300704, "percentage": 12.76, "elapsed_time": "0:06:40", "remaining_time": "0:45:38", "rewards/chosen": 0.002159434351427802, "logps/chosen": -378.46988146551723, "logits/chosen": -56482780.68965517, "rewards/rejected": 0.0010998719930648804, "logps/rejected": -202.78185267857143, "logits/rejected": -52020662.85714286, "rewards/margins": 0.0010595623583629214, "regularization/w1": 0.04689044505357742, "regularization/lipschitz_norm": 46.89044189453125, "kl": 0.019569864496588707, "stability/response_length_mean": 50.25, "stability/response_length_std": 25.549671173095703, "stability/response_length_var": 652.7857055664062, "stability/token_entropy_mean": 4.5078125, "stability/repetition_rate_mean": 0.28109976649284363}
+{"current_steps": 30, "total_steps": 196, "loss": 0.5527, "lr": 2.9e-07, "epoch": 0.15355086372360843, "percentage": 15.31, "elapsed_time": "0:08:00", "remaining_time": "0:44:18", "rewards/chosen": -0.003406525858296644, "logps/chosen": -468.6152999161074, "logits/chosen": -57732102.87248322, "rewards/rejected": 0.0015189916069744624, "logps/rejected": -198.2326845760234, "logits/rejected": -51518679.578947365, "rewards/margins": -0.004925517465271106, "regularization/w1": 0.05276590213179588, "regularization/lipschitz_norm": 52.765899658203125, "kl": 0.018669696524739265, "stability/response_length_mean": 48.375, "stability/response_length_std": 28.93311309814453, "stability/response_length_var": 837.125, "stability/token_entropy_mean": 4.595703125, "stability/repetition_rate_mean": 0.158203125}
+{"current_steps": 35, "total_steps": 196, "loss": 0.5528, "lr": 3.4000000000000003e-07, "epoch": 0.17914267434420986, "percentage": 17.86, "elapsed_time": "0:09:21", "remaining_time": "0:43:04", "rewards/chosen": 0.003288782095607323, "logps/chosen": -385.9036293512658, "logits/chosen": -57640635.94936709, "rewards/rejected": 0.0027985008042535663, "logps/rejected": -195.09850019290124, "logits/rejected": -53904560.98765432, "rewards/margins": 0.0004902812913537566, "regularization/w1": 0.0461605004966259, "regularization/lipschitz_norm": 46.16049575805664, "kl": 0.03436283394694328, "stability/response_length_mean": 56.875, "stability/response_length_std": 20.152544021606445, "stability/response_length_var": 406.125, "stability/token_entropy_mean": 5.416015625, "stability/repetition_rate_mean": 0.3537946343421936}
+{"current_steps": 40, "total_steps": 196, "loss": 0.5509, "lr": 3.8999999999999997e-07, "epoch": 0.20473448496481125, "percentage": 20.41, "elapsed_time": "0:10:42", "remaining_time": "0:41:45", "rewards/chosen": 0.0011051620356738567, "logps/chosen": -421.8916015625, "logits/chosen": -56632812.8, "rewards/rejected": 0.0027551772072911263, "logps/rejected": -219.2817138671875, "logits/rejected": -50224707.2, "rewards/margins": -0.0016500151716172696, "regularization/w1": 0.05284709855914116, "regularization/lipschitz_norm": 52.84709548950195, "kl": 0.027318641543388367, "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": 0.5532, "lr": 4.3999999999999997e-07, "epoch": 0.23032629558541268, "percentage": 22.96, "elapsed_time": "0:12:03", "remaining_time": "0:40:26", "rewards/chosen": 0.002814074090663219, "logps/chosen": -352.3003355704698, "logits/chosen": -52704039.51677852, "rewards/rejected": 0.0029790101692690487, "logps/rejected": -243.84438961988303, "logits/rejected": -50776093.94152047, "rewards/margins": -0.00016493607860582982, "regularization/w1": 0.05639222264289856, "regularization/lipschitz_norm": 56.39221954345703, "kl": 0.04743483290076256, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.87324333190918, "stability/response_length_var": 892.4107055664062, "stability/token_entropy_mean": 4.767578125, "stability/repetition_rate_mean": 0.0994017943739891}
+{"current_steps": 50, "total_steps": 196, "loss": 0.5494, "lr": 4.9e-07, "epoch": 0.2559181062060141, "percentage": 25.51, "elapsed_time": "0:13:23", "remaining_time": "0:39:05", "rewards/chosen": 0.006553848717836726, "logps/chosen": -375.13357277684565, "logits/chosen": -54039232.4295302, "rewards/rejected": 0.003993244896158141, "logps/rejected": -180.2006807383041, "logits/rejected": -48303834.57309941, "rewards/margins": 0.0025606038216785853, "regularization/w1": 0.04168223217129707, "regularization/lipschitz_norm": 41.682228088378906, "kl": 0.051512785255908966, "stability/response_length_mean": 55.375, "stability/response_length_std": 16.361431121826172, "stability/response_length_var": 267.6964416503906, "stability/token_entropy_mean": 4.068359375, "stability/repetition_rate_mean": 0.3249044120311737}
+{"current_steps": 55, "total_steps": 196, "loss": 0.5531, "lr": 5.4e-07, "epoch": 0.28150991682661547, "percentage": 28.06, "elapsed_time": "0:14:43", "remaining_time": "0:37:43", "rewards/chosen": 0.0043661816519980105, "logps/chosen": -344.4683618012422, "logits/chosen": -56948316.22360248, "rewards/rejected": 0.002374995804432803, "logps/rejected": -230.42428262578616, "logits/rejected": -49577713.50943396, "rewards/margins": 0.0019911858475652076, "regularization/w1": 0.04550762102007866, "regularization/lipschitz_norm": 45.507625579833984, "kl": 0.05490356683731079, "stability/response_length_mean": 24.5, "stability/response_length_std": 32.710853576660156, "stability/response_length_var": 1070.0, "stability/token_entropy_mean": 4.6484375, "stability/repetition_rate_mean": 0.1015625}
+{"current_steps": 60, "total_steps": 196, "loss": 0.5496, "lr": 5.9e-07, "epoch": 0.30710172744721687, "percentage": 30.61, "elapsed_time": "0:16:04", "remaining_time": "0:36:26", "rewards/chosen": 0.010953712463378906, "logps/chosen": -359.61126893939394, "logits/chosen": -52669911.660606064, "rewards/rejected": 0.007483173954871393, "logps/rejected": -178.15209173387098, "logits/rejected": -47820423.43225806, "rewards/margins": 0.0034705385085075126, "regularization/w1": 0.046306367963552475, "regularization/lipschitz_norm": 46.30636215209961, "kl": 0.05875139310956001, "stability/response_length_mean": 52.625, "stability/response_length_std": 21.75800323486328, "stability/response_length_var": 473.41070556640625, "stability/token_entropy_mean": 5.314453125, "stability/repetition_rate_mean": 0.20815381407737732}
+{"current_steps": 65, "total_steps": 196, "loss": 0.5544, "lr": 6.4e-07, "epoch": 0.3326935380678183, "percentage": 33.16, "elapsed_time": "0:17:25", "remaining_time": "0:35:06", "rewards/chosen": 0.012180716760696904, "logps/chosen": -381.825, "logits/chosen": -57561999.69032258, "rewards/rejected": 0.007880073605161724, "logps/rejected": -183.50355113636363, "logits/rejected": -47265897.5030303, "rewards/margins": 0.00430064315553518, "regularization/w1": 0.053168147802352905, "regularization/lipschitz_norm": 53.16814041137695, "kl": 0.08697053045034409, "stability/response_length_mean": 17.875, "stability/response_length_std": 23.485177993774414, "stability/response_length_var": 551.5535888671875, "stability/token_entropy_mean": 5.6875, "stability/repetition_rate_mean": 0.20286151766777039}
+{"current_steps": 70, "total_steps": 196, "loss": 0.5558, "lr": 6.9e-07, "epoch": 0.3582853486884197, "percentage": 35.71, "elapsed_time": "0:18:45", "remaining_time": "0:33:45", "rewards/chosen": 0.012132589278682585, "logps/chosen": -379.0071572580645, "logits/chosen": -52762746.21935484, "rewards/rejected": 0.007109479470686479, "logps/rejected": -220.3510890151515, "logits/rejected": -49031695.515151516, "rewards/margins": 0.005023109807996106, "regularization/w1": 0.05391382798552513, "regularization/lipschitz_norm": 53.91382598876953, "kl": 0.09267856180667877, "stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.3349609375, "stability/repetition_rate_mean": 0.119140625}
+{"current_steps": 75, "total_steps": 196, "loss": 0.5574, "lr": 7.4e-07, "epoch": 0.3838771593090211, "percentage": 38.27, "elapsed_time": "0:20:06", "remaining_time": "0:32:26", "rewards/chosen": 0.013364916717683946, "logps/chosen": -314.1371674408784, "logits/chosen": -49496755.89189189, "rewards/rejected": 0.007764597964841266, "logps/rejected": -228.80718568313952, "logits/rejected": -47891048.18604651, "rewards/margins": 0.0056003187528426805, "regularization/w1": 0.05378229543566704, "regularization/lipschitz_norm": 53.78229904174805, "kl": 0.10173000395298004, "stability/response_length_mean": 22.75, "stability/response_length_std": 31.07479476928711, "stability/response_length_var": 965.6428833007812, "stability/token_entropy_mean": 2.16015625, "stability/repetition_rate_mean": 0.06745792925357819}
+{"current_steps": 80, "total_steps": 196, "loss": 0.5561, "lr": 7.9e-07, "epoch": 0.4094689699296225, "percentage": 40.82, "elapsed_time": "0:21:27", "remaining_time": "0:31:06", "rewards/chosen": 0.01607713478290482, "logps/chosen": -403.86936051324506, "logits/chosen": -58616967.62913907, "rewards/rejected": 0.00929838524767633, "logps/rejected": -200.9628328402367, "logits/rejected": -53431344.47337278, "rewards/margins": 0.006778749535228491, "regularization/w1": 0.047661323100328445, "regularization/lipschitz_norm": 47.66132354736328, "kl": 0.13186340034008026, "stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.25390625, "stability/repetition_rate_mean": 0.17916667461395264}
+{"current_steps": 85, "total_steps": 196, "loss": 0.5594, "lr": 8.399999999999999e-07, "epoch": 0.4350607805502239, "percentage": 43.37, "elapsed_time": "0:22:47", "remaining_time": "0:29:46", "rewards/chosen": 0.020604954996416645, "logps/chosen": -358.1264616935484, "logits/chosen": -51464935.22580645, "rewards/rejected": 0.00958243716846813, "logps/rejected": -230.3289772727273, "logits/rejected": -47132777.5030303, "rewards/margins": 0.011022517827948516, "regularization/w1": 0.055119942873716354, "regularization/lipschitz_norm": 55.11994552612305, "kl": 0.1272238940000534, "stability/response_length_mean": 32.5, "stability/response_length_std": 33.67915725708008, "stability/response_length_var": 1134.2857666015625, "stability/token_entropy_mean": 4.9453125, "stability/repetition_rate_mean": 0.11328125}
+{"current_steps": 90, "total_steps": 196, "loss": 0.5657, "lr": 8.9e-07, "epoch": 0.46065259117082535, "percentage": 45.92, "elapsed_time": "0:24:08", "remaining_time": "0:28:26", "rewards/chosen": 0.026595650575099848, "logps/chosen": -380.5234124599359, "logits/chosen": -58414500.102564104, "rewards/rejected": 0.009931363710543005, "logps/rejected": -187.12747713414635, "logits/rejected": -50554680.19512195, "rewards/margins": 0.016664286864556843, "regularization/w1": 0.06698118895292282, "regularization/lipschitz_norm": 66.9811782836914, "kl": 0.1703931838274002, "stability/response_length_mean": 30.625, "stability/response_length_std": 28.779644012451172, "stability/response_length_var": 828.2678833007812, "stability/token_entropy_mean": 3.736328125, "stability/repetition_rate_mean": 0.3812144994735718}
+{"current_steps": 95, "total_steps": 196, "loss": 0.5682, "lr": 9.399999999999999e-07, "epoch": 0.48624440179142675, "percentage": 48.47, "elapsed_time": "0:25:28", "remaining_time": "0:27:05", "rewards/chosen": 0.026378269436993177, "logps/chosen": -314.7296034414557, "logits/chosen": -50386140.35443038, "rewards/rejected": 0.005254786691547912, "logps/rejected": -207.09116994598764, "logits/rejected": -48979835.25925926, "rewards/margins": 0.021123482745445267, "regularization/w1": 0.06674359738826752, "regularization/lipschitz_norm": 66.74359130859375, "kl": 0.16741442680358887, "stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.689453125, "stability/repetition_rate_mean": 0.20546108484268188}
+{"current_steps": 100, "total_steps": 196, "loss": 0.5714, "lr": 9.9e-07, "epoch": 0.5118362124120281, "percentage": 51.02, "elapsed_time": "0:26:49", "remaining_time": "0:25:45", "rewards/chosen": 0.030437920544598554, "logps/chosen": -359.56579919763516, "logits/chosen": -54716139.24324324, "rewards/rejected": 0.018727043340372486, "logps/rejected": -216.39162427325581, "logits/rejected": -50905552.37209302, "rewards/margins": 0.011710877204226068, "regularization/w1": 0.0671883150935173, "regularization/lipschitz_norm": 67.18830871582031, "kl": 0.20850256085395813, "stability/response_length_mean": 40.75, "stability/response_length_std": 31.093177795410156, "stability/response_length_var": 966.7857055664062, "stability/token_entropy_mean": 4.140625, "stability/repetition_rate_mean": 0.24010416865348816}
+{"current_steps": 105, "total_steps": 196, "loss": 0.5793, "lr": 9.999815554351224e-07, "epoch": 0.5374280230326296, "percentage": 53.57, "elapsed_time": "0:28:10", "remaining_time": "0:24:24", "rewards/chosen": 0.03818985353033227, "logps/chosen": -399.5205666415663, "logits/chosen": -59376942.26506024, "rewards/rejected": 0.016095796188750823, "logps/rejected": -209.8428622159091, "logits/rejected": -53793147.01298701, "rewards/margins": 0.022094057341581444, "regularization/w1": 0.09085484594106674, "regularization/lipschitz_norm": 90.85485076904297, "kl": 0.2234809845685959, "stability/response_length_mean": 29.0, "stability/response_length_std": 29.89505386352539, "stability/response_length_var": 893.7142944335938, "stability/token_entropy_mean": 3.40625, "stability/repetition_rate_mean": 0.2014508843421936}
+{"current_steps": 110, "total_steps": 196, "loss": 0.5834, "lr": 9.999066267225444e-07, "epoch": 0.5630198336532309, "percentage": 56.12, "elapsed_time": "0:29:31", "remaining_time": "0:23:05", "rewards/chosen": 0.036506069711892, "logps/chosen": -378.19820689006025, "logits/chosen": -57391949.10843374, "rewards/rejected": 0.01970642108421821, "logps/rejected": -205.89108157467533, "logits/rejected": -53613863.896103896, "rewards/margins": 0.01679964862767379, "regularization/w1": 0.08991190046072006, "regularization/lipschitz_norm": 89.9118881225586, "kl": 0.26629599928855896, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.911476135253906, "stability/response_length_var": 894.6964111328125, "stability/token_entropy_mean": 4.3857421875, "stability/repetition_rate_mean": 0.31048768758773804}
+{"current_steps": 115, "total_steps": 196, "loss": 0.5942, "lr": 9.997740697079592e-07, "epoch": 0.5886116442738324, "percentage": 58.67, "elapsed_time": "0:30:53", "remaining_time": "0:21:45", "rewards/chosen": 0.052421219741241844, "logps/chosen": -364.06833465189874, "logits/chosen": -60606263.088607594, "rewards/rejected": 0.02105676097634398, "logps/rejected": -200.9859905478395, "logits/rejected": -53428704.39506173, "rewards/margins": 0.031364458764897865, "regularization/w1": 0.10365422070026398, "regularization/lipschitz_norm": 103.65421295166016, "kl": 0.32535991072654724, "stability/response_length_mean": 49.0, "stability/response_length_std": 27.795169830322266, "stability/response_length_var": 772.5714111328125, "stability/token_entropy_mean": 4.830078125, "stability/repetition_rate_mean": 0.3151041865348816}
+{"current_steps": 120, "total_steps": 196, "loss": 0.6004, "lr": 9.995838996722914e-07, "epoch": 0.6142034548944337, "percentage": 61.22, "elapsed_time": "0:32:12", "remaining_time": "0:20:23", "rewards/chosen": 0.05425496818193423, "logps/chosen": -383.7023845996732, "logits/chosen": -57748794.5620915, "rewards/rejected": 0.026231080472112416, "logps/rejected": -211.95403723802394, "logits/rejected": -54355410.01197605, "rewards/margins": 0.028023887709821817, "regularization/w1": 0.10987582057714462, "regularization/lipschitz_norm": 109.87580871582031, "kl": 0.3553595244884491, "stability/response_length_mean": 27.875, "stability/response_length_std": 30.31943130493164, "stability/response_length_var": 919.2678833007812, "stability/token_entropy_mean": 4.64453125, "stability/repetition_rate_mean": 0.1723366528749466}
+{"current_steps": 125, "total_steps": 196, "loss": 0.6157, "lr": 9.993361385379874e-07, "epoch": 0.6397952655150352, "percentage": 63.78, "elapsed_time": "0:33:33", "remaining_time": "0:19:03", "rewards/chosen": 0.05468781353676156, "logps/chosen": -416.8324058219178, "logits/chosen": -60685662.68493151, "rewards/rejected": 0.025684822564837576, "logps/rejected": -216.1103627873563, "logits/rejected": -55654294.06896552, "rewards/margins": 0.029002990971923985, "regularization/w1": 0.12617388367652893, "regularization/lipschitz_norm": 126.17388153076172, "kl": 0.39908552169799805, "stability/response_length_mean": 15.875, "stability/response_length_std": 24.584766387939453, "stability/response_length_var": 604.4107055664062, "stability/token_entropy_mean": 3.49609375, "stability/repetition_rate_mean": 0.15937499701976776}
+{"current_steps": 130, "total_steps": 196, "loss": 0.6243, "lr": 9.99030814866488e-07, "epoch": 0.6653870761356366, "percentage": 66.33, "elapsed_time": "0:34:53", "remaining_time": "0:17:42", "rewards/chosen": 0.06732863149335308, "logps/chosen": -391.98465221774194, "logits/chosen": -57391097.393548384, "rewards/rejected": 0.02722955183549361, "logps/rejected": -183.64661458333333, "logits/rejected": -53830773.915151514, "rewards/margins": 0.04009907965785947, "regularization/w1": 0.1440272331237793, "regularization/lipschitz_norm": 144.0272216796875, "kl": 0.44654884934425354, "stability/response_length_mean": 38.0, "stability/response_length_std": 29.13515281677246, "stability/response_length_var": 848.8571166992188, "stability/token_entropy_mean": 3.91796875, "stability/repetition_rate_mean": 0.22330255806446075}
+{"current_steps": 135, "total_steps": 196, "loss": 0.6335, "lr": 9.986679638549358e-07, "epoch": 0.690978886756238, "percentage": 68.88, "elapsed_time": "0:36:13", "remaining_time": "0:16:22", "rewards/chosen": 0.07913663197148797, "logps/chosen": -401.0247795245399, "logits/chosen": -59841391.50920245, "rewards/rejected": 0.03488306179168118, "logps/rejected": -247.52191978503186, "logits/rejected": -55011230.1656051, "rewards/margins": 0.044253570179806793, "regularization/w1": 0.167570099234581, "regularization/lipschitz_norm": 167.570068359375, "kl": 0.4990025460720062, "stability/response_length_mean": 29.875, "stability/response_length_std": 29.502723693847656, "stability/response_length_var": 870.4107055664062, "stability/token_entropy_mean": 3.9013671875, "stability/repetition_rate_mean": 0.2270585298538208}
+{"current_steps": 140, "total_steps": 196, "loss": 0.6499, "lr": 9.982476273321175e-07, "epoch": 0.7165706973768394, "percentage": 71.43, "elapsed_time": "0:37:34", "remaining_time": "0:15:01", "rewards/chosen": 0.08244560126102332, "logps/chosen": -366.36526988636365, "logits/chosen": -57072553.11515152, "rewards/rejected": 0.012183857733203518, "logps/rejected": -189.91263860887096, "logits/rejected": -50065289.08387097, "rewards/margins": 0.0702617435278198, "regularization/w1": 0.18331506848335266, "regularization/lipschitz_norm": 183.31504821777344, "kl": 0.4819028973579407, "stability/response_length_mean": 56.25, "stability/response_length_std": 21.920310974121094, "stability/response_length_var": 480.5, "stability/token_entropy_mean": 5.134765625, "stability/repetition_rate_mean": 0.228515625}
+{"current_steps": 145, "total_steps": 196, "loss": 0.6604, "lr": 9.977698537536417e-07, "epoch": 0.7421625079974408, "percentage": 73.98, "elapsed_time": "0:38:54", "remaining_time": "0:13:40", "rewards/chosen": 0.08324084224471126, "logps/chosen": -329.439406061747, "logits/chosen": -51148250.98795181, "rewards/rejected": 0.03314851166365983, "logps/rejected": -199.1564275568182, "logits/rejected": -51665840.20779221, "rewards/margins": 0.050092330581051434, "regularization/w1": 0.1634482592344284, "regularization/lipschitz_norm": 163.44825744628906, "kl": 0.5741755366325378, "stability/response_length_mean": 22.625, "stability/response_length_std": 26.359804153442383, "stability/response_length_var": 694.8392944335938, "stability/token_entropy_mean": 5.6640625, "stability/repetition_rate_mean": 0.21016408503055573}
+{"current_steps": 150, "total_steps": 196, "loss": 0.6752, "lr": 9.972346981963546e-07, "epoch": 0.7677543186180422, "percentage": 76.53, "elapsed_time": "0:40:15", "remaining_time": "0:12:20", "rewards/chosen": 0.11221859875847312, "logps/chosen": -396.8122242647059, "logits/chosen": -55426325.08235294, "rewards/rejected": 0.03900526364644368, "logps/rejected": -207.33588541666666, "logits/rejected": -50193100.8, "rewards/margins": 0.07321333511202943, "regularization/w1": 0.1853603571653366, "regularization/lipschitz_norm": 185.36038208007812, "kl": 0.6443031430244446, "stability/response_length_mean": 29.375, "stability/response_length_std": 29.193626403808594, "stability/response_length_var": 852.2678833007812, "stability/token_entropy_mean": 4.58984375, "stability/repetition_rate_mean": 0.23066024482250214}
+{"current_steps": 155, "total_steps": 196, "loss": 0.7031, "lr": 9.966422223519885e-07, "epoch": 0.7933461292386437, "percentage": 79.08, "elapsed_time": "0:41:36", "remaining_time": "0:11:00", "rewards/chosen": 0.09209441016702091, "logps/chosen": -360.06833639705883, "logits/chosen": -54025167.8117647, "rewards/rejected": 0.026456499099731447, "logps/rejected": -191.04104166666667, "logits/rejected": -49363421.86666667, "rewards/margins": 0.06563791106728947, "regularization/w1": 0.2233385592699051, "regularization/lipschitz_norm": 223.3385467529297, "kl": 0.6507260799407959, "stability/response_length_mean": 58.125, "stability/response_length_std": 10.960155487060547, "stability/response_length_var": 120.125, "stability/token_entropy_mean": 4.5673828125, "stability/repetition_rate_mean": 0.3115246295928955}
+{"current_steps": 160, "total_steps": 196, "loss": 0.7229, "lr": 9.959924945200523e-07, "epoch": 0.818937939859245, "percentage": 81.63, "elapsed_time": "0:42:57", "remaining_time": "0:09:39", "rewards/chosen": 0.13677164486476354, "logps/chosen": -415.52762276785717, "logits/chosen": -55248286.47619048, "rewards/rejected": 0.041667740595968145, "logps/rejected": -186.48067434210526, "logits/rejected": -48961236.21052632, "rewards/margins": 0.09510390426879539, "regularization/w1": 0.23188279569149017, "regularization/lipschitz_norm": 231.88279724121094, "kl": 0.7176083326339722, "stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.484375, "stability/repetition_rate_mean": 0.30851560831069946}
+{"current_steps": 165, "total_steps": 196, "loss": 0.7436, "lr": 9.952855895999568e-07, "epoch": 0.8445297504798465, "percentage": 84.18, "elapsed_time": "0:44:17", "remaining_time": "0:08:19", "rewards/chosen": 0.12150281051109577, "logps/chosen": -369.70171066810343, "logits/chosen": -54404449.10344828, "rewards/rejected": 0.047391715115063814, "logps/rejected": -212.1182041952055, "logits/rejected": -49778060.2739726, "rewards/margins": 0.07411109539603195, "regularization/w1": 0.22178728878498077, "regularization/lipschitz_norm": 221.78726196289062, "kl": 0.770777702331543, "stability/response_length_mean": 22.5, "stability/response_length_std": 28.33977508544922, "stability/response_length_var": 803.1428833007812, "stability/token_entropy_mean": 4.87109375, "stability/repetition_rate_mean": 0.2411106526851654}
+{"current_steps": 170, "total_steps": 196, "loss": 0.7621, "lr": 9.9452158908238e-07, "epoch": 0.8701215611004478, "percentage": 86.73, "elapsed_time": "0:45:38", "remaining_time": "0:06:58", "rewards/chosen": 0.16493017433663093, "logps/chosen": -439.69253883136093, "logits/chosen": -59132721.988165684, "rewards/rejected": 0.06748862613905345, "logps/rejected": -190.58294960678808, "logits/rejected": -52588998.357615896, "rewards/margins": 0.09744154819757748, "regularization/w1": 0.27300870418548584, "regularization/lipschitz_norm": 273.0086975097656, "kl": 0.9223539233207703, "stability/response_length_mean": 44.125, "stability/response_length_std": 28.43256378173828, "stability/response_length_var": 808.4107055664062, "stability/token_entropy_mean": 3.59228515625, "stability/repetition_rate_mean": 0.3512404263019562}
+{"current_steps": 175, "total_steps": 196, "loss": 0.7733, "lr": 9.937005810398745e-07, "epoch": 0.8957133717210493, "percentage": 89.29, "elapsed_time": "0:46:59", "remaining_time": "0:05:38", "rewards/chosen": 0.15406293754118033, "logps/chosen": -354.8965549698795, "logits/chosen": -56663225.06024096, "rewards/rejected": 0.050444962142349836, "logps/rejected": -194.3030768060065, "logits/rejected": -48029942.02597403, "rewards/margins": 0.1036179753988305, "regularization/w1": 0.3229624629020691, "regularization/lipschitz_norm": 322.96240234375, "kl": 0.9431293606758118, "stability/response_length_mean": 26.875, "stability/response_length_std": 25.570281982421875, "stability/response_length_var": 653.8392944335938, "stability/token_entropy_mean": 4.849609375, "stability/repetition_rate_mean": 0.19270984828472137}
+{"current_steps": 180, "total_steps": 196, "loss": 0.8001, "lr": 9.928226601167138e-07, "epoch": 0.9213051823416507, "percentage": 91.84, "elapsed_time": "0:48:19", "remaining_time": "0:04:17", "rewards/chosen": 0.1740191524014151, "logps/chosen": -374.4056269171779, "logits/chosen": -52662171.48466258, "rewards/rejected": 0.0745979722138423, "logps/rejected": -210.5522243232484, "logits/rejected": -47984167.133757964, "rewards/margins": 0.0994211801875728, "regularization/w1": 0.3154919147491455, "regularization/lipschitz_norm": 315.4919128417969, "kl": 1.0352585315704346, "stability/response_length_mean": 35.25, "stability/response_length_std": 31.129911422729492, "stability/response_length_var": 969.0714111328125, "stability/token_entropy_mean": 3.46875, "stability/repetition_rate_mean": 0.2008928656578064}
+{"current_steps": 185, "total_steps": 196, "loss": 0.8358, "lr": 9.918879275179817e-07, "epoch": 0.946896992962252, "percentage": 94.39, "elapsed_time": "0:49:39", "remaining_time": "0:02:57", "rewards/chosen": 0.16936941500063296, "logps/chosen": -310.3076774691358, "logits/chosen": -54614294.12345679, "rewards/rejected": 0.05014253266250031, "logps/rejected": -196.53772745253164, "logits/rejected": -49507992.30379747, "rewards/margins": 0.11922688233813265, "regularization/w1": 0.3928832709789276, "regularization/lipschitz_norm": 392.8832702636719, "kl": 1.078233003616333, "stability/response_length_mean": 32.875, "stability/response_length_std": 33.30996322631836, "stability/response_length_var": 1109.5535888671875, "stability/token_entropy_mean": 3.705078125, "stability/repetition_rate_mean": 0.19140625}
+{"current_steps": 190, "total_steps": 196, "loss": 0.8981, "lr": 9.90896490997906e-07, "epoch": 0.9724888035828535, "percentage": 96.94, "elapsed_time": "0:50:59", "remaining_time": "0:01:36", "rewards/chosen": 0.21182136764069517, "logps/chosen": -390.28709767964074, "logits/chosen": -53080229.556886226, "rewards/rejected": 0.07398576674118541, "logps/rejected": -210.45874183006535, "logits/rejected": -49681528.47058824, "rewards/margins": 0.13783560089950975, "regularization/w1": 0.4389554560184479, "regularization/lipschitz_norm": 438.9554748535156, "kl": 1.122229814529419, "stability/response_length_mean": 18.625, "stability/response_length_std": 26.52189064025879, "stability/response_length_var": 703.4107055664062, "stability/token_entropy_mean": 4.580078125, "stability/repetition_rate_mean": 0.19138340651988983}
+{"current_steps": 195, "total_steps": 196, "loss": 0.8839, "lr": 9.898484648474362e-07, "epoch": 0.9980806142034548, "percentage": 99.49, "elapsed_time": "0:52:19", "remaining_time": "0:00:16", "rewards/chosen": 0.19376148100822202, "logps/chosen": -390.84732862903223, "logits/chosen": -60036446.14193548, "rewards/rejected": 0.08309277043198095, "logps/rejected": -197.08839962121212, "logits/rejected": -53804826.375757575, "rewards/margins": 0.11066871057624107, "regularization/w1": 0.37205204367637634, "regularization/lipschitz_norm": 372.052001953125, "kl": 1.1505008935928345, "stability/response_length_mean": 49.0, "stability/response_length_std": 27.82085609436035, "stability/response_length_var": 774.0, "stability/token_entropy_mean": 5.1875, "stability/repetition_rate_mean": 0.3934151828289032}
+{"current_steps": 196, "total_steps": 196, "epoch": 1.0, "percentage": 100.0, "elapsed_time": "0:52:32", "remaining_time": "0:00:00"}

mistral7b_kto_w1_gamma0.001/trainer_state.json

@@ -0,0 +1,974 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 196,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.025591810620601407,
+      "grad_norm": 0.5678192377090454,
+      "kl": 0.009494942612946033,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59717214.315789476,
+      "logits/rejected": -50418761.14285714,
+      "logps/chosen": -439.0025699013158,
+      "logps/rejected": -198.39857700892858,
+      "loss": 0.5319,
+      "loss/base_kto": 0.49962836503982544,
+      "metrics/advantage_var": 0.04574162885546684,
+      "regularization/lipschitz_norm": 36.127017974853516,
+      "regularization/w1": 0.03612701967358589,
+      "rewards/chosen": 0.004304667836741397,
+      "rewards/margins": 0.0032234100442855877,
+      "rewards/rejected": 0.0010812577924558095,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.6058618426322937,
+      "kl": 0.017178865149617195,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59305279.16883117,
+      "logits/rejected": -50020561.73493976,
+      "logps/chosen": -429.8796672077922,
+      "logps/rejected": -192.88354198042168,
+      "loss": 0.5474,
+      "loss/base_kto": 0.4991108477115631,
+      "metrics/advantage_var": 0.07220327109098434,
+      "regularization/lipschitz_norm": 51.88991165161133,
+      "regularization/w1": 0.051889922469854355,
+      "rewards/chosen": 0.00588285071509225,
+      "rewards/margins": 0.007441361728100276,
+      "rewards/rejected": -0.0015585110130080257,
+      "stability/repetition_rate_mean": 0.3605554699897766,
+      "stability/response_length_mean": 44.25,
+      "stability/response_length_std": 28.13614845275879,
+      "stability/response_length_var": 791.6428833007812,
+      "stability/token_entropy_mean": 3.3544921875,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.6088977456092834,
+      "kl": 0.013535099104046822,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56235889.26060606,
+      "logits/rejected": -51918445.006451614,
+      "logps/chosen": -404.12386363636364,
+      "logps/rejected": -200.66649445564516,
+      "loss": 0.5528,
+      "loss/base_kto": 0.5002279281616211,
+      "metrics/advantage_var": 0.057587385177612305,
+      "regularization/lipschitz_norm": 47.16032409667969,
+      "regularization/w1": 0.047160327434539795,
+      "rewards/chosen": 0.0006950111100167939,
+      "rewards/margins": -0.0020621884311981795,
+      "rewards/rejected": 0.0027571995412149736,
+      "stability/repetition_rate_mean": 0.2651909589767456,
+      "stability/response_length_mean": 27.125,
+      "stability/response_length_std": 30.79163932800293,
+      "stability/response_length_var": 948.125,
+      "stability/token_entropy_mean": 3.8203125,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.6186808347702026,
+      "kl": 0.014954889193177223,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51655552.0,
+      "logits/rejected": -48047257.6,
+      "logps/chosen": -343.684423828125,
+      "logps/rejected": -225.644677734375,
+      "loss": 0.5531,
+      "loss/base_kto": 0.500113308429718,
+      "metrics/advantage_var": 0.0725272074341774,
+      "regularization/lipschitz_norm": 51.32513427734375,
+      "regularization/w1": 0.05132513865828514,
+      "rewards/chosen": 0.0012291071005165577,
+      "rewards/margins": -0.0006385699845850468,
+      "rewards/rejected": 0.0018676770851016045,
+      "stability/repetition_rate_mean": 0.2174789160490036,
+      "stability/response_length_mean": 44.0,
+      "stability/response_length_std": 28.0611572265625,
+      "stability/response_length_var": 787.4285888671875,
+      "stability/token_entropy_mean": 4.462890625,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.5771578550338745,
+      "kl": 0.019569864496588707,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56482780.68965517,
+      "logits/rejected": -52020662.85714286,
+      "logps/chosen": -378.46988146551723,
+      "logps/rejected": -202.78185267857143,
+      "loss": 0.5539,
+      "loss/base_kto": 0.49986138939857483,
+      "metrics/advantage_var": 0.05885173752903938,
+      "regularization/lipschitz_norm": 46.89044189453125,
+      "regularization/w1": 0.04689044505357742,
+      "rewards/chosen": 0.002159434351427802,
+      "rewards/margins": 0.0010595623583629214,
+      "rewards/rejected": 0.0010998719930648804,
+      "stability/repetition_rate_mean": 0.28109976649284363,
+      "stability/response_length_mean": 50.25,
+      "stability/response_length_std": 25.549671173095703,
+      "stability/response_length_var": 652.7857055664062,
+      "stability/token_entropy_mean": 4.5078125,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.603465735912323,
+      "kl": 0.018669696524739265,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57732102.87248322,
+      "logits/rejected": -51518679.578947365,
+      "logps/chosen": -468.6152999161074,
+      "logps/rejected": -198.2326845760234,
+      "loss": 0.5527,
+      "loss/base_kto": 0.5005181431770325,
+      "metrics/advantage_var": 0.07389438152313232,
+      "regularization/lipschitz_norm": 52.765899658203125,
+      "regularization/w1": 0.05276590213179588,
+      "rewards/chosen": -0.003406525858296644,
+      "rewards/margins": -0.004925517465271106,
+      "rewards/rejected": 0.0015189916069744624,
+      "stability/repetition_rate_mean": 0.158203125,
+      "stability/response_length_mean": 48.375,
+      "stability/response_length_std": 28.93311309814453,
+      "stability/response_length_var": 837.125,
+      "stability/token_entropy_mean": 4.595703125,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6781691908836365,
+      "kl": 0.03436283394694328,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57640635.94936709,
+      "logits/rejected": -53904560.98765432,
+      "logps/chosen": -385.9036293512658,
+      "logps/rejected": -195.09850019290124,
+      "loss": 0.5528,
+      "loss/base_kto": 0.4999621510505676,
+      "metrics/advantage_var": 0.05386241897940636,
+      "regularization/lipschitz_norm": 46.16049575805664,
+      "regularization/w1": 0.0461605004966259,
+      "rewards/chosen": 0.003288782095607323,
+      "rewards/margins": 0.0004902812913537566,
+      "rewards/rejected": 0.0027985008042535663,
+      "stability/repetition_rate_mean": 0.3537946343421936,
+      "stability/response_length_mean": 56.875,
+      "stability/response_length_std": 20.152544021606445,
+      "stability/response_length_var": 406.125,
+      "stability/token_entropy_mean": 5.416015625,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.5843126177787781,
+      "kl": 0.027318641543388367,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56632812.8,
+      "logits/rejected": -50224707.2,
+      "logps/chosen": -421.8916015625,
+      "logps/rejected": -219.2817138671875,
+      "loss": 0.5509,
+      "loss/base_kto": 0.5002434849739075,
+      "metrics/advantage_var": 0.07726090401411057,
+      "regularization/lipschitz_norm": 52.84709548950195,
+      "regularization/w1": 0.05284709855914116,
+      "rewards/chosen": 0.0011051620356738567,
+      "rewards/margins": -0.0016500151716172696,
+      "rewards/rejected": 0.0027551772072911263,
+      "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": 0.6399968862533569,
+      "kl": 0.04743483290076256,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52704039.51677852,
+      "logits/rejected": -50776093.94152047,
+      "logps/chosen": -352.3003355704698,
+      "logps/rejected": -243.84438961988303,
+      "loss": 0.5532,
+      "loss/base_kto": 0.4999566078186035,
+      "metrics/advantage_var": 0.07218363136053085,
+      "regularization/lipschitz_norm": 56.39221954345703,
+      "regularization/w1": 0.05639222264289856,
+      "rewards/chosen": 0.002814074090663219,
+      "rewards/margins": -0.00016493607860582982,
+      "rewards/rejected": 0.0029790101692690487,
+      "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": 4.767578125,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.5783849954605103,
+      "kl": 0.051512785255908966,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54039232.4295302,
+      "logits/rejected": -48303834.57309941,
+      "logps/chosen": -375.13357277684565,
+      "logps/rejected": -180.2006807383041,
+      "loss": 0.5494,
+      "loss/base_kto": 0.4995388090610504,
+      "metrics/advantage_var": 0.04794999957084656,
+      "regularization/lipschitz_norm": 41.682228088378906,
+      "regularization/w1": 0.04168223217129707,
+      "rewards/chosen": 0.006553848717836726,
+      "rewards/margins": 0.0025606038216785853,
+      "rewards/rejected": 0.003993244896158141,
+      "stability/repetition_rate_mean": 0.3249044120311737,
+      "stability/response_length_mean": 55.375,
+      "stability/response_length_std": 16.361431121826172,
+      "stability/response_length_var": 267.6964416503906,
+      "stability/token_entropy_mean": 4.068359375,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.6621458530426025,
+      "kl": 0.05490356683731079,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56948316.22360248,
+      "logits/rejected": -49577713.50943396,
+      "logps/chosen": -344.4683618012422,
+      "logps/rejected": -230.42428262578616,
+      "loss": 0.5531,
+      "loss/base_kto": 0.4997487962245941,
+      "metrics/advantage_var": 0.05243971571326256,
+      "regularization/lipschitz_norm": 45.507625579833984,
+      "regularization/w1": 0.04550762102007866,
+      "rewards/chosen": 0.0043661816519980105,
+      "rewards/margins": 0.0019911858475652076,
+      "rewards/rejected": 0.002374995804432803,
+      "stability/repetition_rate_mean": 0.1015625,
+      "stability/response_length_mean": 24.5,
+      "stability/response_length_std": 32.710853576660156,
+      "stability/response_length_var": 1070.0,
+      "stability/token_entropy_mean": 4.6484375,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.5836600661277771,
+      "kl": 0.05875139310956001,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52669911.660606064,
+      "logits/rejected": -47820423.43225806,
+      "logps/chosen": -359.61126893939394,
+      "logps/rejected": -178.15209173387098,
+      "loss": 0.5496,
+      "loss/base_kto": 0.4995574951171875,
+      "metrics/advantage_var": 0.05988660454750061,
+      "regularization/lipschitz_norm": 46.30636215209961,
+      "regularization/w1": 0.046306367963552475,
+      "rewards/chosen": 0.010953712463378906,
+      "rewards/margins": 0.0034705385085075126,
+      "rewards/rejected": 0.007483173954871393,
+      "stability/repetition_rate_mean": 0.20815381407737732,
+      "stability/response_length_mean": 52.625,
+      "stability/response_length_std": 21.75800323486328,
+      "stability/response_length_var": 473.41070556640625,
+      "stability/token_entropy_mean": 5.314453125,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.5745318531990051,
+      "kl": 0.08697053045034409,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57561999.69032258,
+      "logits/rejected": -47265897.5030303,
+      "logps/chosen": -381.825,
+      "logps/rejected": -183.50355113636363,
+      "loss": 0.5544,
+      "loss/base_kto": 0.4995245635509491,
+      "metrics/advantage_var": 0.0773346945643425,
+      "regularization/lipschitz_norm": 53.16814041137695,
+      "regularization/w1": 0.053168147802352905,
+      "rewards/chosen": 0.012180716760696904,
+      "rewards/margins": 0.00430064315553518,
+      "rewards/rejected": 0.007880073605161724,
+      "stability/repetition_rate_mean": 0.20286151766777039,
+      "stability/response_length_mean": 17.875,
+      "stability/response_length_std": 23.485177993774414,
+      "stability/response_length_var": 551.5535888671875,
+      "stability/token_entropy_mean": 5.6875,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5525624752044678,
+      "kl": 0.09267856180667877,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52762746.21935484,
+      "logits/rejected": -49031695.515151516,
+      "logps/chosen": -379.0071572580645,
+      "logps/rejected": -220.3510890151515,
+      "loss": 0.5558,
+      "loss/base_kto": 0.4993644654750824,
+      "metrics/advantage_var": 0.07783074676990509,
+      "regularization/lipschitz_norm": 53.91382598876953,
+      "regularization/w1": 0.05391382798552513,
+      "rewards/chosen": 0.012132589278682585,
+      "rewards/margins": 0.005023109807996106,
+      "rewards/rejected": 0.007109479470686479,
+      "stability/repetition_rate_mean": 0.119140625,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.3349609375,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.6093087792396545,
+      "kl": 0.10173000395298004,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49496755.89189189,
+      "logits/rejected": -47891048.18604651,
+      "logps/chosen": -314.1371674408784,
+      "logps/rejected": -228.80718568313952,
+      "loss": 0.5574,
+      "loss/base_kto": 0.4992505609989166,
+      "metrics/advantage_var": 0.07528817653656006,
+      "regularization/lipschitz_norm": 53.78229904174805,
+      "regularization/w1": 0.05378229543566704,
+      "rewards/chosen": 0.013364916717683946,
+      "rewards/margins": 0.0056003187528426805,
+      "rewards/rejected": 0.007764597964841266,
+      "stability/repetition_rate_mean": 0.06745792925357819,
+      "stability/response_length_mean": 22.75,
+      "stability/response_length_std": 31.07479476928711,
+      "stability/response_length_var": 965.6428833007812,
+      "stability/token_entropy_mean": 2.16015625,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.6032807230949402,
+      "kl": 0.13186340034008026,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58616967.62913907,
+      "logits/rejected": -53431344.47337278,
+      "logps/chosen": -403.86936051324506,
+      "logps/rejected": -200.9628328402367,
+      "loss": 0.5561,
+      "loss/base_kto": 0.4992029666900635,
+      "metrics/advantage_var": 0.0667102262377739,
+      "regularization/lipschitz_norm": 47.66132354736328,
+      "regularization/w1": 0.047661323100328445,
+      "rewards/chosen": 0.01607713478290482,
+      "rewards/margins": 0.006778749535228491,
+      "rewards/rejected": 0.00929838524767633,
+      "stability/repetition_rate_mean": 0.17916667461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.25390625,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.6124517321586609,
+      "kl": 0.1272238940000534,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51464935.22580645,
+      "logits/rejected": -47132777.5030303,
+      "logps/chosen": -358.1264616935484,
+      "logps/rejected": -230.3289772727273,
+      "loss": 0.5594,
+      "loss/base_kto": 0.4986273944377899,
+      "metrics/advantage_var": 0.0937027856707573,
+      "regularization/lipschitz_norm": 55.11994552612305,
+      "regularization/w1": 0.055119942873716354,
+      "rewards/chosen": 0.020604954996416645,
+      "rewards/margins": 0.011022517827948516,
+      "rewards/rejected": 0.00958243716846813,
+      "stability/repetition_rate_mean": 0.11328125,
+      "stability/response_length_mean": 32.5,
+      "stability/response_length_std": 33.67915725708008,
+      "stability/response_length_var": 1134.2857666015625,
+      "stability/token_entropy_mean": 4.9453125,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.6401332020759583,
+      "kl": 0.1703931838274002,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58414500.102564104,
+      "logits/rejected": -50554680.19512195,
+      "logps/chosen": -380.5234124599359,
+      "logps/rejected": -187.12747713414635,
+      "loss": 0.5657,
+      "loss/base_kto": 0.4978005886077881,
+      "metrics/advantage_var": 0.12542474269866943,
+      "regularization/lipschitz_norm": 66.9811782836914,
+      "regularization/w1": 0.06698118895292282,
+      "rewards/chosen": 0.026595650575099848,
+      "rewards/margins": 0.016664286864556843,
+      "rewards/rejected": 0.009931363710543005,
+      "stability/repetition_rate_mean": 0.3812144994735718,
+      "stability/response_length_mean": 30.625,
+      "stability/response_length_std": 28.779644012451172,
+      "stability/response_length_var": 828.2678833007812,
+      "stability/token_entropy_mean": 3.736328125,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.6237964034080505,
+      "kl": 0.16741442680358887,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50386140.35443038,
+      "logits/rejected": -48979835.25925926,
+      "logps/chosen": -314.7296034414557,
+      "logps/rejected": -207.09116994598764,
+      "loss": 0.5682,
+      "loss/base_kto": 0.49736830592155457,
+      "metrics/advantage_var": 0.1331707090139389,
+      "regularization/lipschitz_norm": 66.74359130859375,
+      "regularization/w1": 0.06674359738826752,
+      "rewards/chosen": 0.026378269436993177,
+      "rewards/margins": 0.021123482745445267,
+      "rewards/rejected": 0.005254786691547912,
+      "stability/repetition_rate_mean": 0.20546108484268188,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 31.1264705657959,
+      "stability/response_length_var": 968.8571166992188,
+      "stability/token_entropy_mean": 3.689453125,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.6090478301048279,
+      "kl": 0.20850256085395813,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54716139.24324324,
+      "logits/rejected": -50905552.37209302,
+      "logps/chosen": -359.56579919763516,
+      "logps/rejected": -216.39162427325581,
+      "loss": 0.5714,
+      "loss/base_kto": 0.4984969198703766,
+      "metrics/advantage_var": 0.13034458458423615,
+      "regularization/lipschitz_norm": 67.18830871582031,
+      "regularization/w1": 0.0671883150935173,
+      "rewards/chosen": 0.030437920544598554,
+      "rewards/margins": 0.011710877204226068,
+      "rewards/rejected": 0.018727043340372486,
+      "stability/repetition_rate_mean": 0.24010416865348816,
+      "stability/response_length_mean": 40.75,
+      "stability/response_length_std": 31.093177795410156,
+      "stability/response_length_var": 966.7857055664062,
+      "stability/token_entropy_mean": 4.140625,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6770948171615601,
+      "kl": 0.2234809845685959,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59376942.26506024,
+      "logits/rejected": -53793147.01298701,
+      "logps/chosen": -399.5205666415663,
+      "logps/rejected": -209.8428622159091,
+      "loss": 0.5793,
+      "loss/base_kto": 0.4972439706325531,
+      "metrics/advantage_var": 0.22079260647296906,
+      "regularization/lipschitz_norm": 90.85485076904297,
+      "regularization/w1": 0.09085484594106674,
+      "rewards/chosen": 0.03818985353033227,
+      "rewards/margins": 0.022094057341581444,
+      "rewards/rejected": 0.016095796188750823,
+      "stability/repetition_rate_mean": 0.2014508843421936,
+      "stability/response_length_mean": 29.0,
+      "stability/response_length_std": 29.89505386352539,
+      "stability/response_length_var": 893.7142944335938,
+      "stability/token_entropy_mean": 3.40625,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.6547015309333801,
+      "kl": 0.26629599928855896,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57391949.10843374,
+      "logits/rejected": -53613863.896103896,
+      "logps/chosen": -378.19820689006025,
+      "logps/rejected": -205.89108157467533,
+      "loss": 0.5834,
+      "loss/base_kto": 0.4977465569972992,
+      "metrics/advantage_var": 0.21738283336162567,
+      "regularization/lipschitz_norm": 89.9118881225586,
+      "regularization/w1": 0.08991190046072006,
+      "rewards/chosen": 0.036506069711892,
+      "rewards/margins": 0.01679964862767379,
+      "rewards/rejected": 0.01970642108421821,
+      "stability/repetition_rate_mean": 0.31048768758773804,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.911476135253906,
+      "stability/response_length_var": 894.6964111328125,
+      "stability/token_entropy_mean": 4.3857421875,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.6483462452888489,
+      "kl": 0.32535991072654724,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60606263.088607594,
+      "logits/rejected": -53428704.39506173,
+      "logps/chosen": -364.06833465189874,
+      "logps/rejected": -200.9859905478395,
+      "loss": 0.5942,
+      "loss/base_kto": 0.49606046080589294,
+      "metrics/advantage_var": 0.26823315024375916,
+      "regularization/lipschitz_norm": 103.65421295166016,
+      "regularization/w1": 0.10365422070026398,
+      "rewards/chosen": 0.052421219741241844,
+      "rewards/margins": 0.031364458764897865,
+      "rewards/rejected": 0.02105676097634398,
+      "stability/repetition_rate_mean": 0.3151041865348816,
+      "stability/response_length_mean": 49.0,
+      "stability/response_length_std": 27.795169830322266,
+      "stability/response_length_var": 772.5714111328125,
+      "stability/token_entropy_mean": 4.830078125,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.6184374690055847,
+      "kl": 0.3553595244884491,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57748794.5620915,
+      "logits/rejected": -54355410.01197605,
+      "logps/chosen": -383.7023845996732,
+      "logps/rejected": -211.95403723802394,
+      "loss": 0.6004,
+      "loss/base_kto": 0.4964432418346405,
+      "metrics/advantage_var": 0.3870581090450287,
+      "regularization/lipschitz_norm": 109.87580871582031,
+      "regularization/w1": 0.10987582057714462,
+      "rewards/chosen": 0.05425496818193423,
+      "rewards/margins": 0.028023887709821817,
+      "rewards/rejected": 0.026231080472112416,
+      "stability/repetition_rate_mean": 0.1723366528749466,
+      "stability/response_length_mean": 27.875,
+      "stability/response_length_std": 30.31943130493164,
+      "stability/response_length_var": 919.2678833007812,
+      "stability/token_entropy_mean": 4.64453125,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.6122660636901855,
+      "kl": 0.39908552169799805,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60685662.68493151,
+      "logits/rejected": -55654294.06896552,
+      "logps/chosen": -416.8324058219178,
+      "logps/rejected": -216.1103627873563,
+      "loss": 0.6157,
+      "loss/base_kto": 0.4964618682861328,
+      "metrics/advantage_var": 0.5276581645011902,
+      "regularization/lipschitz_norm": 126.17388153076172,
+      "regularization/w1": 0.12617388367652893,
+      "rewards/chosen": 0.05468781353676156,
+      "rewards/margins": 0.029002990971923985,
+      "rewards/rejected": 0.025684822564837576,
+      "stability/repetition_rate_mean": 0.15937499701976776,
+      "stability/response_length_mean": 15.875,
+      "stability/response_length_std": 24.584766387939453,
+      "stability/response_length_var": 604.4107055664062,
+      "stability/token_entropy_mean": 3.49609375,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.5939815044403076,
+      "kl": 0.44654884934425354,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57391097.393548384,
+      "logits/rejected": -53830773.915151514,
+      "logps/chosen": -391.98465221774194,
+      "logps/rejected": -183.64661458333333,
+      "loss": 0.6243,
+      "loss/base_kto": 0.49498701095581055,
+      "metrics/advantage_var": 0.7337404489517212,
+      "regularization/lipschitz_norm": 144.0272216796875,
+      "regularization/w1": 0.1440272331237793,
+      "rewards/chosen": 0.06732863149335308,
+      "rewards/margins": 0.04009907965785947,
+      "rewards/rejected": 0.02722955183549361,
+      "stability/repetition_rate_mean": 0.22330255806446075,
+      "stability/response_length_mean": 38.0,
+      "stability/response_length_std": 29.13515281677246,
+      "stability/response_length_var": 848.8571166992188,
+      "stability/token_entropy_mean": 3.91796875,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5939186215400696,
+      "kl": 0.4990025460720062,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59841391.50920245,
+      "logits/rejected": -55011230.1656051,
+      "logps/chosen": -401.0247795245399,
+      "logps/rejected": -247.52191978503186,
+      "loss": 0.6335,
+      "loss/base_kto": 0.494366854429245,
+      "metrics/advantage_var": 0.9336435198783875,
+      "regularization/lipschitz_norm": 167.570068359375,
+      "regularization/w1": 0.167570099234581,
+      "rewards/chosen": 0.07913663197148797,
+      "rewards/margins": 0.044253570179806793,
+      "rewards/rejected": 0.03488306179168118,
+      "stability/repetition_rate_mean": 0.2270585298538208,
+      "stability/response_length_mean": 29.875,
+      "stability/response_length_std": 29.502723693847656,
+      "stability/response_length_var": 870.4107055664062,
+      "stability/token_entropy_mean": 3.9013671875,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.7301244139671326,
+      "kl": 0.4819028973579407,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57072553.11515152,
+      "logits/rejected": -50065289.08387097,
+      "logps/chosen": -366.36526988636365,
+      "logps/rejected": -189.91263860887096,
+      "loss": 0.6499,
+      "loss/base_kto": 0.4916805922985077,
+      "metrics/advantage_var": 1.7661739587783813,
+      "regularization/lipschitz_norm": 183.31504821777344,
+      "regularization/w1": 0.18331506848335266,
+      "rewards/chosen": 0.08244560126102332,
+      "rewards/margins": 0.0702617435278198,
+      "rewards/rejected": 0.012183857733203518,
+      "stability/repetition_rate_mean": 0.228515625,
+      "stability/response_length_mean": 56.25,
+      "stability/response_length_std": 21.920310974121094,
+      "stability/response_length_var": 480.5,
+      "stability/token_entropy_mean": 5.134765625,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.6743220090866089,
+      "kl": 0.5741755366325378,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51148250.98795181,
+      "logits/rejected": -51665840.20779221,
+      "logps/chosen": -329.439406061747,
+      "logps/rejected": -199.1564275568182,
+      "loss": 0.6604,
+      "loss/base_kto": 0.49365338683128357,
+      "metrics/advantage_var": 0.856822669506073,
+      "regularization/lipschitz_norm": 163.44825744628906,
+      "regularization/w1": 0.1634482592344284,
+      "rewards/chosen": 0.08324084224471126,
+      "rewards/margins": 0.050092330581051434,
+      "rewards/rejected": 0.03314851166365983,
+      "stability/repetition_rate_mean": 0.21016408503055573,
+      "stability/response_length_mean": 22.625,
+      "stability/response_length_std": 26.359804153442383,
+      "stability/response_length_var": 694.8392944335938,
+      "stability/token_entropy_mean": 5.6640625,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.6430969834327698,
+      "kl": 0.6443031430244446,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55426325.08235294,
+      "logits/rejected": -50193100.8,
+      "logps/chosen": -396.8122242647059,
+      "logps/rejected": -207.33588541666666,
+      "loss": 0.6752,
+      "loss/base_kto": 0.4903334677219391,
+      "metrics/advantage_var": 1.2321922779083252,
+      "regularization/lipschitz_norm": 185.36038208007812,
+      "regularization/w1": 0.1853603571653366,
+      "rewards/chosen": 0.11221859875847312,
+      "rewards/margins": 0.07321333511202943,
+      "rewards/rejected": 0.03900526364644368,
+      "stability/repetition_rate_mean": 0.23066024482250214,
+      "stability/response_length_mean": 29.375,
+      "stability/response_length_std": 29.193626403808594,
+      "stability/response_length_var": 852.2678833007812,
+      "stability/token_entropy_mean": 4.58984375,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.6289855241775513,
+      "kl": 0.6507260799407959,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54025167.8117647,
+      "logits/rejected": -49363421.86666667,
+      "logps/chosen": -360.06833639705883,
+      "logps/rejected": -191.04104166666667,
+      "loss": 0.7031,
+      "loss/base_kto": 0.49213314056396484,
+      "metrics/advantage_var": 2.246683120727539,
+      "regularization/lipschitz_norm": 223.3385467529297,
+      "regularization/w1": 0.2233385592699051,
+      "rewards/chosen": 0.09209441016702091,
+      "rewards/margins": 0.06563791106728947,
+      "rewards/rejected": 0.026456499099731447,
+      "stability/repetition_rate_mean": 0.3115246295928955,
+      "stability/response_length_mean": 58.125,
+      "stability/response_length_std": 10.960155487060547,
+      "stability/response_length_var": 120.125,
+      "stability/token_entropy_mean": 4.5673828125,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.619614839553833,
+      "kl": 0.7176083326339722,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55248286.47619048,
+      "logits/rejected": -48961236.21052632,
+      "logps/chosen": -415.52762276785717,
+      "logps/rejected": -186.48067434210526,
+      "loss": 0.7229,
+      "loss/base_kto": 0.487854927778244,
+      "metrics/advantage_var": 2.0578365325927734,
+      "regularization/lipschitz_norm": 231.88279724121094,
+      "regularization/w1": 0.23188279569149017,
+      "rewards/chosen": 0.13677164486476354,
+      "rewards/margins": 0.09510390426879539,
+      "rewards/rejected": 0.041667740595968145,
+      "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": 4.484375,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.7249348163604736,
+      "kl": 0.770777702331543,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54404449.10344828,
+      "logits/rejected": -49778060.2739726,
+      "logps/chosen": -369.70171066810343,
+      "logps/rejected": -212.1182041952055,
+      "loss": 0.7436,
+      "loss/base_kto": 0.49073153734207153,
+      "metrics/advantage_var": 1.810006022453308,
+      "regularization/lipschitz_norm": 221.78726196289062,
+      "regularization/w1": 0.22178728878498077,
+      "rewards/chosen": 0.12150281051109577,
+      "rewards/margins": 0.07411109539603195,
+      "rewards/rejected": 0.047391715115063814,
+      "stability/repetition_rate_mean": 0.2411106526851654,
+      "stability/response_length_mean": 22.5,
+      "stability/response_length_std": 28.33977508544922,
+      "stability/response_length_var": 803.1428833007812,
+      "stability/token_entropy_mean": 4.87109375,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.621860682964325,
+      "kl": 0.9223539233207703,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59132721.988165684,
+      "logits/rejected": -52588998.357615896,
+      "logps/chosen": -439.69253883136093,
+      "logps/rejected": -190.58294960678808,
+      "loss": 0.7621,
+      "loss/base_kto": 0.48786044120788574,
+      "metrics/advantage_var": 2.32456111907959,
+      "regularization/lipschitz_norm": 273.0086975097656,
+      "regularization/w1": 0.27300870418548584,
+      "rewards/chosen": 0.16493017433663093,
+      "rewards/margins": 0.09744154819757748,
+      "rewards/rejected": 0.06748862613905345,
+      "stability/repetition_rate_mean": 0.3512404263019562,
+      "stability/response_length_mean": 44.125,
+      "stability/response_length_std": 28.43256378173828,
+      "stability/response_length_var": 808.4107055664062,
+      "stability/token_entropy_mean": 3.59228515625,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.7261084318161011,
+      "kl": 0.9431293606758118,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56663225.06024096,
+      "logits/rejected": -48029942.02597403,
+      "logps/chosen": -354.8965549698795,
+      "logps/rejected": -194.3030768060065,
+      "loss": 0.7733,
+      "loss/base_kto": 0.4871121346950531,
+      "metrics/advantage_var": 4.2855119705200195,
+      "regularization/lipschitz_norm": 322.96240234375,
+      "regularization/w1": 0.3229624629020691,
+      "rewards/chosen": 0.15406293754118033,
+      "rewards/margins": 0.1036179753988305,
+      "rewards/rejected": 0.050444962142349836,
+      "stability/repetition_rate_mean": 0.19270984828472137,
+      "stability/response_length_mean": 26.875,
+      "stability/response_length_std": 25.570281982421875,
+      "stability/response_length_var": 653.8392944335938,
+      "stability/token_entropy_mean": 4.849609375,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.6424126625061035,
+      "kl": 1.0352585315704346,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52662171.48466258,
+      "logits/rejected": -47984167.133757964,
+      "logps/chosen": -374.4056269171779,
+      "logps/rejected": -210.5522243232484,
+      "loss": 0.8001,
+      "loss/base_kto": 0.4874431788921356,
+      "metrics/advantage_var": 4.169354438781738,
+      "regularization/lipschitz_norm": 315.4919128417969,
+      "regularization/w1": 0.3154919147491455,
+      "rewards/chosen": 0.1740191524014151,
+      "rewards/margins": 0.0994211801875728,
+      "rewards/rejected": 0.0745979722138423,
+      "stability/repetition_rate_mean": 0.2008928656578064,
+      "stability/response_length_mean": 35.25,
+      "stability/response_length_std": 31.129911422729492,
+      "stability/response_length_var": 969.0714111328125,
+      "stability/token_entropy_mean": 3.46875,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.6501648426055908,
+      "kl": 1.078233003616333,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54614294.12345679,
+      "logits/rejected": -49507992.30379747,
+      "logps/chosen": -310.3076774691358,
+      "logps/rejected": -196.53772745253164,
+      "loss": 0.8358,
+      "loss/base_kto": 0.48679420351982117,
+      "metrics/advantage_var": 7.409839630126953,
+      "regularization/lipschitz_norm": 392.8832702636719,
+      "regularization/w1": 0.3928832709789276,
+      "rewards/chosen": 0.16936941500063296,
+      "rewards/margins": 0.11922688233813265,
+      "rewards/rejected": 0.05014253266250031,
+      "stability/repetition_rate_mean": 0.19140625,
+      "stability/response_length_mean": 32.875,
+      "stability/response_length_std": 33.30996322631836,
+      "stability/response_length_var": 1109.5535888671875,
+      "stability/token_entropy_mean": 3.705078125,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.6980246901512146,
+      "kl": 1.122229814529419,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53080229.556886226,
+      "logits/rejected": -49681528.47058824,
+      "logps/chosen": -390.28709767964074,
+      "logps/rejected": -210.45874183006535,
+      "loss": 0.8981,
+      "loss/base_kto": 0.483151912689209,
+      "metrics/advantage_var": 7.226253032684326,
+      "regularization/lipschitz_norm": 438.9554748535156,
+      "regularization/w1": 0.4389554560184479,
+      "rewards/chosen": 0.21182136764069517,
+      "rewards/margins": 0.13783560089950975,
+      "rewards/rejected": 0.07398576674118541,
+      "stability/repetition_rate_mean": 0.19138340651988983,
+      "stability/response_length_mean": 18.625,
+      "stability/response_length_std": 26.52189064025879,
+      "stability/response_length_var": 703.4107055664062,
+      "stability/token_entropy_mean": 4.580078125,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.6413615942001343,
+      "kl": 1.1505008935928345,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60036446.14193548,
+      "logits/rejected": -53804826.375757575,
+      "logps/chosen": -390.84732862903223,
+      "logps/rejected": -197.08839962121212,
+      "loss": 0.8839,
+      "loss/base_kto": 0.4860563278198242,
+      "metrics/advantage_var": 4.479440212249756,
+      "regularization/lipschitz_norm": 372.052001953125,
+      "regularization/w1": 0.37205204367637634,
+      "rewards/chosen": 0.19376148100822202,
+      "rewards/margins": 0.11066871057624107,
+      "rewards/rejected": 0.08309277043198095,
+      "stability/repetition_rate_mean": 0.3934151828289032,
+      "stability/response_length_mean": 49.0,
+      "stability/response_length_std": 27.82085609436035,
+      "stability/response_length_var": 774.0,
+      "stability/token_entropy_mean": 5.1875,
+      "step": 195
+    },
+    {
+      "epoch": 1.0,
+      "step": 196,
+      "total_flos": 3.1861866257511875e+18,
+      "train_loss": 0.6270372308030421,
+      "train_runtime": 3152.8646,
+      "train_samples_per_second": 31.717,
+      "train_steps_per_second": 0.062
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 196,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 200,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.1861866257511875e+18,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}

mistral7b_kto_w1_gamma0.001/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:137a09d8908cce7eb46d8c86b2908fab853f68db7a0ae41f0299b4a8db223316
+size 8081