Upload folder using huggingface_hub

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/README.md

@@ -0,0 +1,64 @@
+---
+library_name: peft
+license: other
+base_model: mistralai/Mistral-7B-v0.1
+tags:
+- base_model:adapter:mistralai/Mistral-7B-v0.1
+- llama-factory
+- lora
+- transformers
+pipeline_tag: text-generation
+model-index:
+- name: mistral7b_kto_mmd_rho0.005_lambda_kr0.01
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+# mistral7b_kto_mmd_rho0.005_lambda_kr0.01
+
+This model is a fine-tuned version of [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) on the shp_kto_train dataset.
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 1e-06
+- train_batch_size: 8
+- eval_batch_size: 8
+- seed: 42
+- distributed_type: multi-GPU
+- num_devices: 8
+- gradient_accumulation_steps: 8
+- total_train_batch_size: 512
+- total_eval_batch_size: 64
+- optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
+- lr_scheduler_type: cosine
+- lr_scheduler_warmup_steps: 100
+- num_epochs: 1.0
+
+### Training results
+
+
+
+### Framework versions
+
+- PEFT 0.17.1
+- Transformers 4.57.1
+- Pytorch 2.10.0+cu128
+- Datasets 4.0.0
+- Tokenizers 0.22.2

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/adapter_config.json

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42a750dad12fd4ff1c2fc904e71a7853533c98ee2478b29837a68fc8452bc14e
+size 42002584

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.5200024314072668,
+    "train_runtime": 3159.9923,
+    "train_samples_per_second": 31.646,
+    "train_steps_per_second": 0.062
+}

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/chat_template.jinja

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/README.md

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/adapter_config.json

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42a750dad12fd4ff1c2fc904e71a7853533c98ee2478b29837a68fc8452bc14e
+size 42002584

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/chat_template.jinja

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/global_step196/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:539fffea780e3cb9e1a6d7a025ff2d15a56b0910e4c892a5fe8de51124c5f322
+size 31467333

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/global_step196/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a306f2a953145b4a63a964002232a7d32735cc20f5965dacf74a8a47d4743d94
+size 31467333

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/global_step196/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3aeaac8f38a36defb70347dc53e2fe4ead3429ba2c126072d47d31d7a3f0cca
+size 31467333

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/global_step196/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d67e6ba361d4d2a9a39f8c26e7437ddead92ddd977cfe5e7d313c7d7af5be1b
+size 31467397

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/global_step196/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1053783aaf8b929a0cdbe013b9bde7d5264347746ab7006490cbd658da66dba7
+size 31467397

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/global_step196/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:82659dcefe3de378a4a02a631e58051bd9f8315d9a435ca0197767935041dec6
+size 31467397

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/global_step196/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46f48369c7dddb7619e6e0a70bdd839a5cd1a3bd6860c21db7cd3685855bf21c
+size 31467397

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/global_step196/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e5beb5214719b07f0a6206e89f6bc5dd6fa519d3d33744889e0be95cc8a6c05
+size 31467397

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/global_step196/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2069b6704037dbeba648f0ac223a7bb4a6f0d3daa3a065ced478b71a0af93047
+size 42288409

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/latest

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/rng_state_0.pth

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/rng_state_1.pth

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/rng_state_2.pth

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/rng_state_3.pth

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/rng_state_4.pth

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/rng_state_5.pth

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/rng_state_6.pth

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/rng_state_7.pth

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/scheduler.pt

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/special_tokens_map.json

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/tokenizer.model

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/tokenizer_config.json

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/trainer_state.json

@@ -0,0 +1,965 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 196,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.025591810620601407,
+      "grad_norm": 0.5693546533584595,
+      "kl": 0.005381508264690638,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59717632.0,
+      "logits/rejected": -50410139.428571425,
+      "logps/chosen": -439.0407072368421,
+      "logps/rejected": -198.40538969494048,
+      "loss": 0.5055,
+      "loss/base_kto": 0.49999600648880005,
+      "metrics/advantage_var": 0.04680101200938225,
+      "regularization/mmd": 0.00584295904263854,
+      "regularization/rkhs_norm": 1.168591856956482,
+      "rewards/chosen": 0.0004892349733333839,
+      "rewards/margins": 8.741460675210289e-05,
+      "rewards/rejected": 0.000401820366581281,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.6020964980125427,
+      "kl": 0.013140608556568623,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59290737.03896104,
+      "logits/rejected": -49998308.240963854,
+      "logps/chosen": -429.92283887987014,
+      "logps/rejected": -192.86274002259037,
+      "loss": 0.5087,
+      "loss/base_kto": 0.4998807907104492,
+      "metrics/advantage_var": 0.06702536344528198,
+      "regularization/mmd": 0.009019351564347744,
+      "regularization/rkhs_norm": 1.803870439529419,
+      "rewards/chosen": 0.0015674108615169277,
+      "rewards/margins": 0.0010466737603613363,
+      "rewards/rejected": 0.0005207371011555913,
+      "stability/repetition_rate_mean": 0.3805190920829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.3486328125,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.6099647283554077,
+      "kl": 0.010814945213496685,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56234691.49090909,
+      "logits/rejected": -51924714.52903226,
+      "logps/chosen": -404.1313446969697,
+      "logps/rejected": -200.673828125,
+      "loss": 0.5087,
+      "loss/base_kto": 0.5002461671829224,
+      "metrics/advantage_var": 0.0660475566983223,
+      "regularization/mmd": 0.008868320845067501,
+      "regularization/rkhs_norm": 1.773664116859436,
+      "rewards/chosen": -5.2523883906277746e-05,
+      "rewards/margins": -0.0020758919964088372,
+      "rewards/rejected": 0.0020233681125025596,
+      "stability/repetition_rate_mean": 0.2747395634651184,
+      "stability/response_length_mean": 34.0,
+      "stability/response_length_std": 32.27117156982422,
+      "stability/response_length_var": 1041.4285888671875,
+      "stability/token_entropy_mean": 3.55078125,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.618466854095459,
+      "kl": 0.01798226684331894,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51667507.2,
+      "logits/rejected": -48054137.6,
+      "logps/chosen": -343.664892578125,
+      "logps/rejected": -225.62744140625,
+      "loss": 0.5094,
+      "loss/base_kto": 0.5001111030578613,
+      "metrics/advantage_var": 0.06671873480081558,
+      "regularization/mmd": 0.00892628263682127,
+      "regularization/rkhs_norm": 1.785256266593933,
+      "rewards/chosen": 0.003181634098291397,
+      "rewards/margins": -0.0004089426249265669,
+      "rewards/rejected": 0.003590576723217964,
+      "stability/repetition_rate_mean": 0.2040860652923584,
+      "stability/response_length_mean": 42.125,
+      "stability/response_length_std": 27.01553726196289,
+      "stability/response_length_var": 729.8392944335938,
+      "stability/token_entropy_mean": 4.453125,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.5777862668037415,
+      "kl": 0.01633647456765175,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56475732.74482758,
+      "logits/rejected": -52022096.45714286,
+      "logps/chosen": -378.50290948275864,
+      "logps/rejected": -202.79145089285714,
+      "loss": 0.5089,
+      "loss/base_kto": 0.5001418590545654,
+      "metrics/advantage_var": 0.0671335980296135,
+      "regularization/mmd": 0.00950543861836195,
+      "regularization/rkhs_norm": 1.901087760925293,
+      "rewards/chosen": -0.001145503541518902,
+      "rewards/margins": -0.0012863660915731798,
+      "rewards/rejected": 0.0001408625500542777,
+      "stability/repetition_rate_mean": 0.18034331500530243,
+      "stability/response_length_mean": 43.25,
+      "stability/response_length_std": 28.444181442260742,
+      "stability/response_length_var": 809.0714111328125,
+      "stability/token_entropy_mean": 4.7890625,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.6039624214172363,
+      "kl": 0.013820863328874111,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57726357.47651007,
+      "logits/rejected": -51529943.578947365,
+      "logps/chosen": -468.5825293624161,
+      "logps/rejected": -198.25539108187135,
+      "loss": 0.5094,
+      "loss/base_kto": 0.49989691376686096,
+      "metrics/advantage_var": 0.07160170376300812,
+      "regularization/mmd": 0.009396458975970745,
+      "regularization/rkhs_norm": 1.8792918920516968,
+      "rewards/chosen": -0.00012647900605361733,
+      "rewards/margins": 0.0006231328833561227,
+      "rewards/rejected": -0.00074961188940974,
+      "stability/repetition_rate_mean": 0.185546875,
+      "stability/response_length_mean": 48.375,
+      "stability/response_length_std": 28.93311309814453,
+      "stability/response_length_var": 837.125,
+      "stability/token_entropy_mean": 4.5625,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6789125204086304,
+      "kl": 0.023283982649445534,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57636902.88607595,
+      "logits/rejected": -53890566.32098766,
+      "logps/chosen": -385.9214547072785,
+      "logps/rejected": -195.13126929012347,
+      "loss": 0.509,
+      "loss/base_kto": 0.4997556209564209,
+      "metrics/advantage_var": 0.0707424208521843,
+      "regularization/mmd": 0.009495963342487812,
+      "regularization/rkhs_norm": 1.8991928100585938,
+      "rewards/chosen": 0.0015070421409003343,
+      "rewards/margins": 0.001985956171440471,
+      "rewards/rejected": -0.00047891403054013665,
+      "stability/repetition_rate_mean": 0.4045758843421936,
+      "stability/response_length_mean": 50.5,
+      "stability/response_length_std": 25.048524856567383,
+      "stability/response_length_var": 627.4285888671875,
+      "stability/token_entropy_mean": 5.46484375,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.5862559676170349,
+      "kl": 0.026907920837402344,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56639987.2,
+      "logits/rejected": -50230240.0,
+      "logps/chosen": -421.881982421875,
+      "logps/rejected": -219.275439453125,
+      "loss": 0.5092,
+      "loss/base_kto": 0.5001746416091919,
+      "metrics/advantage_var": 0.0762791857123375,
+      "regularization/mmd": 0.009773665107786655,
+      "regularization/rkhs_norm": 1.954732894897461,
+      "rewards/chosen": 0.002065117657184601,
+      "rewards/margins": -0.0013176284730434416,
+      "rewards/rejected": 0.0033827461302280426,
+      "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": 4.9013671875,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.6387529969215393,
+      "kl": 0.035909831523895264,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52695871.5704698,
+      "logits/rejected": -50777555.0877193,
+      "logps/chosen": -352.296875,
+      "logps/rejected": -243.83088907163742,
+      "loss": 0.5092,
+      "loss/base_kto": 0.5001186728477478,
+      "metrics/advantage_var": 0.07618097215890884,
+      "regularization/mmd": 0.009779557585716248,
+      "regularization/rkhs_norm": 1.955911636352539,
+      "rewards/chosen": 0.0031585363333657282,
+      "rewards/margins": -0.0011703405746967665,
+      "rewards/rejected": 0.004328876908062495,
+      "stability/repetition_rate_mean": 0.0974486693739891,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.87324333190918,
+      "stability/response_length_var": 892.4107055664062,
+      "stability/token_entropy_mean": 4.796875,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.5790932774543762,
+      "kl": 0.04191715270280838,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54039015.946308725,
+      "logits/rejected": -48296516.865497075,
+      "logps/chosen": -375.1137793624161,
+      "logps/rejected": -180.1954495614035,
+      "loss": 0.5085,
+      "loss/base_kto": 0.49944159388542175,
+      "metrics/advantage_var": 0.05880240350961685,
+      "regularization/mmd": 0.008453378453850746,
+      "regularization/rkhs_norm": 1.6906756162643433,
+      "rewards/chosen": 0.008534903494303659,
+      "rewards/margins": 0.0040195142414792655,
+      "rewards/rejected": 0.004515389252824393,
+      "stability/repetition_rate_mean": 0.2932637929916382,
+      "stability/response_length_mean": 53.375,
+      "stability/response_length_std": 21.480472564697266,
+      "stability/response_length_var": 461.41070556640625,
+      "stability/token_entropy_mean": 4.115234375,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.6598390936851501,
+      "kl": 0.05445736274123192,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56959065.04347826,
+      "logits/rejected": -49578112.80503145,
+      "logps/chosen": -344.4817546583851,
+      "logps/rejected": -230.40470224056602,
+      "loss": 0.5091,
+      "loss/base_kto": 0.5001540184020996,
+      "metrics/advantage_var": 0.0567297525703907,
+      "regularization/mmd": 0.008397291414439678,
+      "regularization/rkhs_norm": 1.6794582605361938,
+      "rewards/chosen": 0.0030251911708286832,
+      "rewards/margins": -0.0013078764442484973,
+      "rewards/rejected": 0.0043330676150771805,
+      "stability/repetition_rate_mean": 0.11003988981246948,
+      "stability/response_length_mean": 22.375,
+      "stability/response_length_std": 30.236862182617188,
+      "stability/response_length_var": 914.2678833007812,
+      "stability/token_entropy_mean": 4.515625,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.5815622210502625,
+      "kl": 0.05094662308692932,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52672549.236363634,
+      "logits/rejected": -47823535.07096774,
+      "logps/chosen": -359.6089488636364,
+      "logps/rejected": -178.15403225806452,
+      "loss": 0.5085,
+      "loss/base_kto": 0.49942919611930847,
+      "metrics/advantage_var": 0.05574983358383179,
+      "regularization/mmd": 0.007710353937000036,
+      "regularization/rkhs_norm": 1.5420708656311035,
+      "rewards/chosen": 0.011186000795075387,
+      "rewards/margins": 0.0038980508946020696,
+      "rewards/rejected": 0.007287949900473318,
+      "stability/repetition_rate_mean": 0.1785714328289032,
+      "stability/response_length_mean": 47.25,
+      "stability/response_length_std": 28.684242248535156,
+      "stability/response_length_var": 822.7857055664062,
+      "stability/token_entropy_mean": 4.447265625,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.5738525390625,
+      "kl": 0.07379233092069626,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57558207.58709677,
+      "logits/rejected": -47281902.93333333,
+      "logps/chosen": -381.80806451612904,
+      "logps/rejected": -183.5120146780303,
+      "loss": 0.509,
+      "loss/base_kto": 0.49920693039894104,
+      "metrics/advantage_var": 0.07559790462255478,
+      "regularization/mmd": 0.009121461771428585,
+      "regularization/rkhs_norm": 1.8242921829223633,
+      "rewards/chosen": 0.01387463846514302,
+      "rewards/margins": 0.006840014481008461,
+      "rewards/rejected": 0.007034623984134559,
+      "stability/repetition_rate_mean": 0.18141163885593414,
+      "stability/response_length_mean": 21.5,
+      "stability/response_length_std": 26.635099411010742,
+      "stability/response_length_var": 709.4285888671875,
+      "stability/token_entropy_mean": 5.423828125,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5529053211212158,
+      "kl": 0.08646707981824875,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52772190.14193548,
+      "logits/rejected": -49029666.13333333,
+      "logps/chosen": -379.00892137096776,
+      "logps/rejected": -220.35814393939393,
+      "loss": 0.5089,
+      "loss/base_kto": 0.4993232190608978,
+      "metrics/advantage_var": 0.07030369341373444,
+      "regularization/mmd": 0.008873376064002514,
+      "regularization/rkhs_norm": 1.7746753692626953,
+      "rewards/chosen": 0.011954639803978704,
+      "rewards/margins": 0.005551152425078003,
+      "rewards/rejected": 0.006403487378900701,
+      "stability/repetition_rate_mean": 0.18359375,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.73046875,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.610389769077301,
+      "kl": 0.09753876179456711,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49507348.75675676,
+      "logits/rejected": -47883686.697674416,
+      "logps/chosen": -314.1405458192568,
+      "logps/rejected": -228.81352198401163,
+      "loss": 0.5095,
+      "loss/base_kto": 0.49918660521507263,
+      "metrics/advantage_var": 0.07326646149158478,
+      "regularization/mmd": 0.00923239067196846,
+      "regularization/rkhs_norm": 1.8464781045913696,
+      "rewards/chosen": 0.013025604389809273,
+      "rewards/margins": 0.005891873767584243,
+      "rewards/rejected": 0.00713373062222503,
+      "stability/repetition_rate_mean": 0.08548677712678909,
+      "stability/response_length_mean": 22.75,
+      "stability/response_length_std": 31.07479476928711,
+      "stability/response_length_var": 965.6428833007812,
+      "stability/token_entropy_mean": 2.150390625,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.6044193506240845,
+      "kl": 0.1186995729804039,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58622508.0794702,
+      "logits/rejected": -53452569.75147929,
+      "logps/chosen": -403.86501448675494,
+      "logps/rejected": -200.98915957840237,
+      "loss": 0.5092,
+      "loss/base_kto": 0.49880704283714294,
+      "metrics/advantage_var": 0.06772632896900177,
+      "regularization/mmd": 0.008787428960204124,
+      "regularization/rkhs_norm": 1.7574856281280518,
+      "rewards/chosen": 0.016511326594068513,
+      "rewards/margins": 0.009846882590483691,
+      "rewards/rejected": 0.006664444003584822,
+      "stability/repetition_rate_mean": 0.18307292461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.234375,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.6132326126098633,
+      "kl": 0.12901714444160461,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51477778.16774193,
+      "logits/rejected": -47139790.35151515,
+      "logps/chosen": -358.1587701612903,
+      "logps/rejected": -230.33401988636365,
+      "loss": 0.5098,
+      "loss/base_kto": 0.49895045161247253,
+      "metrics/advantage_var": 0.082537442445755,
+      "regularization/mmd": 0.00979132205247879,
+      "regularization/rkhs_norm": 1.9582645893096924,
+      "rewards/chosen": 0.017374797021189044,
+      "rewards/margins": 0.00829642441265744,
+      "rewards/rejected": 0.009078372608531605,
+      "stability/repetition_rate_mean": 0.09735831618309021,
+      "stability/response_length_mean": 31.875,
+      "stability/response_length_std": 33.0516357421875,
+      "stability/response_length_var": 1092.4107666015625,
+      "stability/token_entropy_mean": 4.90625,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.6399518847465515,
+      "kl": 0.17196393013000488,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58408185.43589743,
+      "logits/rejected": -50558657.56097561,
+      "logps/chosen": -380.54271834935895,
+      "logps/rejected": -187.10587366615854,
+      "loss": 0.51,
+      "loss/base_kto": 0.4982968270778656,
+      "metrics/advantage_var": 0.10217716544866562,
+      "regularization/mmd": 0.011231121607124805,
+      "regularization/rkhs_norm": 2.2462244033813477,
+      "rewards/chosen": 0.024664967488019895,
+      "rewards/margins": 0.012572131468550424,
+      "rewards/rejected": 0.01209283601946947,
+      "stability/repetition_rate_mean": 0.2192234843969345,
+      "stability/response_length_mean": 33.625,
+      "stability/response_length_std": 29.442134857177734,
+      "stability/response_length_var": 866.8392944335938,
+      "stability/token_entropy_mean": 3.7060546875,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.6170993447303772,
+      "kl": 0.1672627478837967,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50378544.60759494,
+      "logits/rejected": -48969532.04938272,
+      "logps/chosen": -314.7658722310127,
+      "logps/rejected": -207.07737750771605,
+      "loss": 0.5107,
+      "loss/base_kto": 0.49796175956726074,
+      "metrics/advantage_var": 0.14120100438594818,
+      "regularization/mmd": 0.012432265095412731,
+      "regularization/rkhs_norm": 2.48645281791687,
+      "rewards/chosen": 0.022749973248831833,
+      "rewards/margins": 0.01611456490621136,
+      "rewards/rejected": 0.0066354083426204725,
+      "stability/repetition_rate_mean": 0.19374233484268188,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 31.1264705657959,
+      "stability/response_length_var": 968.8571166992188,
+      "stability/token_entropy_mean": 3.751953125,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.6102624535560608,
+      "kl": 0.20016632974147797,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54725538.5945946,
+      "logits/rejected": -50922793.674418606,
+      "logps/chosen": -359.54178103885135,
+      "logps/rejected": -216.3764989098837,
+      "loss": 0.511,
+      "loss/base_kto": 0.4985528886318207,
+      "metrics/advantage_var": 0.12865900993347168,
+      "regularization/mmd": 0.01211935468018055,
+      "regularization/rkhs_norm": 2.423870801925659,
+      "rewards/chosen": 0.03283950444814321,
+      "rewards/margins": 0.01260079583006786,
+      "rewards/rejected": 0.02023870861807535,
+      "stability/repetition_rate_mean": 0.23773986101150513,
+      "stability/response_length_mean": 37.5,
+      "stability/response_length_std": 29.640705108642578,
+      "stability/response_length_var": 878.5714111328125,
+      "stability/token_entropy_mean": 4.15625,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6752254366874695,
+      "kl": 0.22721095383167267,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59377960.09638554,
+      "logits/rejected": -53795347.948051944,
+      "logps/chosen": -399.49143448795184,
+      "logps/rejected": -209.84331879058442,
+      "loss": 0.5118,
+      "loss/base_kto": 0.4968232214450836,
+      "metrics/advantage_var": 0.2163117676973343,
+      "regularization/mmd": 0.015834027901291847,
+      "regularization/rkhs_norm": 3.1668055057525635,
+      "rewards/chosen": 0.04110269374158009,
+      "rewards/margins": 0.02505132391351692,
+      "rewards/rejected": 0.016051369828063173,
+      "stability/repetition_rate_mean": 0.189453125,
+      "stability/response_length_mean": 35.25,
+      "stability/response_length_std": 31.494897842407227,
+      "stability/response_length_var": 991.9285888671875,
+      "stability/token_entropy_mean": 3.11328125,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.6515340805053711,
+      "kl": 0.2592513859272003,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57376342.361445785,
+      "logits/rejected": -53605545.55844156,
+      "logps/chosen": -378.17446347891564,
+      "logps/rejected": -205.87290736607142,
+      "loss": 0.5132,
+      "loss/base_kto": 0.49767494201660156,
+      "metrics/advantage_var": 0.22866569459438324,
+      "regularization/mmd": 0.016469214111566544,
+      "regularization/rkhs_norm": 3.2938427925109863,
+      "rewards/chosen": 0.03888099164847868,
+      "rewards/margins": 0.01735842323064468,
+      "rewards/rejected": 0.021522568417833995,
+      "stability/repetition_rate_mean": 0.3042376935482025,
+      "stability/response_length_mean": 35.875,
+      "stability/response_length_std": 29.811012268066406,
+      "stability/response_length_var": 888.6964111328125,
+      "stability/token_entropy_mean": 4.0595703125,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.649266242980957,
+      "kl": 0.31319910287857056,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60592257.62025317,
+      "logits/rejected": -53408388.74074074,
+      "logps/chosen": -364.08156151107596,
+      "logps/rejected": -200.96959394290124,
+      "loss": 0.5148,
+      "loss/base_kto": 0.49650001525878906,
+      "metrics/advantage_var": 0.25559911131858826,
+      "regularization/mmd": 0.01740214042365551,
+      "regularization/rkhs_norm": 3.480428457260132,
+      "rewards/chosen": 0.051097362856321694,
+      "rewards/margins": 0.028401050293402737,
+      "rewards/rejected": 0.022696312562918958,
+      "stability/repetition_rate_mean": 0.3111979365348816,
+      "stability/response_length_mean": 49.25,
+      "stability/response_length_std": 27.316923141479492,
+      "stability/response_length_var": 746.2142944335938,
+      "stability/token_entropy_mean": 4.77734375,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.6187500357627869,
+      "kl": 0.34892186522483826,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57730777.51633987,
+      "logits/rejected": -54345593.1017964,
+      "logps/chosen": -383.69628267973854,
+      "logps/rejected": -211.9707616017964,
+      "loss": 0.5155,
+      "loss/base_kto": 0.49620935320854187,
+      "metrics/advantage_var": 0.3712993860244751,
+      "regularization/mmd": 0.020150119438767433,
+      "regularization/rkhs_norm": 4.03002405166626,
+      "rewards/chosen": 0.05486513274947023,
+      "rewards/margins": 0.030305783392459158,
+      "rewards/rejected": 0.024559349357011075,
+      "stability/repetition_rate_mean": 0.18938209116458893,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.687660217285156,
+      "stability/response_length_var": 881.3571166992188,
+      "stability/token_entropy_mean": 4.5625,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.611441969871521,
+      "kl": 0.3908672630786896,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60671656.32876712,
+      "logits/rejected": -55648673.83908046,
+      "logps/chosen": -416.7914704623288,
+      "logps/rejected": -216.10171964798852,
+      "loss": 0.5184,
+      "loss/base_kto": 0.4961169362068176,
+      "metrics/advantage_var": 0.5313876271247864,
+      "regularization/mmd": 0.023098159581422806,
+      "regularization/rkhs_norm": 4.619632244110107,
+      "rewards/chosen": 0.058781075150999305,
+      "rewards/margins": 0.03223435023960551,
+      "rewards/rejected": 0.02654672491139379,
+      "stability/repetition_rate_mean": 0.16215278208255768,
+      "stability/response_length_mean": 15.25,
+      "stability/response_length_std": 24.93276596069336,
+      "stability/response_length_var": 621.6428833007812,
+      "stability/token_entropy_mean": 3.640625,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.5946590304374695,
+      "kl": 0.4510687291622162,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57380811.1483871,
+      "logits/rejected": -53835955.97575758,
+      "logps/chosen": -391.98538306451616,
+      "logps/rejected": -183.6492897727273,
+      "loss": 0.5201,
+      "loss/base_kto": 0.49492835998535156,
+      "metrics/advantage_var": 0.8019859194755554,
+      "regularization/mmd": 0.02793542854487896,
+      "regularization/rkhs_norm": 5.587085723876953,
+      "rewards/chosen": 0.06725460175544985,
+      "rewards/margins": 0.040291827305321125,
+      "rewards/rejected": 0.026962774450128728,
+      "stability/repetition_rate_mean": 0.20181818306446075,
+      "stability/response_length_mean": 38.0,
+      "stability/response_length_std": 29.13515281677246,
+      "stability/response_length_var": 848.8571166992188,
+      "stability/token_entropy_mean": 3.99609375,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5928955674171448,
+      "kl": 0.49744081497192383,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59853874.25766871,
+      "logits/rejected": -55008706.03821656,
+      "logps/chosen": -401.01260544478527,
+      "logps/rejected": -247.52005374203821,
+      "loss": 0.5209,
+      "loss/base_kto": 0.4942546486854553,
+      "metrics/advantage_var": 1.02751624584198,
+      "regularization/mmd": 0.03274526074528694,
+      "regularization/rkhs_norm": 6.549052715301514,
+      "rewards/chosen": 0.08035813080021209,
+      "rewards/margins": 0.04529071829624675,
+      "rewards/rejected": 0.03506741250396534,
+      "stability/repetition_rate_mean": 0.2230902761220932,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.936838150024414,
+      "stability/response_length_var": 896.2142944335938,
+      "stability/token_entropy_mean": 4.078125,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.7285244464874268,
+      "kl": 0.4897715151309967,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57075643.733333334,
+      "logits/rejected": -50076533.26451613,
+      "logps/chosen": -366.3785511363636,
+      "logps/rejected": -189.9105216733871,
+      "loss": 0.523,
+      "loss/base_kto": 0.49187538027763367,
+      "metrics/advantage_var": 1.7579011917114258,
+      "regularization/mmd": 0.03410874679684639,
+      "regularization/rkhs_norm": 6.821749210357666,
+      "rewards/chosen": 0.08111698844216086,
+      "rewards/margins": 0.06872179263498077,
+      "rewards/rejected": 0.012395195807180097,
+      "stability/repetition_rate_mean": 0.220703125,
+      "stability/response_length_mean": 56.25,
+      "stability/response_length_std": 21.920310974121094,
+      "stability/response_length_var": 480.5,
+      "stability/token_entropy_mean": 5.052734375,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.6769351959228516,
+      "kl": 0.5653125047683716,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51149342.84337349,
+      "logits/rejected": -51664423.896103896,
+      "logps/chosen": -329.4778567394578,
+      "logps/rejected": -199.1466492491883,
+      "loss": 0.5245,
+      "loss/base_kto": 0.4942997097969055,
+      "metrics/advantage_var": 0.7970510125160217,
+      "regularization/mmd": 0.029787013307213783,
+      "regularization/rkhs_norm": 5.95740270614624,
+      "rewards/chosen": 0.07939604104283345,
+      "rewards/margins": 0.045269124553789654,
+      "rewards/rejected": 0.03412691648904379,
+      "stability/repetition_rate_mean": 0.26596763730049133,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.73423957824707,
+      "stability/response_length_var": 884.125,
+      "stability/token_entropy_mean": 5.228515625,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.6468893885612488,
+      "kl": 0.6396333575248718,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55421066.54117647,
+      "logits/rejected": -50181959.68,
+      "logps/chosen": -396.82509191176473,
+      "logps/rejected": -207.365703125,
+      "loss": 0.5281,
+      "loss/base_kto": 0.49031582474708557,
+      "metrics/advantage_var": 1.2932097911834717,
+      "regularization/mmd": 0.036661840975284576,
+      "regularization/rkhs_norm": 7.33236837387085,
+      "rewards/chosen": 0.11093491946949678,
+      "rewards/margins": 0.0749119384616029,
+      "rewards/rejected": 0.03602298100789388,
+      "stability/repetition_rate_mean": 0.2291666716337204,
+      "stability/response_length_mean": 35.25,
+      "stability/response_length_std": 31.01957893371582,
+      "stability/response_length_var": 962.2142944335938,
+      "stability/token_entropy_mean": 4.66796875,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.6322670578956604,
+      "kl": 0.6517454981803894,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54015186.823529415,
+      "logits/rejected": -49345460.906666666,
+      "logps/chosen": -360.07118566176473,
+      "logps/rejected": -191.05200520833333,
+      "loss": 0.5318,
+      "loss/base_kto": 0.4920540750026703,
+      "metrics/advantage_var": 2.2298808097839355,
+      "regularization/mmd": 0.0417025052011013,
+      "regularization/rkhs_norm": 8.340499877929688,
+      "rewards/chosen": 0.09180937935324276,
+      "rewards/margins": 0.0664488141676959,
+      "rewards/rejected": 0.025360565185546875,
+      "stability/repetition_rate_mean": 0.3406159281730652,
+      "stability/response_length_mean": 61.375,
+      "stability/response_length_std": 7.424621105194092,
+      "stability/response_length_var": 55.125,
+      "stability/token_entropy_mean": 4.5703125,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.619647204875946,
+      "kl": 0.7206474542617798,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55251065.9047619,
+      "logits/rejected": -48947941.05263158,
+      "logps/chosen": -415.55845424107144,
+      "logps/rejected": -186.46698961759867,
+      "loss": 0.5348,
+      "loss/base_kto": 0.48844751715660095,
+      "metrics/advantage_var": 1.9670952558517456,
+      "regularization/mmd": 0.04304032027721405,
+      "regularization/rkhs_norm": 8.608064651489258,
+      "rewards/chosen": 0.13368967601231166,
+      "rewards/margins": 0.0906540558750468,
+      "rewards/rejected": 0.043035620137264856,
+      "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": 4.44140625,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.720912516117096,
+      "kl": 0.7607746124267578,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54404996.4137931,
+      "logits/rejected": -49794777.42465753,
+      "logps/chosen": -369.7616738505747,
+      "logps/rejected": -212.15198255565068,
+      "loss": 0.5371,
+      "loss/base_kto": 0.4911351203918457,
+      "metrics/advantage_var": 1.7204967737197876,
+      "regularization/mmd": 0.04163478687405586,
+      "regularization/rkhs_norm": 8.326957702636719,
+      "rewards/chosen": 0.11550897839425624,
+      "rewards/margins": 0.07149485110711007,
+      "rewards/rejected": 0.04401412728714616,
+      "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": 4.8984375,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.6247206926345825,
+      "kl": 0.924032986164093,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59132631.10059172,
+      "logits/rejected": -52570654.51655629,
+      "logps/chosen": -439.69008875739644,
+      "logps/rejected": -190.55952504139074,
+      "loss": 0.5386,
+      "loss/base_kto": 0.48805713653564453,
+      "metrics/advantage_var": 2.3215138912200928,
+      "regularization/mmd": 0.0507260225713253,
+      "regularization/rkhs_norm": 10.145204544067383,
+      "rewards/chosen": 0.16517957427797939,
+      "rewards/margins": 0.09534949453615406,
+      "rewards/rejected": 0.06983007974182533,
+      "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": 3.5888671875,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.7177579998970032,
+      "kl": 0.9547041058540344,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56654465.54216868,
+      "logits/rejected": -48027790.96103896,
+      "logps/chosen": -354.88453030873495,
+      "logps/rejected": -194.3028612012987,
+      "loss": 0.5416,
+      "loss/base_kto": 0.48682698607444763,
+      "metrics/advantage_var": 4.120774745941162,
+      "regularization/mmd": 0.06023603305220604,
+      "regularization/rkhs_norm": 12.047205924987793,
+      "rewards/chosen": 0.15526514168245248,
+      "rewards/margins": 0.10479820168199273,
+      "rewards/rejected": 0.050466940000459745,
+      "stability/repetition_rate_mean": 0.2005799561738968,
+      "stability/response_length_mean": 27.125,
+      "stability/response_length_std": 25.737340927124023,
+      "stability/response_length_var": 662.4107055664062,
+      "stability/token_entropy_mean": 4.8115234375,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.6431882977485657,
+      "kl": 1.0281428098678589,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52652606.82208589,
+      "logits/rejected": -47977778.5477707,
+      "logps/chosen": -374.4046922929448,
+      "logps/rejected": -210.54264530254778,
+      "loss": 0.5458,
+      "loss/base_kto": 0.48759326338768005,
+      "metrics/advantage_var": 4.256661415100098,
+      "regularization/mmd": 0.062359750270843506,
+      "regularization/rkhs_norm": 12.471949577331543,
+      "rewards/chosen": 0.17411304544086106,
+      "rewards/margins": 0.09855715900823765,
+      "rewards/rejected": 0.07555588643262341,
+      "stability/repetition_rate_mean": 0.16484375298023224,
+      "stability/response_length_mean": 34.75,
+      "stability/response_length_std": 31.54928207397461,
+      "stability/response_length_var": 995.3571166992188,
+      "stability/token_entropy_mean": 3.6171875,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.6504088044166565,
+      "kl": 1.0762888193130493,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54606911.209876545,
+      "logits/rejected": -49525303.088607594,
+      "logps/chosen": -310.3205536265432,
+      "logps/rejected": -196.54314181170886,
+      "loss": 0.5519,
+      "loss/base_kto": 0.4868957996368408,
+      "metrics/advantage_var": 7.445571422576904,
+      "regularization/mmd": 0.07445289939641953,
+      "regularization/rkhs_norm": 14.890579223632812,
+      "rewards/chosen": 0.16808104809419608,
+      "rewards/margins": 0.11847921255361774,
+      "rewards/rejected": 0.049601835540578336,
+      "stability/repetition_rate_mean": 0.19921875,
+      "stability/response_length_mean": 32.875,
+      "stability/response_length_std": 33.30996322631836,
+      "stability/response_length_var": 1109.5535888671875,
+      "stability/token_entropy_mean": 3.716796875,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.7037036418914795,
+      "kl": 1.1190913915634155,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53083148.263473056,
+      "logits/rejected": -49686661.85620915,
+      "logps/chosen": -390.29261788922156,
+      "logps/rejected": -210.45340584150327,
+      "loss": 0.5607,
+      "loss/base_kto": 0.4832470118999481,
+      "metrics/advantage_var": 7.204617500305176,
+      "regularization/mmd": 0.0810161754488945,
+      "regularization/rkhs_norm": 16.203235626220703,
+      "rewards/chosen": 0.21127086342451815,
+      "rewards/margins": 0.1367532206740586,
+      "rewards/rejected": 0.07451764275045956,
+      "stability/repetition_rate_mean": 0.19138340651988983,
+      "stability/response_length_mean": 18.25,
+      "stability/response_length_std": 26.778188705444336,
+      "stability/response_length_var": 717.0714111328125,
+      "stability/token_entropy_mean": 4.83203125,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.6389921307563782,
+      "kl": 1.1464111804962158,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60020286.76129032,
+      "logits/rejected": -53779549.09090909,
+      "logps/chosen": -390.8505040322581,
+      "logps/rejected": -197.09185606060606,
+      "loss": 0.5603,
+      "loss/base_kto": 0.48602548241615295,
+      "metrics/advantage_var": 4.388164520263672,
+      "regularization/mmd": 0.06844492256641388,
+      "regularization/rkhs_norm": 13.688984870910645,
+      "rewards/chosen": 0.1934424861784904,
+      "rewards/margins": 0.11069587062414322,
+      "rewards/rejected": 0.08274661555434719,
+      "stability/repetition_rate_mean": 0.3953683078289032,
+      "stability/response_length_mean": 49.0,
+      "stability/response_length_std": 27.82085609436035,
+      "stability/response_length_var": 774.0,
+      "stability/token_entropy_mean": 5.16796875,
+      "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_mmd_rho0.005_lambda_kr0.01/checkpoint-196/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7afba6a96f544b9b3a7bd795e549d676d18d66bf7bce223ed504d9c66dfaf4c7
+size 8081

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/checkpoint-196/zero_to_fp32.py

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/special_tokens_map.json

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/stability_metrics.jsonl

@@ -0,0 +1,38 @@
+{"stability/response_length_mean": 46.125, "stability/response_length_std": 27.231481552124023, "stability/response_length_var": 741.5535888671875, "stability/token_entropy_mean": 3.3486328125, "stability/repetition_rate_mean": 0.3805190920829773, "step": 10}
+{"stability/response_length_mean": 34.0, "stability/response_length_std": 32.27117156982422, "stability/response_length_var": 1041.4285888671875, "stability/token_entropy_mean": 3.55078125, "stability/repetition_rate_mean": 0.2747395634651184, "step": 15}
+{"stability/response_length_mean": 42.125, "stability/response_length_std": 27.01553726196289, "stability/response_length_var": 729.8392944335938, "stability/token_entropy_mean": 4.453125, "stability/repetition_rate_mean": 0.2040860652923584, "step": 20}
+{"stability/response_length_mean": 43.25, "stability/response_length_std": 28.444181442260742, "stability/response_length_var": 809.0714111328125, "stability/token_entropy_mean": 4.7890625, "stability/repetition_rate_mean": 0.18034331500530243, "step": 25}
+{"stability/response_length_mean": 48.375, "stability/response_length_std": 28.93311309814453, "stability/response_length_var": 837.125, "stability/token_entropy_mean": 4.5625, "stability/repetition_rate_mean": 0.185546875, "step": 30}
+{"stability/response_length_mean": 50.5, "stability/response_length_std": 25.048524856567383, "stability/response_length_var": 627.4285888671875, "stability/token_entropy_mean": 5.46484375, "stability/repetition_rate_mean": 0.4045758843421936, "step": 35}
+{"stability/response_length_mean": 33.125, "stability/response_length_std": 33.01704025268555, "stability/response_length_var": 1090.125, "stability/token_entropy_mean": 4.9013671875, "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": 4.796875, "stability/repetition_rate_mean": 0.0974486693739891, "step": 45}
+{"stability/response_length_mean": 53.375, "stability/response_length_std": 21.480472564697266, "stability/response_length_var": 461.41070556640625, "stability/token_entropy_mean": 4.115234375, "stability/repetition_rate_mean": 0.2932637929916382, "step": 50}
+{"stability/response_length_mean": 22.375, "stability/response_length_std": 30.236862182617188, "stability/response_length_var": 914.2678833007812, "stability/token_entropy_mean": 4.515625, "stability/repetition_rate_mean": 0.11003988981246948, "step": 55}
+{"stability/response_length_mean": 47.25, "stability/response_length_std": 28.684242248535156, "stability/response_length_var": 822.7857055664062, "stability/token_entropy_mean": 4.447265625, "stability/repetition_rate_mean": 0.1785714328289032, "step": 60}
+{"stability/response_length_mean": 21.5, "stability/response_length_std": 26.635099411010742, "stability/response_length_var": 709.4285888671875, "stability/token_entropy_mean": 5.423828125, "stability/repetition_rate_mean": 0.18141163885593414, "step": 65}
+{"stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.73046875, "stability/repetition_rate_mean": 0.18359375, "step": 70}
+{"stability/response_length_mean": 22.75, "stability/response_length_std": 31.07479476928711, "stability/response_length_var": 965.6428833007812, "stability/token_entropy_mean": 2.150390625, "stability/repetition_rate_mean": 0.08548677712678909, "step": 75}
+{"stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.234375, "stability/repetition_rate_mean": 0.18307292461395264, "step": 80}
+{"stability/response_length_mean": 31.875, "stability/response_length_std": 33.0516357421875, "stability/response_length_var": 1092.4107666015625, "stability/token_entropy_mean": 4.90625, "stability/repetition_rate_mean": 0.09735831618309021, "step": 85}
+{"stability/response_length_mean": 33.625, "stability/response_length_std": 29.442134857177734, "stability/response_length_var": 866.8392944335938, "stability/token_entropy_mean": 3.7060546875, "stability/repetition_rate_mean": 0.2192234843969345, "step": 90}
+{"stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.751953125, "stability/repetition_rate_mean": 0.19374233484268188, "step": 95}
+{"stability/response_length_mean": 37.5, "stability/response_length_std": 29.640705108642578, "stability/response_length_var": 878.5714111328125, "stability/token_entropy_mean": 4.15625, "stability/repetition_rate_mean": 0.23773986101150513, "step": 100}
+{"stability/response_length_mean": 35.25, "stability/response_length_std": 31.494897842407227, "stability/response_length_var": 991.9285888671875, "stability/token_entropy_mean": 3.11328125, "stability/repetition_rate_mean": 0.189453125, "step": 105}
+{"stability/response_length_mean": 35.875, "stability/response_length_std": 29.811012268066406, "stability/response_length_var": 888.6964111328125, "stability/token_entropy_mean": 4.0595703125, "stability/repetition_rate_mean": 0.3042376935482025, "step": 110}
+{"stability/response_length_mean": 49.25, "stability/response_length_std": 27.316923141479492, "stability/response_length_var": 746.2142944335938, "stability/token_entropy_mean": 4.77734375, "stability/repetition_rate_mean": 0.3111979365348816, "step": 115}
+{"stability/response_length_mean": 29.25, "stability/response_length_std": 29.687660217285156, "stability/response_length_var": 881.3571166992188, "stability/token_entropy_mean": 4.5625, "stability/repetition_rate_mean": 0.18938209116458893, "step": 120}
+{"stability/response_length_mean": 15.25, "stability/response_length_std": 24.93276596069336, "stability/response_length_var": 621.6428833007812, "stability/token_entropy_mean": 3.640625, "stability/repetition_rate_mean": 0.16215278208255768, "step": 125}
+{"stability/response_length_mean": 38.0, "stability/response_length_std": 29.13515281677246, "stability/response_length_var": 848.8571166992188, "stability/token_entropy_mean": 3.99609375, "stability/repetition_rate_mean": 0.20181818306446075, "step": 130}
+{"stability/response_length_mean": 29.25, "stability/response_length_std": 29.936838150024414, "stability/response_length_var": 896.2142944335938, "stability/token_entropy_mean": 4.078125, "stability/repetition_rate_mean": 0.2230902761220932, "step": 135}
+{"stability/response_length_mean": 56.25, "stability/response_length_std": 21.920310974121094, "stability/response_length_var": 480.5, "stability/token_entropy_mean": 5.052734375, "stability/repetition_rate_mean": 0.220703125, "step": 140}
+{"stability/response_length_mean": 28.875, "stability/response_length_std": 29.73423957824707, "stability/response_length_var": 884.125, "stability/token_entropy_mean": 5.228515625, "stability/repetition_rate_mean": 0.26596763730049133, "step": 145}
+{"stability/response_length_mean": 35.25, "stability/response_length_std": 31.01957893371582, "stability/response_length_var": 962.2142944335938, "stability/token_entropy_mean": 4.66796875, "stability/repetition_rate_mean": 0.2291666716337204, "step": 150}
+{"stability/response_length_mean": 61.375, "stability/response_length_std": 7.424621105194092, "stability/response_length_var": 55.125, "stability/token_entropy_mean": 4.5703125, "stability/repetition_rate_mean": 0.3406159281730652, "step": 155}
+{"stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.44140625, "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": 4.8984375, "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": 3.5888671875, "stability/repetition_rate_mean": 0.3512404263019562, "step": 170}
+{"stability/response_length_mean": 27.125, "stability/response_length_std": 25.737340927124023, "stability/response_length_var": 662.4107055664062, "stability/token_entropy_mean": 4.8115234375, "stability/repetition_rate_mean": 0.2005799561738968, "step": 175}
+{"stability/response_length_mean": 34.75, "stability/response_length_std": 31.54928207397461, "stability/response_length_var": 995.3571166992188, "stability/token_entropy_mean": 3.6171875, "stability/repetition_rate_mean": 0.16484375298023224, "step": 180}
+{"stability/response_length_mean": 32.875, "stability/response_length_std": 33.30996322631836, "stability/response_length_var": 1109.5535888671875, "stability/token_entropy_mean": 3.716796875, "stability/repetition_rate_mean": 0.19921875, "step": 185}
+{"stability/response_length_mean": 18.25, "stability/response_length_std": 26.778188705444336, "stability/response_length_var": 717.0714111328125, "stability/token_entropy_mean": 4.83203125, "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": 5.16796875, "stability/repetition_rate_mean": 0.3953683078289032, "step": 195}

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/tokenizer.model

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/tokenizer_config.json

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

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "total_flos": 3.1861866257511875e+18,
+    "train_loss": 0.5200024314072668,
+    "train_runtime": 3159.9923,
+    "train_samples_per_second": 31.646,
+    "train_steps_per_second": 0.062
+}

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/trainer_log.jsonl

@@ -0,0 +1,40 @@
+{"current_steps": 5, "total_steps": 196, "loss": 0.5055, "lr": 4e-08, "epoch": 0.025591810620601407, "percentage": 2.55, "elapsed_time": "0:01:20", "remaining_time": "0:50:58", "rewards/chosen": 0.0004892349733333839, "logps/chosen": -439.0407072368421, "logits/chosen": -59717632.0, "rewards/rejected": 0.000401820366581281, "logps/rejected": -198.40538969494048, "logits/rejected": -50410139.428571425, "rewards/margins": 8.741460675210289e-05, "regularization/mmd": 0.00584295904263854, "regularization/rkhs_norm": 1.168591856956482, "kl": 0.005381508264690638}
+{"current_steps": 10, "total_steps": 196, "loss": 0.5087, "lr": 9e-08, "epoch": 0.05118362124120281, "percentage": 5.1, "elapsed_time": "0:02:41", "remaining_time": "0:49:55", "rewards/chosen": 0.0015674108615169277, "logps/chosen": -429.92283887987014, "logits/chosen": -59290737.03896104, "rewards/rejected": 0.0005207371011555913, "logps/rejected": -192.86274002259037, "logits/rejected": -49998308.240963854, "rewards/margins": 0.0010466737603613363, "regularization/mmd": 0.009019351564347744, "regularization/rkhs_norm": 1.803870439529419, "kl": 0.013140608556568623, "stability/response_length_mean": 46.125, "stability/response_length_std": 27.231481552124023, "stability/response_length_var": 741.5535888671875, "stability/token_entropy_mean": 3.3486328125, "stability/repetition_rate_mean": 0.3805190920829773}
+{"current_steps": 15, "total_steps": 196, "loss": 0.5087, "lr": 1.4e-07, "epoch": 0.07677543186180422, "percentage": 7.65, "elapsed_time": "0:04:01", "remaining_time": "0:48:28", "rewards/chosen": -5.2523883906277746e-05, "logps/chosen": -404.1313446969697, "logits/chosen": -56234691.49090909, "rewards/rejected": 0.0020233681125025596, "logps/rejected": -200.673828125, "logits/rejected": -51924714.52903226, "rewards/margins": -0.0020758919964088372, "regularization/mmd": 0.008868320845067501, "regularization/rkhs_norm": 1.773664116859436, "kl": 0.010814945213496685, "stability/response_length_mean": 34.0, "stability/response_length_std": 32.27117156982422, "stability/response_length_var": 1041.4285888671875, "stability/token_entropy_mean": 3.55078125, "stability/repetition_rate_mean": 0.2747395634651184}
+{"current_steps": 20, "total_steps": 196, "loss": 0.5094, "lr": 1.8999999999999998e-07, "epoch": 0.10236724248240563, "percentage": 10.2, "elapsed_time": "0:05:22", "remaining_time": "0:47:19", "rewards/chosen": 0.003181634098291397, "logps/chosen": -343.664892578125, "logits/chosen": -51667507.2, "rewards/rejected": 0.003590576723217964, "logps/rejected": -225.62744140625, "logits/rejected": -48054137.6, "rewards/margins": -0.0004089426249265669, "regularization/mmd": 0.00892628263682127, "regularization/rkhs_norm": 1.785256266593933, "kl": 0.01798226684331894, "stability/response_length_mean": 42.125, "stability/response_length_std": 27.01553726196289, "stability/response_length_var": 729.8392944335938, "stability/token_entropy_mean": 4.453125, "stability/repetition_rate_mean": 0.2040860652923584}
+{"current_steps": 25, "total_steps": 196, "loss": 0.5089, "lr": 2.4e-07, "epoch": 0.12795905310300704, "percentage": 12.76, "elapsed_time": "0:06:42", "remaining_time": "0:45:54", "rewards/chosen": -0.001145503541518902, "logps/chosen": -378.50290948275864, "logits/chosen": -56475732.74482758, "rewards/rejected": 0.0001408625500542777, "logps/rejected": -202.79145089285714, "logits/rejected": -52022096.45714286, "rewards/margins": -0.0012863660915731798, "regularization/mmd": 0.00950543861836195, "regularization/rkhs_norm": 1.901087760925293, "kl": 0.01633647456765175, "stability/response_length_mean": 43.25, "stability/response_length_std": 28.444181442260742, "stability/response_length_var": 809.0714111328125, "stability/token_entropy_mean": 4.7890625, "stability/repetition_rate_mean": 0.18034331500530243}
+{"current_steps": 30, "total_steps": 196, "loss": 0.5094, "lr": 2.9e-07, "epoch": 0.15355086372360843, "percentage": 15.31, "elapsed_time": "0:08:02", "remaining_time": "0:44:31", "rewards/chosen": -0.00012647900605361733, "logps/chosen": -468.5825293624161, "logits/chosen": -57726357.47651007, "rewards/rejected": -0.00074961188940974, "logps/rejected": -198.25539108187135, "logits/rejected": -51529943.578947365, "rewards/margins": 0.0006231328833561227, "regularization/mmd": 0.009396458975970745, "regularization/rkhs_norm": 1.8792918920516968, "kl": 0.013820863328874111, "stability/response_length_mean": 48.375, "stability/response_length_std": 28.93311309814453, "stability/response_length_var": 837.125, "stability/token_entropy_mean": 4.5625, "stability/repetition_rate_mean": 0.185546875}
+{"current_steps": 35, "total_steps": 196, "loss": 0.509, "lr": 3.4000000000000003e-07, "epoch": 0.17914267434420986, "percentage": 17.86, "elapsed_time": "0:09:24", "remaining_time": "0:43:16", "rewards/chosen": 0.0015070421409003343, "logps/chosen": -385.9214547072785, "logits/chosen": -57636902.88607595, "rewards/rejected": -0.00047891403054013665, "logps/rejected": -195.13126929012347, "logits/rejected": -53890566.32098766, "rewards/margins": 0.001985956171440471, "regularization/mmd": 0.009495963342487812, "regularization/rkhs_norm": 1.8991928100585938, "kl": 0.023283982649445534, "stability/response_length_mean": 50.5, "stability/response_length_std": 25.048524856567383, "stability/response_length_var": 627.4285888671875, "stability/token_entropy_mean": 5.46484375, "stability/repetition_rate_mean": 0.4045758843421936}
+{"current_steps": 40, "total_steps": 196, "loss": 0.5092, "lr": 3.8999999999999997e-07, "epoch": 0.20473448496481125, "percentage": 20.41, "elapsed_time": "0:10:45", "remaining_time": "0:41:56", "rewards/chosen": 0.002065117657184601, "logps/chosen": -421.881982421875, "logits/chosen": -56639987.2, "rewards/rejected": 0.0033827461302280426, "logps/rejected": -219.275439453125, "logits/rejected": -50230240.0, "rewards/margins": -0.0013176284730434416, "regularization/mmd": 0.009773665107786655, "regularization/rkhs_norm": 1.954732894897461, "kl": 0.026907920837402344, "stability/response_length_mean": 33.125, "stability/response_length_std": 33.01704025268555, "stability/response_length_var": 1090.125, "stability/token_entropy_mean": 4.9013671875, "stability/repetition_rate_mean": 0.13671875}
+{"current_steps": 45, "total_steps": 196, "loss": 0.5092, "lr": 4.3999999999999997e-07, "epoch": 0.23032629558541268, "percentage": 22.96, "elapsed_time": "0:12:05", "remaining_time": "0:40:35", "rewards/chosen": 0.0031585363333657282, "logps/chosen": -352.296875, "logits/chosen": -52695871.5704698, "rewards/rejected": 0.004328876908062495, "logps/rejected": -243.83088907163742, "logits/rejected": -50777555.0877193, "rewards/margins": -0.0011703405746967665, "regularization/mmd": 0.009779557585716248, "regularization/rkhs_norm": 1.955911636352539, "kl": 0.035909831523895264, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.87324333190918, "stability/response_length_var": 892.4107055664062, "stability/token_entropy_mean": 4.796875, "stability/repetition_rate_mean": 0.0974486693739891}
+{"current_steps": 50, "total_steps": 196, "loss": 0.5085, "lr": 4.9e-07, "epoch": 0.2559181062060141, "percentage": 25.51, "elapsed_time": "0:13:26", "remaining_time": "0:39:13", "rewards/chosen": 0.008534903494303659, "logps/chosen": -375.1137793624161, "logits/chosen": -54039015.946308725, "rewards/rejected": 0.004515389252824393, "logps/rejected": -180.1954495614035, "logits/rejected": -48296516.865497075, "rewards/margins": 0.0040195142414792655, "regularization/mmd": 0.008453378453850746, "regularization/rkhs_norm": 1.6906756162643433, "kl": 0.04191715270280838, "stability/response_length_mean": 53.375, "stability/response_length_std": 21.480472564697266, "stability/response_length_var": 461.41070556640625, "stability/token_entropy_mean": 4.115234375, "stability/repetition_rate_mean": 0.2932637929916382}
+{"current_steps": 55, "total_steps": 196, "loss": 0.5091, "lr": 5.4e-07, "epoch": 0.28150991682661547, "percentage": 28.06, "elapsed_time": "0:14:46", "remaining_time": "0:37:51", "rewards/chosen": 0.0030251911708286832, "logps/chosen": -344.4817546583851, "logits/chosen": -56959065.04347826, "rewards/rejected": 0.0043330676150771805, "logps/rejected": -230.40470224056602, "logits/rejected": -49578112.80503145, "rewards/margins": -0.0013078764442484973, "regularization/mmd": 0.008397291414439678, "regularization/rkhs_norm": 1.6794582605361938, "kl": 0.05445736274123192, "stability/response_length_mean": 22.375, "stability/response_length_std": 30.236862182617188, "stability/response_length_var": 914.2678833007812, "stability/token_entropy_mean": 4.515625, "stability/repetition_rate_mean": 0.11003988981246948}
+{"current_steps": 60, "total_steps": 196, "loss": 0.5085, "lr": 5.9e-07, "epoch": 0.30710172744721687, "percentage": 30.61, "elapsed_time": "0:16:07", "remaining_time": "0:36:33", "rewards/chosen": 0.011186000795075387, "logps/chosen": -359.6089488636364, "logits/chosen": -52672549.236363634, "rewards/rejected": 0.007287949900473318, "logps/rejected": -178.15403225806452, "logits/rejected": -47823535.07096774, "rewards/margins": 0.0038980508946020696, "regularization/mmd": 0.007710353937000036, "regularization/rkhs_norm": 1.5420708656311035, "kl": 0.05094662308692932, "stability/response_length_mean": 47.25, "stability/response_length_std": 28.684242248535156, "stability/response_length_var": 822.7857055664062, "stability/token_entropy_mean": 4.447265625, "stability/repetition_rate_mean": 0.1785714328289032}
+{"current_steps": 65, "total_steps": 196, "loss": 0.509, "lr": 6.4e-07, "epoch": 0.3326935380678183, "percentage": 33.16, "elapsed_time": "0:17:28", "remaining_time": "0:35:13", "rewards/chosen": 0.01387463846514302, "logps/chosen": -381.80806451612904, "logits/chosen": -57558207.58709677, "rewards/rejected": 0.007034623984134559, "logps/rejected": -183.5120146780303, "logits/rejected": -47281902.93333333, "rewards/margins": 0.006840014481008461, "regularization/mmd": 0.009121461771428585, "regularization/rkhs_norm": 1.8242921829223633, "kl": 0.07379233092069626, "stability/response_length_mean": 21.5, "stability/response_length_std": 26.635099411010742, "stability/response_length_var": 709.4285888671875, "stability/token_entropy_mean": 5.423828125, "stability/repetition_rate_mean": 0.18141163885593414}
+{"current_steps": 70, "total_steps": 196, "loss": 0.5089, "lr": 6.9e-07, "epoch": 0.3582853486884197, "percentage": 35.71, "elapsed_time": "0:18:48", "remaining_time": "0:33:51", "rewards/chosen": 0.011954639803978704, "logps/chosen": -379.00892137096776, "logits/chosen": -52772190.14193548, "rewards/rejected": 0.006403487378900701, "logps/rejected": -220.35814393939393, "logits/rejected": -49029666.13333333, "rewards/margins": 0.005551152425078003, "regularization/mmd": 0.008873376064002514, "regularization/rkhs_norm": 1.7746753692626953, "kl": 0.08646707981824875, "stability/response_length_mean": 33.875, "stability/response_length_std": 32.36151885986328, "stability/response_length_var": 1047.267822265625, "stability/token_entropy_mean": 5.73046875, "stability/repetition_rate_mean": 0.18359375}
+{"current_steps": 75, "total_steps": 196, "loss": 0.5095, "lr": 7.4e-07, "epoch": 0.3838771593090211, "percentage": 38.27, "elapsed_time": "0:20:09", "remaining_time": "0:32:31", "rewards/chosen": 0.013025604389809273, "logps/chosen": -314.1405458192568, "logits/chosen": -49507348.75675676, "rewards/rejected": 0.00713373062222503, "logps/rejected": -228.81352198401163, "logits/rejected": -47883686.697674416, "rewards/margins": 0.005891873767584243, "regularization/mmd": 0.00923239067196846, "regularization/rkhs_norm": 1.8464781045913696, "kl": 0.09753876179456711, "stability/response_length_mean": 22.75, "stability/response_length_std": 31.07479476928711, "stability/response_length_var": 965.6428833007812, "stability/token_entropy_mean": 2.150390625, "stability/repetition_rate_mean": 0.08548677712678909}
+{"current_steps": 80, "total_steps": 196, "loss": 0.5092, "lr": 7.9e-07, "epoch": 0.4094689699296225, "percentage": 40.82, "elapsed_time": "0:21:30", "remaining_time": "0:31:11", "rewards/chosen": 0.016511326594068513, "logps/chosen": -403.86501448675494, "logits/chosen": -58622508.0794702, "rewards/rejected": 0.006664444003584822, "logps/rejected": -200.98915957840237, "logits/rejected": -53452569.75147929, "rewards/margins": 0.009846882590483691, "regularization/mmd": 0.008787428960204124, "regularization/rkhs_norm": 1.7574856281280518, "kl": 0.1186995729804039, "stability/response_length_mean": 42.0, "stability/response_length_std": 30.481843948364258, "stability/response_length_var": 929.1428833007812, "stability/token_entropy_mean": 5.234375, "stability/repetition_rate_mean": 0.18307292461395264}
+{"current_steps": 85, "total_steps": 196, "loss": 0.5098, "lr": 8.399999999999999e-07, "epoch": 0.4350607805502239, "percentage": 43.37, "elapsed_time": "0:22:51", "remaining_time": "0:29:50", "rewards/chosen": 0.017374797021189044, "logps/chosen": -358.1587701612903, "logits/chosen": -51477778.16774193, "rewards/rejected": 0.009078372608531605, "logps/rejected": -230.33401988636365, "logits/rejected": -47139790.35151515, "rewards/margins": 0.00829642441265744, "regularization/mmd": 0.00979132205247879, "regularization/rkhs_norm": 1.9582645893096924, "kl": 0.12901714444160461, "stability/response_length_mean": 31.875, "stability/response_length_std": 33.0516357421875, "stability/response_length_var": 1092.4107666015625, "stability/token_entropy_mean": 4.90625, "stability/repetition_rate_mean": 0.09735831618309021}
+{"current_steps": 90, "total_steps": 196, "loss": 0.51, "lr": 8.9e-07, "epoch": 0.46065259117082535, "percentage": 45.92, "elapsed_time": "0:24:12", "remaining_time": "0:28:30", "rewards/chosen": 0.024664967488019895, "logps/chosen": -380.54271834935895, "logits/chosen": -58408185.43589743, "rewards/rejected": 0.01209283601946947, "logps/rejected": -187.10587366615854, "logits/rejected": -50558657.56097561, "rewards/margins": 0.012572131468550424, "regularization/mmd": 0.011231121607124805, "regularization/rkhs_norm": 2.2462244033813477, "kl": 0.17196393013000488, "stability/response_length_mean": 33.625, "stability/response_length_std": 29.442134857177734, "stability/response_length_var": 866.8392944335938, "stability/token_entropy_mean": 3.7060546875, "stability/repetition_rate_mean": 0.2192234843969345}
+{"current_steps": 95, "total_steps": 196, "loss": 0.5107, "lr": 9.399999999999999e-07, "epoch": 0.48624440179142675, "percentage": 48.47, "elapsed_time": "0:25:33", "remaining_time": "0:27:09", "rewards/chosen": 0.022749973248831833, "logps/chosen": -314.7658722310127, "logits/chosen": -50378544.60759494, "rewards/rejected": 0.0066354083426204725, "logps/rejected": -207.07737750771605, "logits/rejected": -48969532.04938272, "rewards/margins": 0.01611456490621136, "regularization/mmd": 0.012432265095412731, "regularization/rkhs_norm": 2.48645281791687, "kl": 0.1672627478837967, "stability/response_length_mean": 27.0, "stability/response_length_std": 31.1264705657959, "stability/response_length_var": 968.8571166992188, "stability/token_entropy_mean": 3.751953125, "stability/repetition_rate_mean": 0.19374233484268188}
+{"current_steps": 100, "total_steps": 196, "loss": 0.511, "lr": 9.9e-07, "epoch": 0.5118362124120281, "percentage": 51.02, "elapsed_time": "0:26:53", "remaining_time": "0:25:49", "rewards/chosen": 0.03283950444814321, "logps/chosen": -359.54178103885135, "logits/chosen": -54725538.5945946, "rewards/rejected": 0.02023870861807535, "logps/rejected": -216.3764989098837, "logits/rejected": -50922793.674418606, "rewards/margins": 0.01260079583006786, "regularization/mmd": 0.01211935468018055, "regularization/rkhs_norm": 2.423870801925659, "kl": 0.20016632974147797, "stability/response_length_mean": 37.5, "stability/response_length_std": 29.640705108642578, "stability/response_length_var": 878.5714111328125, "stability/token_entropy_mean": 4.15625, "stability/repetition_rate_mean": 0.23773986101150513}
+{"current_steps": 105, "total_steps": 196, "loss": 0.5118, "lr": 9.999815554351224e-07, "epoch": 0.5374280230326296, "percentage": 53.57, "elapsed_time": "0:28:14", "remaining_time": "0:24:28", "rewards/chosen": 0.04110269374158009, "logps/chosen": -399.49143448795184, "logits/chosen": -59377960.09638554, "rewards/rejected": 0.016051369828063173, "logps/rejected": -209.84331879058442, "logits/rejected": -53795347.948051944, "rewards/margins": 0.02505132391351692, "regularization/mmd": 0.015834027901291847, "regularization/rkhs_norm": 3.1668055057525635, "kl": 0.22721095383167267, "stability/response_length_mean": 35.25, "stability/response_length_std": 31.494897842407227, "stability/response_length_var": 991.9285888671875, "stability/token_entropy_mean": 3.11328125, "stability/repetition_rate_mean": 0.189453125}
+{"current_steps": 110, "total_steps": 196, "loss": 0.5132, "lr": 9.999066267225444e-07, "epoch": 0.5630198336532309, "percentage": 56.12, "elapsed_time": "0:29:36", "remaining_time": "0:23:08", "rewards/chosen": 0.03888099164847868, "logps/chosen": -378.17446347891564, "logits/chosen": -57376342.361445785, "rewards/rejected": 0.021522568417833995, "logps/rejected": -205.87290736607142, "logits/rejected": -53605545.55844156, "rewards/margins": 0.01735842323064468, "regularization/mmd": 0.016469214111566544, "regularization/rkhs_norm": 3.2938427925109863, "kl": 0.2592513859272003, "stability/response_length_mean": 35.875, "stability/response_length_std": 29.811012268066406, "stability/response_length_var": 888.6964111328125, "stability/token_entropy_mean": 4.0595703125, "stability/repetition_rate_mean": 0.3042376935482025}
+{"current_steps": 115, "total_steps": 196, "loss": 0.5148, "lr": 9.997740697079592e-07, "epoch": 0.5886116442738324, "percentage": 58.67, "elapsed_time": "0:30:57", "remaining_time": "0:21:48", "rewards/chosen": 0.051097362856321694, "logps/chosen": -364.08156151107596, "logits/chosen": -60592257.62025317, "rewards/rejected": 0.022696312562918958, "logps/rejected": -200.96959394290124, "logits/rejected": -53408388.74074074, "rewards/margins": 0.028401050293402737, "regularization/mmd": 0.01740214042365551, "regularization/rkhs_norm": 3.480428457260132, "kl": 0.31319910287857056, "stability/response_length_mean": 49.25, "stability/response_length_std": 27.316923141479492, "stability/response_length_var": 746.2142944335938, "stability/token_entropy_mean": 4.77734375, "stability/repetition_rate_mean": 0.3111979365348816}
+{"current_steps": 120, "total_steps": 196, "loss": 0.5155, "lr": 9.995838996722914e-07, "epoch": 0.6142034548944337, "percentage": 61.22, "elapsed_time": "0:32:17", "remaining_time": "0:20:26", "rewards/chosen": 0.05486513274947023, "logps/chosen": -383.69628267973854, "logits/chosen": -57730777.51633987, "rewards/rejected": 0.024559349357011075, "logps/rejected": -211.9707616017964, "logits/rejected": -54345593.1017964, "rewards/margins": 0.030305783392459158, "regularization/mmd": 0.020150119438767433, "regularization/rkhs_norm": 4.03002405166626, "kl": 0.34892186522483826, "stability/response_length_mean": 29.25, "stability/response_length_std": 29.687660217285156, "stability/response_length_var": 881.3571166992188, "stability/token_entropy_mean": 4.5625, "stability/repetition_rate_mean": 0.18938209116458893}
+{"current_steps": 125, "total_steps": 196, "loss": 0.5184, "lr": 9.993361385379874e-07, "epoch": 0.6397952655150352, "percentage": 63.78, "elapsed_time": "0:33:38", "remaining_time": "0:19:06", "rewards/chosen": 0.058781075150999305, "logps/chosen": -416.7914704623288, "logits/chosen": -60671656.32876712, "rewards/rejected": 0.02654672491139379, "logps/rejected": -216.10171964798852, "logits/rejected": -55648673.83908046, "rewards/margins": 0.03223435023960551, "regularization/mmd": 0.023098159581422806, "regularization/rkhs_norm": 4.619632244110107, "kl": 0.3908672630786896, "stability/response_length_mean": 15.25, "stability/response_length_std": 24.93276596069336, "stability/response_length_var": 621.6428833007812, "stability/token_entropy_mean": 3.640625, "stability/repetition_rate_mean": 0.16215278208255768}
+{"current_steps": 130, "total_steps": 196, "loss": 0.5201, "lr": 9.99030814866488e-07, "epoch": 0.6653870761356366, "percentage": 66.33, "elapsed_time": "0:34:58", "remaining_time": "0:17:45", "rewards/chosen": 0.06725460175544985, "logps/chosen": -391.98538306451616, "logits/chosen": -57380811.1483871, "rewards/rejected": 0.026962774450128728, "logps/rejected": -183.6492897727273, "logits/rejected": -53835955.97575758, "rewards/margins": 0.040291827305321125, "regularization/mmd": 0.02793542854487896, "regularization/rkhs_norm": 5.587085723876953, "kl": 0.4510687291622162, "stability/response_length_mean": 38.0, "stability/response_length_std": 29.13515281677246, "stability/response_length_var": 848.8571166992188, "stability/token_entropy_mean": 3.99609375, "stability/repetition_rate_mean": 0.20181818306446075}
+{"current_steps": 135, "total_steps": 196, "loss": 0.5209, "lr": 9.986679638549358e-07, "epoch": 0.690978886756238, "percentage": 68.88, "elapsed_time": "0:36:18", "remaining_time": "0:16:24", "rewards/chosen": 0.08035813080021209, "logps/chosen": -401.01260544478527, "logits/chosen": -59853874.25766871, "rewards/rejected": 0.03506741250396534, "logps/rejected": -247.52005374203821, "logits/rejected": -55008706.03821656, "rewards/margins": 0.04529071829624675, "regularization/mmd": 0.03274526074528694, "regularization/rkhs_norm": 6.549052715301514, "kl": 0.49744081497192383, "stability/response_length_mean": 29.25, "stability/response_length_std": 29.936838150024414, "stability/response_length_var": 896.2142944335938, "stability/token_entropy_mean": 4.078125, "stability/repetition_rate_mean": 0.2230902761220932}
+{"current_steps": 140, "total_steps": 196, "loss": 0.523, "lr": 9.982476273321175e-07, "epoch": 0.7165706973768394, "percentage": 71.43, "elapsed_time": "0:37:39", "remaining_time": "0:15:03", "rewards/chosen": 0.08111698844216086, "logps/chosen": -366.3785511363636, "logits/chosen": -57075643.733333334, "rewards/rejected": 0.012395195807180097, "logps/rejected": -189.9105216733871, "logits/rejected": -50076533.26451613, "rewards/margins": 0.06872179263498077, "regularization/mmd": 0.03410874679684639, "regularization/rkhs_norm": 6.821749210357666, "kl": 0.4897715151309967, "stability/response_length_mean": 56.25, "stability/response_length_std": 21.920310974121094, "stability/response_length_var": 480.5, "stability/token_entropy_mean": 5.052734375, "stability/repetition_rate_mean": 0.220703125}
+{"current_steps": 145, "total_steps": 196, "loss": 0.5245, "lr": 9.977698537536417e-07, "epoch": 0.7421625079974408, "percentage": 73.98, "elapsed_time": "0:38:59", "remaining_time": "0:13:42", "rewards/chosen": 0.07939604104283345, "logps/chosen": -329.4778567394578, "logits/chosen": -51149342.84337349, "rewards/rejected": 0.03412691648904379, "logps/rejected": -199.1466492491883, "logits/rejected": -51664423.896103896, "rewards/margins": 0.045269124553789654, "regularization/mmd": 0.029787013307213783, "regularization/rkhs_norm": 5.95740270614624, "kl": 0.5653125047683716, "stability/response_length_mean": 28.875, "stability/response_length_std": 29.73423957824707, "stability/response_length_var": 884.125, "stability/token_entropy_mean": 5.228515625, "stability/repetition_rate_mean": 0.26596763730049133}
+{"current_steps": 150, "total_steps": 196, "loss": 0.5281, "lr": 9.972346981963546e-07, "epoch": 0.7677543186180422, "percentage": 76.53, "elapsed_time": "0:40:20", "remaining_time": "0:12:22", "rewards/chosen": 0.11093491946949678, "logps/chosen": -396.82509191176473, "logits/chosen": -55421066.54117647, "rewards/rejected": 0.03602298100789388, "logps/rejected": -207.365703125, "logits/rejected": -50181959.68, "rewards/margins": 0.0749119384616029, "regularization/mmd": 0.036661840975284576, "regularization/rkhs_norm": 7.33236837387085, "kl": 0.6396333575248718, "stability/response_length_mean": 35.25, "stability/response_length_std": 31.01957893371582, "stability/response_length_var": 962.2142944335938, "stability/token_entropy_mean": 4.66796875, "stability/repetition_rate_mean": 0.2291666716337204}
+{"current_steps": 155, "total_steps": 196, "loss": 0.5318, "lr": 9.966422223519885e-07, "epoch": 0.7933461292386437, "percentage": 79.08, "elapsed_time": "0:41:42", "remaining_time": "0:11:01", "rewards/chosen": 0.09180937935324276, "logps/chosen": -360.07118566176473, "logits/chosen": -54015186.823529415, "rewards/rejected": 0.025360565185546875, "logps/rejected": -191.05200520833333, "logits/rejected": -49345460.906666666, "rewards/margins": 0.0664488141676959, "regularization/mmd": 0.0417025052011013, "regularization/rkhs_norm": 8.340499877929688, "kl": 0.6517454981803894, "stability/response_length_mean": 61.375, "stability/response_length_std": 7.424621105194092, "stability/response_length_var": 55.125, "stability/token_entropy_mean": 4.5703125, "stability/repetition_rate_mean": 0.3406159281730652}
+{"current_steps": 160, "total_steps": 196, "loss": 0.5348, "lr": 9.959924945200523e-07, "epoch": 0.818937939859245, "percentage": 81.63, "elapsed_time": "0:43:03", "remaining_time": "0:09:41", "rewards/chosen": 0.13368967601231166, "logps/chosen": -415.55845424107144, "logits/chosen": -55251065.9047619, "rewards/rejected": 0.043035620137264856, "logps/rejected": -186.46698961759867, "logits/rejected": -48947941.05263158, "rewards/margins": 0.0906540558750468, "regularization/mmd": 0.04304032027721405, "regularization/rkhs_norm": 8.608064651489258, "kl": 0.7206474542617798, "stability/response_length_mean": 32.5, "stability/response_length_std": 30.528675079345703, "stability/response_length_var": 932.0, "stability/token_entropy_mean": 4.44140625, "stability/repetition_rate_mean": 0.30851560831069946}
+{"current_steps": 165, "total_steps": 196, "loss": 0.5371, "lr": 9.952855895999568e-07, "epoch": 0.8445297504798465, "percentage": 84.18, "elapsed_time": "0:44:23", "remaining_time": "0:08:20", "rewards/chosen": 0.11550897839425624, "logps/chosen": -369.7616738505747, "logits/chosen": -54404996.4137931, "rewards/rejected": 0.04401412728714616, "logps/rejected": -212.15198255565068, "logits/rejected": -49794777.42465753, "rewards/margins": 0.07149485110711007, "regularization/mmd": 0.04163478687405586, "regularization/rkhs_norm": 8.326957702636719, "kl": 0.7607746124267578, "stability/response_length_mean": 22.5, "stability/response_length_std": 28.33977508544922, "stability/response_length_var": 803.1428833007812, "stability/token_entropy_mean": 4.8984375, "stability/repetition_rate_mean": 0.2411106526851654}
+{"current_steps": 170, "total_steps": 196, "loss": 0.5386, "lr": 9.9452158908238e-07, "epoch": 0.8701215611004478, "percentage": 86.73, "elapsed_time": "0:45:45", "remaining_time": "0:06:59", "rewards/chosen": 0.16517957427797939, "logps/chosen": -439.69008875739644, "logits/chosen": -59132631.10059172, "rewards/rejected": 0.06983007974182533, "logps/rejected": -190.55952504139074, "logits/rejected": -52570654.51655629, "rewards/margins": 0.09534949453615406, "regularization/mmd": 0.0507260225713253, "regularization/rkhs_norm": 10.145204544067383, "kl": 0.924032986164093, "stability/response_length_mean": 44.125, "stability/response_length_std": 28.43256378173828, "stability/response_length_var": 808.4107055664062, "stability/token_entropy_mean": 3.5888671875, "stability/repetition_rate_mean": 0.3512404263019562}
+{"current_steps": 175, "total_steps": 196, "loss": 0.5416, "lr": 9.937005810398745e-07, "epoch": 0.8957133717210493, "percentage": 89.29, "elapsed_time": "0:47:05", "remaining_time": "0:05:39", "rewards/chosen": 0.15526514168245248, "logps/chosen": -354.88453030873495, "logits/chosen": -56654465.54216868, "rewards/rejected": 0.050466940000459745, "logps/rejected": -194.3028612012987, "logits/rejected": -48027790.96103896, "rewards/margins": 0.10479820168199273, "regularization/mmd": 0.06023603305220604, "regularization/rkhs_norm": 12.047205924987793, "kl": 0.9547041058540344, "stability/response_length_mean": 27.125, "stability/response_length_std": 25.737340927124023, "stability/response_length_var": 662.4107055664062, "stability/token_entropy_mean": 4.8115234375, "stability/repetition_rate_mean": 0.2005799561738968}
+{"current_steps": 180, "total_steps": 196, "loss": 0.5458, "lr": 9.928226601167138e-07, "epoch": 0.9213051823416507, "percentage": 91.84, "elapsed_time": "0:48:25", "remaining_time": "0:04:18", "rewards/chosen": 0.17411304544086106, "logps/chosen": -374.4046922929448, "logits/chosen": -52652606.82208589, "rewards/rejected": 0.07555588643262341, "logps/rejected": -210.54264530254778, "logits/rejected": -47977778.5477707, "rewards/margins": 0.09855715900823765, "regularization/mmd": 0.062359750270843506, "regularization/rkhs_norm": 12.471949577331543, "kl": 1.0281428098678589, "stability/response_length_mean": 34.75, "stability/response_length_std": 31.54928207397461, "stability/response_length_var": 995.3571166992188, "stability/token_entropy_mean": 3.6171875, "stability/repetition_rate_mean": 0.16484375298023224}
+{"current_steps": 185, "total_steps": 196, "loss": 0.5519, "lr": 9.918879275179817e-07, "epoch": 0.946896992962252, "percentage": 94.39, "elapsed_time": "0:49:45", "remaining_time": "0:02:57", "rewards/chosen": 0.16808104809419608, "logps/chosen": -310.3205536265432, "logits/chosen": -54606911.209876545, "rewards/rejected": 0.049601835540578336, "logps/rejected": -196.54314181170886, "logits/rejected": -49525303.088607594, "rewards/margins": 0.11847921255361774, "regularization/mmd": 0.07445289939641953, "regularization/rkhs_norm": 14.890579223632812, "kl": 1.0762888193130493, "stability/response_length_mean": 32.875, "stability/response_length_std": 33.30996322631836, "stability/response_length_var": 1109.5535888671875, "stability/token_entropy_mean": 3.716796875, "stability/repetition_rate_mean": 0.19921875}
+{"current_steps": 190, "total_steps": 196, "loss": 0.5607, "lr": 9.90896490997906e-07, "epoch": 0.9724888035828535, "percentage": 96.94, "elapsed_time": "0:51:05", "remaining_time": "0:01:36", "rewards/chosen": 0.21127086342451815, "logps/chosen": -390.29261788922156, "logits/chosen": -53083148.263473056, "rewards/rejected": 0.07451764275045956, "logps/rejected": -210.45340584150327, "logits/rejected": -49686661.85620915, "rewards/margins": 0.1367532206740586, "regularization/mmd": 0.0810161754488945, "regularization/rkhs_norm": 16.203235626220703, "kl": 1.1190913915634155, "stability/response_length_mean": 18.25, "stability/response_length_std": 26.778188705444336, "stability/response_length_var": 717.0714111328125, "stability/token_entropy_mean": 4.83203125, "stability/repetition_rate_mean": 0.19138340651988983}
+{"current_steps": 195, "total_steps": 196, "loss": 0.5603, "lr": 9.898484648474362e-07, "epoch": 0.9980806142034548, "percentage": 99.49, "elapsed_time": "0:52:26", "remaining_time": "0:00:16", "rewards/chosen": 0.1934424861784904, "logps/chosen": -390.8505040322581, "logits/chosen": -60020286.76129032, "rewards/rejected": 0.08274661555434719, "logps/rejected": -197.09185606060606, "logits/rejected": -53779549.09090909, "rewards/margins": 0.11069587062414322, "regularization/mmd": 0.06844492256641388, "regularization/rkhs_norm": 13.688984870910645, "kl": 1.1464111804962158, "stability/response_length_mean": 49.0, "stability/response_length_std": 27.82085609436035, "stability/response_length_var": 774.0, "stability/token_entropy_mean": 5.16796875, "stability/repetition_rate_mean": 0.3953683078289032}
+{"current_steps": 196, "total_steps": 196, "epoch": 1.0, "percentage": 100.0, "elapsed_time": "0:52:39", "remaining_time": "0:00:00"}

mistral7b_kto_mmd_rho0.005_lambda_kr0.01/trainer_state.json

@@ -0,0 +1,974 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 196,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.025591810620601407,
+      "grad_norm": 0.5693546533584595,
+      "kl": 0.005381508264690638,
+      "learning_rate": 4e-08,
+      "logits/chosen": -59717632.0,
+      "logits/rejected": -50410139.428571425,
+      "logps/chosen": -439.0407072368421,
+      "logps/rejected": -198.40538969494048,
+      "loss": 0.5055,
+      "loss/base_kto": 0.49999600648880005,
+      "metrics/advantage_var": 0.04680101200938225,
+      "regularization/mmd": 0.00584295904263854,
+      "regularization/rkhs_norm": 1.168591856956482,
+      "rewards/chosen": 0.0004892349733333839,
+      "rewards/margins": 8.741460675210289e-05,
+      "rewards/rejected": 0.000401820366581281,
+      "step": 5
+    },
+    {
+      "epoch": 0.05118362124120281,
+      "grad_norm": 0.6020964980125427,
+      "kl": 0.013140608556568623,
+      "learning_rate": 9e-08,
+      "logits/chosen": -59290737.03896104,
+      "logits/rejected": -49998308.240963854,
+      "logps/chosen": -429.92283887987014,
+      "logps/rejected": -192.86274002259037,
+      "loss": 0.5087,
+      "loss/base_kto": 0.4998807907104492,
+      "metrics/advantage_var": 0.06702536344528198,
+      "regularization/mmd": 0.009019351564347744,
+      "regularization/rkhs_norm": 1.803870439529419,
+      "rewards/chosen": 0.0015674108615169277,
+      "rewards/margins": 0.0010466737603613363,
+      "rewards/rejected": 0.0005207371011555913,
+      "stability/repetition_rate_mean": 0.3805190920829773,
+      "stability/response_length_mean": 46.125,
+      "stability/response_length_std": 27.231481552124023,
+      "stability/response_length_var": 741.5535888671875,
+      "stability/token_entropy_mean": 3.3486328125,
+      "step": 10
+    },
+    {
+      "epoch": 0.07677543186180422,
+      "grad_norm": 0.6099647283554077,
+      "kl": 0.010814945213496685,
+      "learning_rate": 1.4e-07,
+      "logits/chosen": -56234691.49090909,
+      "logits/rejected": -51924714.52903226,
+      "logps/chosen": -404.1313446969697,
+      "logps/rejected": -200.673828125,
+      "loss": 0.5087,
+      "loss/base_kto": 0.5002461671829224,
+      "metrics/advantage_var": 0.0660475566983223,
+      "regularization/mmd": 0.008868320845067501,
+      "regularization/rkhs_norm": 1.773664116859436,
+      "rewards/chosen": -5.2523883906277746e-05,
+      "rewards/margins": -0.0020758919964088372,
+      "rewards/rejected": 0.0020233681125025596,
+      "stability/repetition_rate_mean": 0.2747395634651184,
+      "stability/response_length_mean": 34.0,
+      "stability/response_length_std": 32.27117156982422,
+      "stability/response_length_var": 1041.4285888671875,
+      "stability/token_entropy_mean": 3.55078125,
+      "step": 15
+    },
+    {
+      "epoch": 0.10236724248240563,
+      "grad_norm": 0.618466854095459,
+      "kl": 0.01798226684331894,
+      "learning_rate": 1.8999999999999998e-07,
+      "logits/chosen": -51667507.2,
+      "logits/rejected": -48054137.6,
+      "logps/chosen": -343.664892578125,
+      "logps/rejected": -225.62744140625,
+      "loss": 0.5094,
+      "loss/base_kto": 0.5001111030578613,
+      "metrics/advantage_var": 0.06671873480081558,
+      "regularization/mmd": 0.00892628263682127,
+      "regularization/rkhs_norm": 1.785256266593933,
+      "rewards/chosen": 0.003181634098291397,
+      "rewards/margins": -0.0004089426249265669,
+      "rewards/rejected": 0.003590576723217964,
+      "stability/repetition_rate_mean": 0.2040860652923584,
+      "stability/response_length_mean": 42.125,
+      "stability/response_length_std": 27.01553726196289,
+      "stability/response_length_var": 729.8392944335938,
+      "stability/token_entropy_mean": 4.453125,
+      "step": 20
+    },
+    {
+      "epoch": 0.12795905310300704,
+      "grad_norm": 0.5777862668037415,
+      "kl": 0.01633647456765175,
+      "learning_rate": 2.4e-07,
+      "logits/chosen": -56475732.74482758,
+      "logits/rejected": -52022096.45714286,
+      "logps/chosen": -378.50290948275864,
+      "logps/rejected": -202.79145089285714,
+      "loss": 0.5089,
+      "loss/base_kto": 0.5001418590545654,
+      "metrics/advantage_var": 0.0671335980296135,
+      "regularization/mmd": 0.00950543861836195,
+      "regularization/rkhs_norm": 1.901087760925293,
+      "rewards/chosen": -0.001145503541518902,
+      "rewards/margins": -0.0012863660915731798,
+      "rewards/rejected": 0.0001408625500542777,
+      "stability/repetition_rate_mean": 0.18034331500530243,
+      "stability/response_length_mean": 43.25,
+      "stability/response_length_std": 28.444181442260742,
+      "stability/response_length_var": 809.0714111328125,
+      "stability/token_entropy_mean": 4.7890625,
+      "step": 25
+    },
+    {
+      "epoch": 0.15355086372360843,
+      "grad_norm": 0.6039624214172363,
+      "kl": 0.013820863328874111,
+      "learning_rate": 2.9e-07,
+      "logits/chosen": -57726357.47651007,
+      "logits/rejected": -51529943.578947365,
+      "logps/chosen": -468.5825293624161,
+      "logps/rejected": -198.25539108187135,
+      "loss": 0.5094,
+      "loss/base_kto": 0.49989691376686096,
+      "metrics/advantage_var": 0.07160170376300812,
+      "regularization/mmd": 0.009396458975970745,
+      "regularization/rkhs_norm": 1.8792918920516968,
+      "rewards/chosen": -0.00012647900605361733,
+      "rewards/margins": 0.0006231328833561227,
+      "rewards/rejected": -0.00074961188940974,
+      "stability/repetition_rate_mean": 0.185546875,
+      "stability/response_length_mean": 48.375,
+      "stability/response_length_std": 28.93311309814453,
+      "stability/response_length_var": 837.125,
+      "stability/token_entropy_mean": 4.5625,
+      "step": 30
+    },
+    {
+      "epoch": 0.17914267434420986,
+      "grad_norm": 0.6789125204086304,
+      "kl": 0.023283982649445534,
+      "learning_rate": 3.4000000000000003e-07,
+      "logits/chosen": -57636902.88607595,
+      "logits/rejected": -53890566.32098766,
+      "logps/chosen": -385.9214547072785,
+      "logps/rejected": -195.13126929012347,
+      "loss": 0.509,
+      "loss/base_kto": 0.4997556209564209,
+      "metrics/advantage_var": 0.0707424208521843,
+      "regularization/mmd": 0.009495963342487812,
+      "regularization/rkhs_norm": 1.8991928100585938,
+      "rewards/chosen": 0.0015070421409003343,
+      "rewards/margins": 0.001985956171440471,
+      "rewards/rejected": -0.00047891403054013665,
+      "stability/repetition_rate_mean": 0.4045758843421936,
+      "stability/response_length_mean": 50.5,
+      "stability/response_length_std": 25.048524856567383,
+      "stability/response_length_var": 627.4285888671875,
+      "stability/token_entropy_mean": 5.46484375,
+      "step": 35
+    },
+    {
+      "epoch": 0.20473448496481125,
+      "grad_norm": 0.5862559676170349,
+      "kl": 0.026907920837402344,
+      "learning_rate": 3.8999999999999997e-07,
+      "logits/chosen": -56639987.2,
+      "logits/rejected": -50230240.0,
+      "logps/chosen": -421.881982421875,
+      "logps/rejected": -219.275439453125,
+      "loss": 0.5092,
+      "loss/base_kto": 0.5001746416091919,
+      "metrics/advantage_var": 0.0762791857123375,
+      "regularization/mmd": 0.009773665107786655,
+      "regularization/rkhs_norm": 1.954732894897461,
+      "rewards/chosen": 0.002065117657184601,
+      "rewards/margins": -0.0013176284730434416,
+      "rewards/rejected": 0.0033827461302280426,
+      "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": 4.9013671875,
+      "step": 40
+    },
+    {
+      "epoch": 0.23032629558541268,
+      "grad_norm": 0.6387529969215393,
+      "kl": 0.035909831523895264,
+      "learning_rate": 4.3999999999999997e-07,
+      "logits/chosen": -52695871.5704698,
+      "logits/rejected": -50777555.0877193,
+      "logps/chosen": -352.296875,
+      "logps/rejected": -243.83088907163742,
+      "loss": 0.5092,
+      "loss/base_kto": 0.5001186728477478,
+      "metrics/advantage_var": 0.07618097215890884,
+      "regularization/mmd": 0.009779557585716248,
+      "regularization/rkhs_norm": 1.955911636352539,
+      "rewards/chosen": 0.0031585363333657282,
+      "rewards/margins": -0.0011703405746967665,
+      "rewards/rejected": 0.004328876908062495,
+      "stability/repetition_rate_mean": 0.0974486693739891,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.87324333190918,
+      "stability/response_length_var": 892.4107055664062,
+      "stability/token_entropy_mean": 4.796875,
+      "step": 45
+    },
+    {
+      "epoch": 0.2559181062060141,
+      "grad_norm": 0.5790932774543762,
+      "kl": 0.04191715270280838,
+      "learning_rate": 4.9e-07,
+      "logits/chosen": -54039015.946308725,
+      "logits/rejected": -48296516.865497075,
+      "logps/chosen": -375.1137793624161,
+      "logps/rejected": -180.1954495614035,
+      "loss": 0.5085,
+      "loss/base_kto": 0.49944159388542175,
+      "metrics/advantage_var": 0.05880240350961685,
+      "regularization/mmd": 0.008453378453850746,
+      "regularization/rkhs_norm": 1.6906756162643433,
+      "rewards/chosen": 0.008534903494303659,
+      "rewards/margins": 0.0040195142414792655,
+      "rewards/rejected": 0.004515389252824393,
+      "stability/repetition_rate_mean": 0.2932637929916382,
+      "stability/response_length_mean": 53.375,
+      "stability/response_length_std": 21.480472564697266,
+      "stability/response_length_var": 461.41070556640625,
+      "stability/token_entropy_mean": 4.115234375,
+      "step": 50
+    },
+    {
+      "epoch": 0.28150991682661547,
+      "grad_norm": 0.6598390936851501,
+      "kl": 0.05445736274123192,
+      "learning_rate": 5.4e-07,
+      "logits/chosen": -56959065.04347826,
+      "logits/rejected": -49578112.80503145,
+      "logps/chosen": -344.4817546583851,
+      "logps/rejected": -230.40470224056602,
+      "loss": 0.5091,
+      "loss/base_kto": 0.5001540184020996,
+      "metrics/advantage_var": 0.0567297525703907,
+      "regularization/mmd": 0.008397291414439678,
+      "regularization/rkhs_norm": 1.6794582605361938,
+      "rewards/chosen": 0.0030251911708286832,
+      "rewards/margins": -0.0013078764442484973,
+      "rewards/rejected": 0.0043330676150771805,
+      "stability/repetition_rate_mean": 0.11003988981246948,
+      "stability/response_length_mean": 22.375,
+      "stability/response_length_std": 30.236862182617188,
+      "stability/response_length_var": 914.2678833007812,
+      "stability/token_entropy_mean": 4.515625,
+      "step": 55
+    },
+    {
+      "epoch": 0.30710172744721687,
+      "grad_norm": 0.5815622210502625,
+      "kl": 0.05094662308692932,
+      "learning_rate": 5.9e-07,
+      "logits/chosen": -52672549.236363634,
+      "logits/rejected": -47823535.07096774,
+      "logps/chosen": -359.6089488636364,
+      "logps/rejected": -178.15403225806452,
+      "loss": 0.5085,
+      "loss/base_kto": 0.49942919611930847,
+      "metrics/advantage_var": 0.05574983358383179,
+      "regularization/mmd": 0.007710353937000036,
+      "regularization/rkhs_norm": 1.5420708656311035,
+      "rewards/chosen": 0.011186000795075387,
+      "rewards/margins": 0.0038980508946020696,
+      "rewards/rejected": 0.007287949900473318,
+      "stability/repetition_rate_mean": 0.1785714328289032,
+      "stability/response_length_mean": 47.25,
+      "stability/response_length_std": 28.684242248535156,
+      "stability/response_length_var": 822.7857055664062,
+      "stability/token_entropy_mean": 4.447265625,
+      "step": 60
+    },
+    {
+      "epoch": 0.3326935380678183,
+      "grad_norm": 0.5738525390625,
+      "kl": 0.07379233092069626,
+      "learning_rate": 6.4e-07,
+      "logits/chosen": -57558207.58709677,
+      "logits/rejected": -47281902.93333333,
+      "logps/chosen": -381.80806451612904,
+      "logps/rejected": -183.5120146780303,
+      "loss": 0.509,
+      "loss/base_kto": 0.49920693039894104,
+      "metrics/advantage_var": 0.07559790462255478,
+      "regularization/mmd": 0.009121461771428585,
+      "regularization/rkhs_norm": 1.8242921829223633,
+      "rewards/chosen": 0.01387463846514302,
+      "rewards/margins": 0.006840014481008461,
+      "rewards/rejected": 0.007034623984134559,
+      "stability/repetition_rate_mean": 0.18141163885593414,
+      "stability/response_length_mean": 21.5,
+      "stability/response_length_std": 26.635099411010742,
+      "stability/response_length_var": 709.4285888671875,
+      "stability/token_entropy_mean": 5.423828125,
+      "step": 65
+    },
+    {
+      "epoch": 0.3582853486884197,
+      "grad_norm": 0.5529053211212158,
+      "kl": 0.08646707981824875,
+      "learning_rate": 6.9e-07,
+      "logits/chosen": -52772190.14193548,
+      "logits/rejected": -49029666.13333333,
+      "logps/chosen": -379.00892137096776,
+      "logps/rejected": -220.35814393939393,
+      "loss": 0.5089,
+      "loss/base_kto": 0.4993232190608978,
+      "metrics/advantage_var": 0.07030369341373444,
+      "regularization/mmd": 0.008873376064002514,
+      "regularization/rkhs_norm": 1.7746753692626953,
+      "rewards/chosen": 0.011954639803978704,
+      "rewards/margins": 0.005551152425078003,
+      "rewards/rejected": 0.006403487378900701,
+      "stability/repetition_rate_mean": 0.18359375,
+      "stability/response_length_mean": 33.875,
+      "stability/response_length_std": 32.36151885986328,
+      "stability/response_length_var": 1047.267822265625,
+      "stability/token_entropy_mean": 5.73046875,
+      "step": 70
+    },
+    {
+      "epoch": 0.3838771593090211,
+      "grad_norm": 0.610389769077301,
+      "kl": 0.09753876179456711,
+      "learning_rate": 7.4e-07,
+      "logits/chosen": -49507348.75675676,
+      "logits/rejected": -47883686.697674416,
+      "logps/chosen": -314.1405458192568,
+      "logps/rejected": -228.81352198401163,
+      "loss": 0.5095,
+      "loss/base_kto": 0.49918660521507263,
+      "metrics/advantage_var": 0.07326646149158478,
+      "regularization/mmd": 0.00923239067196846,
+      "regularization/rkhs_norm": 1.8464781045913696,
+      "rewards/chosen": 0.013025604389809273,
+      "rewards/margins": 0.005891873767584243,
+      "rewards/rejected": 0.00713373062222503,
+      "stability/repetition_rate_mean": 0.08548677712678909,
+      "stability/response_length_mean": 22.75,
+      "stability/response_length_std": 31.07479476928711,
+      "stability/response_length_var": 965.6428833007812,
+      "stability/token_entropy_mean": 2.150390625,
+      "step": 75
+    },
+    {
+      "epoch": 0.4094689699296225,
+      "grad_norm": 0.6044193506240845,
+      "kl": 0.1186995729804039,
+      "learning_rate": 7.9e-07,
+      "logits/chosen": -58622508.0794702,
+      "logits/rejected": -53452569.75147929,
+      "logps/chosen": -403.86501448675494,
+      "logps/rejected": -200.98915957840237,
+      "loss": 0.5092,
+      "loss/base_kto": 0.49880704283714294,
+      "metrics/advantage_var": 0.06772632896900177,
+      "regularization/mmd": 0.008787428960204124,
+      "regularization/rkhs_norm": 1.7574856281280518,
+      "rewards/chosen": 0.016511326594068513,
+      "rewards/margins": 0.009846882590483691,
+      "rewards/rejected": 0.006664444003584822,
+      "stability/repetition_rate_mean": 0.18307292461395264,
+      "stability/response_length_mean": 42.0,
+      "stability/response_length_std": 30.481843948364258,
+      "stability/response_length_var": 929.1428833007812,
+      "stability/token_entropy_mean": 5.234375,
+      "step": 80
+    },
+    {
+      "epoch": 0.4350607805502239,
+      "grad_norm": 0.6132326126098633,
+      "kl": 0.12901714444160461,
+      "learning_rate": 8.399999999999999e-07,
+      "logits/chosen": -51477778.16774193,
+      "logits/rejected": -47139790.35151515,
+      "logps/chosen": -358.1587701612903,
+      "logps/rejected": -230.33401988636365,
+      "loss": 0.5098,
+      "loss/base_kto": 0.49895045161247253,
+      "metrics/advantage_var": 0.082537442445755,
+      "regularization/mmd": 0.00979132205247879,
+      "regularization/rkhs_norm": 1.9582645893096924,
+      "rewards/chosen": 0.017374797021189044,
+      "rewards/margins": 0.00829642441265744,
+      "rewards/rejected": 0.009078372608531605,
+      "stability/repetition_rate_mean": 0.09735831618309021,
+      "stability/response_length_mean": 31.875,
+      "stability/response_length_std": 33.0516357421875,
+      "stability/response_length_var": 1092.4107666015625,
+      "stability/token_entropy_mean": 4.90625,
+      "step": 85
+    },
+    {
+      "epoch": 0.46065259117082535,
+      "grad_norm": 0.6399518847465515,
+      "kl": 0.17196393013000488,
+      "learning_rate": 8.9e-07,
+      "logits/chosen": -58408185.43589743,
+      "logits/rejected": -50558657.56097561,
+      "logps/chosen": -380.54271834935895,
+      "logps/rejected": -187.10587366615854,
+      "loss": 0.51,
+      "loss/base_kto": 0.4982968270778656,
+      "metrics/advantage_var": 0.10217716544866562,
+      "regularization/mmd": 0.011231121607124805,
+      "regularization/rkhs_norm": 2.2462244033813477,
+      "rewards/chosen": 0.024664967488019895,
+      "rewards/margins": 0.012572131468550424,
+      "rewards/rejected": 0.01209283601946947,
+      "stability/repetition_rate_mean": 0.2192234843969345,
+      "stability/response_length_mean": 33.625,
+      "stability/response_length_std": 29.442134857177734,
+      "stability/response_length_var": 866.8392944335938,
+      "stability/token_entropy_mean": 3.7060546875,
+      "step": 90
+    },
+    {
+      "epoch": 0.48624440179142675,
+      "grad_norm": 0.6170993447303772,
+      "kl": 0.1672627478837967,
+      "learning_rate": 9.399999999999999e-07,
+      "logits/chosen": -50378544.60759494,
+      "logits/rejected": -48969532.04938272,
+      "logps/chosen": -314.7658722310127,
+      "logps/rejected": -207.07737750771605,
+      "loss": 0.5107,
+      "loss/base_kto": 0.49796175956726074,
+      "metrics/advantage_var": 0.14120100438594818,
+      "regularization/mmd": 0.012432265095412731,
+      "regularization/rkhs_norm": 2.48645281791687,
+      "rewards/chosen": 0.022749973248831833,
+      "rewards/margins": 0.01611456490621136,
+      "rewards/rejected": 0.0066354083426204725,
+      "stability/repetition_rate_mean": 0.19374233484268188,
+      "stability/response_length_mean": 27.0,
+      "stability/response_length_std": 31.1264705657959,
+      "stability/response_length_var": 968.8571166992188,
+      "stability/token_entropy_mean": 3.751953125,
+      "step": 95
+    },
+    {
+      "epoch": 0.5118362124120281,
+      "grad_norm": 0.6102624535560608,
+      "kl": 0.20016632974147797,
+      "learning_rate": 9.9e-07,
+      "logits/chosen": -54725538.5945946,
+      "logits/rejected": -50922793.674418606,
+      "logps/chosen": -359.54178103885135,
+      "logps/rejected": -216.3764989098837,
+      "loss": 0.511,
+      "loss/base_kto": 0.4985528886318207,
+      "metrics/advantage_var": 0.12865900993347168,
+      "regularization/mmd": 0.01211935468018055,
+      "regularization/rkhs_norm": 2.423870801925659,
+      "rewards/chosen": 0.03283950444814321,
+      "rewards/margins": 0.01260079583006786,
+      "rewards/rejected": 0.02023870861807535,
+      "stability/repetition_rate_mean": 0.23773986101150513,
+      "stability/response_length_mean": 37.5,
+      "stability/response_length_std": 29.640705108642578,
+      "stability/response_length_var": 878.5714111328125,
+      "stability/token_entropy_mean": 4.15625,
+      "step": 100
+    },
+    {
+      "epoch": 0.5374280230326296,
+      "grad_norm": 0.6752254366874695,
+      "kl": 0.22721095383167267,
+      "learning_rate": 9.999815554351224e-07,
+      "logits/chosen": -59377960.09638554,
+      "logits/rejected": -53795347.948051944,
+      "logps/chosen": -399.49143448795184,
+      "logps/rejected": -209.84331879058442,
+      "loss": 0.5118,
+      "loss/base_kto": 0.4968232214450836,
+      "metrics/advantage_var": 0.2163117676973343,
+      "regularization/mmd": 0.015834027901291847,
+      "regularization/rkhs_norm": 3.1668055057525635,
+      "rewards/chosen": 0.04110269374158009,
+      "rewards/margins": 0.02505132391351692,
+      "rewards/rejected": 0.016051369828063173,
+      "stability/repetition_rate_mean": 0.189453125,
+      "stability/response_length_mean": 35.25,
+      "stability/response_length_std": 31.494897842407227,
+      "stability/response_length_var": 991.9285888671875,
+      "stability/token_entropy_mean": 3.11328125,
+      "step": 105
+    },
+    {
+      "epoch": 0.5630198336532309,
+      "grad_norm": 0.6515340805053711,
+      "kl": 0.2592513859272003,
+      "learning_rate": 9.999066267225444e-07,
+      "logits/chosen": -57376342.361445785,
+      "logits/rejected": -53605545.55844156,
+      "logps/chosen": -378.17446347891564,
+      "logps/rejected": -205.87290736607142,
+      "loss": 0.5132,
+      "loss/base_kto": 0.49767494201660156,
+      "metrics/advantage_var": 0.22866569459438324,
+      "regularization/mmd": 0.016469214111566544,
+      "regularization/rkhs_norm": 3.2938427925109863,
+      "rewards/chosen": 0.03888099164847868,
+      "rewards/margins": 0.01735842323064468,
+      "rewards/rejected": 0.021522568417833995,
+      "stability/repetition_rate_mean": 0.3042376935482025,
+      "stability/response_length_mean": 35.875,
+      "stability/response_length_std": 29.811012268066406,
+      "stability/response_length_var": 888.6964111328125,
+      "stability/token_entropy_mean": 4.0595703125,
+      "step": 110
+    },
+    {
+      "epoch": 0.5886116442738324,
+      "grad_norm": 0.649266242980957,
+      "kl": 0.31319910287857056,
+      "learning_rate": 9.997740697079592e-07,
+      "logits/chosen": -60592257.62025317,
+      "logits/rejected": -53408388.74074074,
+      "logps/chosen": -364.08156151107596,
+      "logps/rejected": -200.96959394290124,
+      "loss": 0.5148,
+      "loss/base_kto": 0.49650001525878906,
+      "metrics/advantage_var": 0.25559911131858826,
+      "regularization/mmd": 0.01740214042365551,
+      "regularization/rkhs_norm": 3.480428457260132,
+      "rewards/chosen": 0.051097362856321694,
+      "rewards/margins": 0.028401050293402737,
+      "rewards/rejected": 0.022696312562918958,
+      "stability/repetition_rate_mean": 0.3111979365348816,
+      "stability/response_length_mean": 49.25,
+      "stability/response_length_std": 27.316923141479492,
+      "stability/response_length_var": 746.2142944335938,
+      "stability/token_entropy_mean": 4.77734375,
+      "step": 115
+    },
+    {
+      "epoch": 0.6142034548944337,
+      "grad_norm": 0.6187500357627869,
+      "kl": 0.34892186522483826,
+      "learning_rate": 9.995838996722914e-07,
+      "logits/chosen": -57730777.51633987,
+      "logits/rejected": -54345593.1017964,
+      "logps/chosen": -383.69628267973854,
+      "logps/rejected": -211.9707616017964,
+      "loss": 0.5155,
+      "loss/base_kto": 0.49620935320854187,
+      "metrics/advantage_var": 0.3712993860244751,
+      "regularization/mmd": 0.020150119438767433,
+      "regularization/rkhs_norm": 4.03002405166626,
+      "rewards/chosen": 0.05486513274947023,
+      "rewards/margins": 0.030305783392459158,
+      "rewards/rejected": 0.024559349357011075,
+      "stability/repetition_rate_mean": 0.18938209116458893,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.687660217285156,
+      "stability/response_length_var": 881.3571166992188,
+      "stability/token_entropy_mean": 4.5625,
+      "step": 120
+    },
+    {
+      "epoch": 0.6397952655150352,
+      "grad_norm": 0.611441969871521,
+      "kl": 0.3908672630786896,
+      "learning_rate": 9.993361385379874e-07,
+      "logits/chosen": -60671656.32876712,
+      "logits/rejected": -55648673.83908046,
+      "logps/chosen": -416.7914704623288,
+      "logps/rejected": -216.10171964798852,
+      "loss": 0.5184,
+      "loss/base_kto": 0.4961169362068176,
+      "metrics/advantage_var": 0.5313876271247864,
+      "regularization/mmd": 0.023098159581422806,
+      "regularization/rkhs_norm": 4.619632244110107,
+      "rewards/chosen": 0.058781075150999305,
+      "rewards/margins": 0.03223435023960551,
+      "rewards/rejected": 0.02654672491139379,
+      "stability/repetition_rate_mean": 0.16215278208255768,
+      "stability/response_length_mean": 15.25,
+      "stability/response_length_std": 24.93276596069336,
+      "stability/response_length_var": 621.6428833007812,
+      "stability/token_entropy_mean": 3.640625,
+      "step": 125
+    },
+    {
+      "epoch": 0.6653870761356366,
+      "grad_norm": 0.5946590304374695,
+      "kl": 0.4510687291622162,
+      "learning_rate": 9.99030814866488e-07,
+      "logits/chosen": -57380811.1483871,
+      "logits/rejected": -53835955.97575758,
+      "logps/chosen": -391.98538306451616,
+      "logps/rejected": -183.6492897727273,
+      "loss": 0.5201,
+      "loss/base_kto": 0.49492835998535156,
+      "metrics/advantage_var": 0.8019859194755554,
+      "regularization/mmd": 0.02793542854487896,
+      "regularization/rkhs_norm": 5.587085723876953,
+      "rewards/chosen": 0.06725460175544985,
+      "rewards/margins": 0.040291827305321125,
+      "rewards/rejected": 0.026962774450128728,
+      "stability/repetition_rate_mean": 0.20181818306446075,
+      "stability/response_length_mean": 38.0,
+      "stability/response_length_std": 29.13515281677246,
+      "stability/response_length_var": 848.8571166992188,
+      "stability/token_entropy_mean": 3.99609375,
+      "step": 130
+    },
+    {
+      "epoch": 0.690978886756238,
+      "grad_norm": 0.5928955674171448,
+      "kl": 0.49744081497192383,
+      "learning_rate": 9.986679638549358e-07,
+      "logits/chosen": -59853874.25766871,
+      "logits/rejected": -55008706.03821656,
+      "logps/chosen": -401.01260544478527,
+      "logps/rejected": -247.52005374203821,
+      "loss": 0.5209,
+      "loss/base_kto": 0.4942546486854553,
+      "metrics/advantage_var": 1.02751624584198,
+      "regularization/mmd": 0.03274526074528694,
+      "regularization/rkhs_norm": 6.549052715301514,
+      "rewards/chosen": 0.08035813080021209,
+      "rewards/margins": 0.04529071829624675,
+      "rewards/rejected": 0.03506741250396534,
+      "stability/repetition_rate_mean": 0.2230902761220932,
+      "stability/response_length_mean": 29.25,
+      "stability/response_length_std": 29.936838150024414,
+      "stability/response_length_var": 896.2142944335938,
+      "stability/token_entropy_mean": 4.078125,
+      "step": 135
+    },
+    {
+      "epoch": 0.7165706973768394,
+      "grad_norm": 0.7285244464874268,
+      "kl": 0.4897715151309967,
+      "learning_rate": 9.982476273321175e-07,
+      "logits/chosen": -57075643.733333334,
+      "logits/rejected": -50076533.26451613,
+      "logps/chosen": -366.3785511363636,
+      "logps/rejected": -189.9105216733871,
+      "loss": 0.523,
+      "loss/base_kto": 0.49187538027763367,
+      "metrics/advantage_var": 1.7579011917114258,
+      "regularization/mmd": 0.03410874679684639,
+      "regularization/rkhs_norm": 6.821749210357666,
+      "rewards/chosen": 0.08111698844216086,
+      "rewards/margins": 0.06872179263498077,
+      "rewards/rejected": 0.012395195807180097,
+      "stability/repetition_rate_mean": 0.220703125,
+      "stability/response_length_mean": 56.25,
+      "stability/response_length_std": 21.920310974121094,
+      "stability/response_length_var": 480.5,
+      "stability/token_entropy_mean": 5.052734375,
+      "step": 140
+    },
+    {
+      "epoch": 0.7421625079974408,
+      "grad_norm": 0.6769351959228516,
+      "kl": 0.5653125047683716,
+      "learning_rate": 9.977698537536417e-07,
+      "logits/chosen": -51149342.84337349,
+      "logits/rejected": -51664423.896103896,
+      "logps/chosen": -329.4778567394578,
+      "logps/rejected": -199.1466492491883,
+      "loss": 0.5245,
+      "loss/base_kto": 0.4942997097969055,
+      "metrics/advantage_var": 0.7970510125160217,
+      "regularization/mmd": 0.029787013307213783,
+      "regularization/rkhs_norm": 5.95740270614624,
+      "rewards/chosen": 0.07939604104283345,
+      "rewards/margins": 0.045269124553789654,
+      "rewards/rejected": 0.03412691648904379,
+      "stability/repetition_rate_mean": 0.26596763730049133,
+      "stability/response_length_mean": 28.875,
+      "stability/response_length_std": 29.73423957824707,
+      "stability/response_length_var": 884.125,
+      "stability/token_entropy_mean": 5.228515625,
+      "step": 145
+    },
+    {
+      "epoch": 0.7677543186180422,
+      "grad_norm": 0.6468893885612488,
+      "kl": 0.6396333575248718,
+      "learning_rate": 9.972346981963546e-07,
+      "logits/chosen": -55421066.54117647,
+      "logits/rejected": -50181959.68,
+      "logps/chosen": -396.82509191176473,
+      "logps/rejected": -207.365703125,
+      "loss": 0.5281,
+      "loss/base_kto": 0.49031582474708557,
+      "metrics/advantage_var": 1.2932097911834717,
+      "regularization/mmd": 0.036661840975284576,
+      "regularization/rkhs_norm": 7.33236837387085,
+      "rewards/chosen": 0.11093491946949678,
+      "rewards/margins": 0.0749119384616029,
+      "rewards/rejected": 0.03602298100789388,
+      "stability/repetition_rate_mean": 0.2291666716337204,
+      "stability/response_length_mean": 35.25,
+      "stability/response_length_std": 31.01957893371582,
+      "stability/response_length_var": 962.2142944335938,
+      "stability/token_entropy_mean": 4.66796875,
+      "step": 150
+    },
+    {
+      "epoch": 0.7933461292386437,
+      "grad_norm": 0.6322670578956604,
+      "kl": 0.6517454981803894,
+      "learning_rate": 9.966422223519885e-07,
+      "logits/chosen": -54015186.823529415,
+      "logits/rejected": -49345460.906666666,
+      "logps/chosen": -360.07118566176473,
+      "logps/rejected": -191.05200520833333,
+      "loss": 0.5318,
+      "loss/base_kto": 0.4920540750026703,
+      "metrics/advantage_var": 2.2298808097839355,
+      "regularization/mmd": 0.0417025052011013,
+      "regularization/rkhs_norm": 8.340499877929688,
+      "rewards/chosen": 0.09180937935324276,
+      "rewards/margins": 0.0664488141676959,
+      "rewards/rejected": 0.025360565185546875,
+      "stability/repetition_rate_mean": 0.3406159281730652,
+      "stability/response_length_mean": 61.375,
+      "stability/response_length_std": 7.424621105194092,
+      "stability/response_length_var": 55.125,
+      "stability/token_entropy_mean": 4.5703125,
+      "step": 155
+    },
+    {
+      "epoch": 0.818937939859245,
+      "grad_norm": 0.619647204875946,
+      "kl": 0.7206474542617798,
+      "learning_rate": 9.959924945200523e-07,
+      "logits/chosen": -55251065.9047619,
+      "logits/rejected": -48947941.05263158,
+      "logps/chosen": -415.55845424107144,
+      "logps/rejected": -186.46698961759867,
+      "loss": 0.5348,
+      "loss/base_kto": 0.48844751715660095,
+      "metrics/advantage_var": 1.9670952558517456,
+      "regularization/mmd": 0.04304032027721405,
+      "regularization/rkhs_norm": 8.608064651489258,
+      "rewards/chosen": 0.13368967601231166,
+      "rewards/margins": 0.0906540558750468,
+      "rewards/rejected": 0.043035620137264856,
+      "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": 4.44140625,
+      "step": 160
+    },
+    {
+      "epoch": 0.8445297504798465,
+      "grad_norm": 0.720912516117096,
+      "kl": 0.7607746124267578,
+      "learning_rate": 9.952855895999568e-07,
+      "logits/chosen": -54404996.4137931,
+      "logits/rejected": -49794777.42465753,
+      "logps/chosen": -369.7616738505747,
+      "logps/rejected": -212.15198255565068,
+      "loss": 0.5371,
+      "loss/base_kto": 0.4911351203918457,
+      "metrics/advantage_var": 1.7204967737197876,
+      "regularization/mmd": 0.04163478687405586,
+      "regularization/rkhs_norm": 8.326957702636719,
+      "rewards/chosen": 0.11550897839425624,
+      "rewards/margins": 0.07149485110711007,
+      "rewards/rejected": 0.04401412728714616,
+      "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": 4.8984375,
+      "step": 165
+    },
+    {
+      "epoch": 0.8701215611004478,
+      "grad_norm": 0.6247206926345825,
+      "kl": 0.924032986164093,
+      "learning_rate": 9.9452158908238e-07,
+      "logits/chosen": -59132631.10059172,
+      "logits/rejected": -52570654.51655629,
+      "logps/chosen": -439.69008875739644,
+      "logps/rejected": -190.55952504139074,
+      "loss": 0.5386,
+      "loss/base_kto": 0.48805713653564453,
+      "metrics/advantage_var": 2.3215138912200928,
+      "regularization/mmd": 0.0507260225713253,
+      "regularization/rkhs_norm": 10.145204544067383,
+      "rewards/chosen": 0.16517957427797939,
+      "rewards/margins": 0.09534949453615406,
+      "rewards/rejected": 0.06983007974182533,
+      "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": 3.5888671875,
+      "step": 170
+    },
+    {
+      "epoch": 0.8957133717210493,
+      "grad_norm": 0.7177579998970032,
+      "kl": 0.9547041058540344,
+      "learning_rate": 9.937005810398745e-07,
+      "logits/chosen": -56654465.54216868,
+      "logits/rejected": -48027790.96103896,
+      "logps/chosen": -354.88453030873495,
+      "logps/rejected": -194.3028612012987,
+      "loss": 0.5416,
+      "loss/base_kto": 0.48682698607444763,
+      "metrics/advantage_var": 4.120774745941162,
+      "regularization/mmd": 0.06023603305220604,
+      "regularization/rkhs_norm": 12.047205924987793,
+      "rewards/chosen": 0.15526514168245248,
+      "rewards/margins": 0.10479820168199273,
+      "rewards/rejected": 0.050466940000459745,
+      "stability/repetition_rate_mean": 0.2005799561738968,
+      "stability/response_length_mean": 27.125,
+      "stability/response_length_std": 25.737340927124023,
+      "stability/response_length_var": 662.4107055664062,
+      "stability/token_entropy_mean": 4.8115234375,
+      "step": 175
+    },
+    {
+      "epoch": 0.9213051823416507,
+      "grad_norm": 0.6431882977485657,
+      "kl": 1.0281428098678589,
+      "learning_rate": 9.928226601167138e-07,
+      "logits/chosen": -52652606.82208589,
+      "logits/rejected": -47977778.5477707,
+      "logps/chosen": -374.4046922929448,
+      "logps/rejected": -210.54264530254778,
+      "loss": 0.5458,
+      "loss/base_kto": 0.48759326338768005,
+      "metrics/advantage_var": 4.256661415100098,
+      "regularization/mmd": 0.062359750270843506,
+      "regularization/rkhs_norm": 12.471949577331543,
+      "rewards/chosen": 0.17411304544086106,
+      "rewards/margins": 0.09855715900823765,
+      "rewards/rejected": 0.07555588643262341,
+      "stability/repetition_rate_mean": 0.16484375298023224,
+      "stability/response_length_mean": 34.75,
+      "stability/response_length_std": 31.54928207397461,
+      "stability/response_length_var": 995.3571166992188,
+      "stability/token_entropy_mean": 3.6171875,
+      "step": 180
+    },
+    {
+      "epoch": 0.946896992962252,
+      "grad_norm": 0.6504088044166565,
+      "kl": 1.0762888193130493,
+      "learning_rate": 9.918879275179817e-07,
+      "logits/chosen": -54606911.209876545,
+      "logits/rejected": -49525303.088607594,
+      "logps/chosen": -310.3205536265432,
+      "logps/rejected": -196.54314181170886,
+      "loss": 0.5519,
+      "loss/base_kto": 0.4868957996368408,
+      "metrics/advantage_var": 7.445571422576904,
+      "regularization/mmd": 0.07445289939641953,
+      "regularization/rkhs_norm": 14.890579223632812,
+      "rewards/chosen": 0.16808104809419608,
+      "rewards/margins": 0.11847921255361774,
+      "rewards/rejected": 0.049601835540578336,
+      "stability/repetition_rate_mean": 0.19921875,
+      "stability/response_length_mean": 32.875,
+      "stability/response_length_std": 33.30996322631836,
+      "stability/response_length_var": 1109.5535888671875,
+      "stability/token_entropy_mean": 3.716796875,
+      "step": 185
+    },
+    {
+      "epoch": 0.9724888035828535,
+      "grad_norm": 0.7037036418914795,
+      "kl": 1.1190913915634155,
+      "learning_rate": 9.90896490997906e-07,
+      "logits/chosen": -53083148.263473056,
+      "logits/rejected": -49686661.85620915,
+      "logps/chosen": -390.29261788922156,
+      "logps/rejected": -210.45340584150327,
+      "loss": 0.5607,
+      "loss/base_kto": 0.4832470118999481,
+      "metrics/advantage_var": 7.204617500305176,
+      "regularization/mmd": 0.0810161754488945,
+      "regularization/rkhs_norm": 16.203235626220703,
+      "rewards/chosen": 0.21127086342451815,
+      "rewards/margins": 0.1367532206740586,
+      "rewards/rejected": 0.07451764275045956,
+      "stability/repetition_rate_mean": 0.19138340651988983,
+      "stability/response_length_mean": 18.25,
+      "stability/response_length_std": 26.778188705444336,
+      "stability/response_length_var": 717.0714111328125,
+      "stability/token_entropy_mean": 4.83203125,
+      "step": 190
+    },
+    {
+      "epoch": 0.9980806142034548,
+      "grad_norm": 0.6389921307563782,
+      "kl": 1.1464111804962158,
+      "learning_rate": 9.898484648474362e-07,
+      "logits/chosen": -60020286.76129032,
+      "logits/rejected": -53779549.09090909,
+      "logps/chosen": -390.8505040322581,
+      "logps/rejected": -197.09185606060606,
+      "loss": 0.5603,
+      "loss/base_kto": 0.48602548241615295,
+      "metrics/advantage_var": 4.388164520263672,
+      "regularization/mmd": 0.06844492256641388,
+      "regularization/rkhs_norm": 13.688984870910645,
+      "rewards/chosen": 0.1934424861784904,
+      "rewards/margins": 0.11069587062414322,
+      "rewards/rejected": 0.08274661555434719,
+      "stability/repetition_rate_mean": 0.3953683078289032,
+      "stability/response_length_mean": 49.0,
+      "stability/response_length_std": 27.82085609436035,
+      "stability/response_length_var": 774.0,
+      "stability/token_entropy_mean": 5.16796875,
+      "step": 195
+    },
+    {
+      "epoch": 1.0,
+      "step": 196,
+      "total_flos": 3.1861866257511875e+18,
+      "train_loss": 0.5200024314072668,
+      "train_runtime": 3159.9923,
+      "train_samples_per_second": 31.646,
+      "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_mmd_rho0.005_lambda_kr0.01/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7afba6a96f544b9b3a7bd795e549d676d18d66bf7bce223ed504d9c66dfaf4c7
+size 8081