Upload full trainer checkpoint step=2026 (profile=llama3_2_1b_base_grpo, run_type=grpo)

.gitattributes

@@ -128,3 +128,4 @@ meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-13182/tokenizer.json filter
 meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-13689/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-14196/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-1013/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2026/tokenizer.json filter=lfs diff=lfs merge=lfs -text

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2026/README.md

@@ -0,0 +1,210 @@
+---
+base_model: unsloth/Llama-3.2-1B
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:unsloth/Llama-3.2-1B
+- grpo
+- lora
+- transformers
+- trl
+- unsloth
+---
+
+# 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 0.18.1

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2026/adapter_config.json

@@ -0,0 +1,50 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "LlamaForCausalLM",
+    "parent_library": "transformers.models.llama.modeling_llama",
+    "unsloth_fixed": true
+  },
+  "base_model_name_or_path": "unsloth/Llama-3.2-1B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "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",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "q_proj",
+    "up_proj",
+    "gate_proj",
+    "o_proj",
+    "k_proj",
+    "v_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2026/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5479de16b3ce7006ba65f0f19b1c4bc9b9dc863e35d1352cbc80ccbf131e6006
+size 45118424

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2026/chat_template.jinja

@@ -0,0 +1,12 @@
+{{ bos_token }}{% if messages[0]['role'] == 'system' %}{{ messages[0]['content'] + '
+
+' }}{% set loop_messages = messages[1:] %}{% else %}{{ '' + '
+
+' }}{% set loop_messages = messages %}{% endif %}{% for message in loop_messages %}{% if message['role'] == 'user' %}{{ '### Instruction:
+' + message['content'] + '
+
+' }}{% elif message['role'] == 'assistant' %}{{ '### Response:
+' + message['content'] + eos_token + '
+
+' }}{% else %}{{ raise_exception('Only user and assistant roles are supported!') }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '### Response:
+' }}{% endif %}

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2026/optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:15e1443b8c809866e8e7f27aa31ca4c833c6f43704ec0c11de6e163bee11662f
+size 23162187

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2026/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e32303cd2e7e543bebef6326a35dd6e3a749c0243d974e44b5cbb96a6346f9bf
+size 14645

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2026/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:293f77415e70dbf5ed1d7acb306a6c47160f4cfa8264202a1bb23887ee0ec131
+size 1465

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2026/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a65c6c5f9764771aa485e6a1f5e63d7d9af8477fe0777148c17476ecb2e09a05
+size 17210099

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2026/tokenizer_config.json

@@ -0,0 +1,18 @@
+{
+  "backend": "tokenizers",
+  "bos_token": "<|begin_of_text|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|end_of_text|>",
+  "from_slow": true,
+  "is_local": false,
+  "legacy": false,
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
+  "model_max_length": 131072,
+  "pad_token": "<|finetune_right_pad_id|>",
+  "padding_side": "right",
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null
+}

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2026/trainer_state.json

@@ -0,0 +1,1132 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.003598515124065292,
+  "eval_steps": 500,
+  "global_step": 2026,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 65.68,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 65.88,
+      "completions/max_terminated_length": 65.88,
+      "completions/mean_length": 65.68,
+      "completions/mean_terminated_length": 65.68,
+      "completions/min_length": 65.48,
+      "completions/min_terminated_length": 65.48,
+      "epoch": 8.88083693007229e-05,
+      "frac_reward_zero_std": 0.56,
+      "grad_norm": 0.5864518284797668,
+      "kl": 0.2340605702996254,
+      "learning_rate": 5.439488466064254e-09,
+      "loss": 0.0004230526089668274,
+      "num_tokens": 46552.0,
+      "reward": 3.2782554626464844e-09,
+      "reward_std": 0.47037978172302247,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 3.2782554626464844e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.4703797769546509,
+      "step": 50
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 53.46,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 53.6,
+      "completions/max_terminated_length": 53.6,
+      "completions/mean_length": 53.46,
+      "completions/mean_terminated_length": 53.46,
+      "completions/min_length": 53.24,
+      "completions/min_terminated_length": 53.24,
+      "epoch": 0.0001776167386014458,
+      "frac_reward_zero_std": 0.68,
+      "grad_norm": 7.498858030885458e-05,
+      "kl": 0.293030250146985,
+      "learning_rate": 1.0989986900823695e-08,
+      "loss": 0.0005571873486042023,
+      "num_tokens": 85556.0,
+      "reward": 4.76837158203125e-09,
+      "reward_std": 0.3420943832397461,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 4.76837158203125e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.3420943832397461,
+      "step": 100
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0002914614137262106,
+      "clip_ratio/low_min": 0.0002914614137262106,
+      "clip_ratio/region_mean": 0.0002914614137262106,
+      "completion_length": 62.48,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 62.6,
+      "completions/max_terminated_length": 62.6,
+      "completions/mean_length": 62.48,
+      "completions/mean_terminated_length": 62.48,
+      "completions/min_length": 62.24,
+      "completions/min_terminated_length": 62.24,
+      "epoch": 0.0002664251079021687,
+      "frac_reward_zero_std": 0.56,
+      "grad_norm": 0.2601912319660187,
+      "kl": 0.2793836747482419,
+      "learning_rate": 1.6540485335583138e-08,
+      "loss": 0.0005555346608161927,
+      "num_tokens": 130380.0,
+      "reward": 1.7881393432617187e-09,
+      "reward_std": 0.4703797769546509,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.7881393432617187e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.47037978172302247,
+      "step": 150
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 62.165,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 62.4,
+      "completions/max_terminated_length": 62.4,
+      "completions/mean_length": 62.165,
+      "completions/mean_terminated_length": 62.165,
+      "completions/min_length": 61.84,
+      "completions/min_terminated_length": 61.84,
+      "epoch": 0.0003552334772028916,
+      "frac_reward_zero_std": 0.48,
+      "grad_norm": 0.39951568841934204,
+      "kl": 0.2763713626563549,
+      "learning_rate": 2.2090983770342578e-08,
+      "loss": 0.0009129984676837921,
+      "num_tokens": 174957.0,
+      "reward": 8.940696716308593e-10,
+      "reward_std": 0.5559033823013305,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 8.940696716308593e-10,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.5559033775329589,
+      "step": 200
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 65.825,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 66.04,
+      "completions/max_terminated_length": 66.04,
+      "completions/mean_length": 65.825,
+      "completions/mean_terminated_length": 65.825,
+      "completions/min_length": 65.56,
+      "completions/min_terminated_length": 65.56,
+      "epoch": 0.0004440418465036145,
+      "frac_reward_zero_std": 0.52,
+      "grad_norm": 5.004892227589153e-05,
+      "kl": 0.2560868472605944,
+      "learning_rate": 2.764148220510202e-08,
+      "loss": 0.001215285211801529,
+      "num_tokens": 222266.0,
+      "reward": 4.470348358154297e-09,
+      "reward_std": 0.5131415796279907,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 4.470348358154297e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.5131415748596191,
+      "step": 250
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 68.75,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 69.04,
+      "completions/max_terminated_length": 69.04,
+      "completions/mean_length": 68.75,
+      "completions/mean_terminated_length": 68.75,
+      "completions/min_length": 68.56,
+      "completions/min_terminated_length": 68.56,
+      "epoch": 0.0005328502158043374,
+      "frac_reward_zero_std": 0.6,
+      "grad_norm": 1.926846834976459e-06,
+      "kl": 0.239115035161376,
+      "learning_rate": 3.3191980639861464e-08,
+      "loss": 0.001264408677816391,
+      "num_tokens": 271736.0,
+      "reward": 6.258487701416015e-09,
+      "reward_std": 0.4276179838180542,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 6.258487701416015e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.42761797904968263,
+      "step": 300
+    },
+    {
+      "clip_ratio/high_max": 4.8543689772486686e-05,
+      "clip_ratio/high_mean": 4.8543689772486686e-05,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 4.8543689772486686e-05,
+      "completion_length": 61.07,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 61.2,
+      "completions/max_terminated_length": 61.2,
+      "completions/mean_length": 61.07,
+      "completions/mean_terminated_length": 61.07,
+      "completions/min_length": 60.96,
+      "completions/min_terminated_length": 60.96,
+      "epoch": 0.0006216585851050603,
+      "frac_reward_zero_std": 0.72,
+      "grad_norm": 2.6762343168229563e-06,
+      "kl": 0.2665533754974604,
+      "learning_rate": 3.8742479074620904e-08,
+      "loss": 0.0013578705489635467,
+      "num_tokens": 316350.0,
+      "reward": 2.384185791015625e-09,
+      "reward_std": 0.29933258533477786,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.384185791015625e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.2993325901031494,
+      "step": 350
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 64.22,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 64.6,
+      "completions/max_terminated_length": 64.6,
+      "completions/mean_length": 64.22,
+      "completions/mean_terminated_length": 64.22,
+      "completions/min_length": 63.92,
+      "completions/min_terminated_length": 63.92,
+      "epoch": 0.0007104669544057832,
+      "frac_reward_zero_std": 0.6,
+      "grad_norm": 1.1867748526128707e-06,
+      "kl": 0.2582776914909482,
+      "learning_rate": 4.429297750938035e-08,
+      "loss": 0.000348069965839386,
+      "num_tokens": 362546.0,
+      "reward": 0.0,
+      "reward_std": 0.42761797904968263,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.42761797904968263,
+      "step": 400
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 65.25,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 65.36,
+      "completions/max_terminated_length": 65.36,
+      "completions/mean_length": 65.25,
+      "completions/mean_terminated_length": 65.25,
+      "completions/min_length": 65.04,
+      "completions/min_terminated_length": 65.04,
+      "epoch": 0.000799275323706506,
+      "frac_reward_zero_std": 0.6,
+      "grad_norm": 0.0002190974773839116,
+      "kl": 0.24411882523447276,
+      "learning_rate": 4.984347594413978e-08,
+      "loss": 0.0009650686383247376,
+      "num_tokens": 409412.0,
+      "reward": 1.7881393432617187e-09,
+      "reward_std": 0.4276179838180542,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.7881393432617187e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.4276179838180542,
+      "step": 450
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 65.765,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 65.92,
+      "completions/max_terminated_length": 65.92,
+      "completions/mean_length": 65.765,
+      "completions/mean_terminated_length": 65.765,
+      "completions/min_length": 65.72,
+      "completions/min_terminated_length": 65.72,
+      "epoch": 0.000888083693007229,
+      "frac_reward_zero_std": 0.88,
+      "grad_norm": 0.03857799246907234,
+      "kl": 0.23137393046170474,
+      "learning_rate": 5.5393974378899236e-08,
+      "loss": 0.00039650365710258484,
+      "num_tokens": 456021.0,
+      "reward": -5.960464477539063e-10,
+      "reward_std": 0.1282853889465332,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -5.960464477539063e-10,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.12828539371490477,
+      "step": 500
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 66.25,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 66.64,
+      "completions/max_terminated_length": 66.64,
+      "completions/mean_length": 66.25,
+      "completions/mean_terminated_length": 66.25,
+      "completions/min_length": 66.0,
+      "completions/min_terminated_length": 66.0,
+      "epoch": 0.0009768920623079518,
+      "frac_reward_zero_std": 0.68,
+      "grad_norm": 1.3617865988635458e-05,
+      "kl": 0.2465892593190074,
+      "learning_rate": 6.094447281365867e-08,
+      "loss": 0.0013641029596328734,
+      "num_tokens": 503519.0,
+      "reward": 0.0,
+      "reward_std": 0.3420943832397461,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.3420943832397461,
+      "step": 550
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 8.620689623057842e-05,
+      "clip_ratio/low_min": 8.620689623057842e-05,
+      "clip_ratio/region_mean": 8.620689623057842e-05,
+      "completion_length": 70.03,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 70.16,
+      "completions/max_terminated_length": 70.16,
+      "completions/mean_length": 70.03,
+      "completions/mean_terminated_length": 70.03,
+      "completions/min_length": 69.8,
+      "completions/min_terminated_length": 69.8,
+      "epoch": 0.0010657004316086747,
+      "frac_reward_zero_std": 0.52,
+      "grad_norm": 0.00015144512872211635,
+      "kl": 0.22738730356097223,
+      "learning_rate": 6.649497124841812e-08,
+      "loss": 0.0008844786882400513,
+      "num_tokens": 553309.0,
+      "reward": 3.2782554626464844e-09,
+      "reward_std": 0.5131415700912476,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 3.2782554626464844e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.5131415748596191,
+      "step": 600
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 71.28,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 71.76,
+      "completions/max_terminated_length": 71.76,
+      "completions/mean_length": 71.28,
+      "completions/mean_terminated_length": 71.28,
+      "completions/min_length": 70.52,
+      "completions/min_terminated_length": 70.52,
+      "epoch": 0.0011545088009093977,
+      "frac_reward_zero_std": 0.44,
+      "grad_norm": 8.714160685485695e-06,
+      "kl": 0.23357330340892077,
+      "learning_rate": 7.204546968317756e-08,
+      "loss": -0.00036261290311813356,
+      "num_tokens": 603181.0,
+      "reward": -8.940696716308593e-10,
+      "reward_std": 0.5986651754379273,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -8.940696716308593e-10,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.5986651754379273,
+      "step": 650
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 71.675,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 71.88,
+      "completions/max_terminated_length": 71.88,
+      "completions/mean_length": 71.675,
+      "completions/mean_terminated_length": 71.675,
+      "completions/min_length": 71.52,
+      "completions/min_terminated_length": 71.52,
+      "epoch": 0.0012433171702101206,
+      "frac_reward_zero_std": 0.6,
+      "grad_norm": 0.1935173124074936,
+      "kl": 0.22929298244416713,
+      "learning_rate": 7.7595968117937e-08,
+      "loss": 0.0007024538516998291,
+      "num_tokens": 654188.0,
+      "reward": 2.9802322387695314e-09,
+      "reward_std": 0.4276179885864258,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.9802322387695314e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.42761797904968263,
+      "step": 700
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 65.605,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 65.92,
+      "completions/max_terminated_length": 65.92,
+      "completions/mean_length": 65.605,
+      "completions/mean_terminated_length": 65.605,
+      "completions/min_length": 65.36,
+      "completions/min_terminated_length": 65.36,
+      "epoch": 0.0013321255395108435,
+      "frac_reward_zero_std": 0.64,
+      "grad_norm": 0.41991785168647766,
+      "kl": 0.23588136799633502,
+      "learning_rate": 8.314646655269643e-08,
+      "loss": 0.0017469561100006104,
+      "num_tokens": 700877.0,
+      "reward": 5.960464477539063e-10,
+      "reward_std": 0.3848561763763428,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 5.960464477539063e-10,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.38485618591308596,
+      "step": 750
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 67.435,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 67.6,
+      "completions/max_terminated_length": 67.6,
+      "completions/mean_length": 67.435,
+      "completions/mean_terminated_length": 67.435,
+      "completions/min_length": 67.28,
+      "completions/min_terminated_length": 67.28,
+      "epoch": 0.0014209339088115665,
+      "frac_reward_zero_std": 0.48,
+      "grad_norm": 0.31073635816574097,
+      "kl": 0.2509309166669846,
+      "learning_rate": 8.869696498745589e-08,
+      "loss": 0.0010700391232967377,
+      "num_tokens": 749028.0,
+      "reward": 6.556510925292969e-09,
+      "reward_std": 0.5559033823013305,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 6.556510925292969e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.5559033727645875,
+      "step": 800
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.00014605543576180934,
+      "clip_ratio/low_min": 0.00014605543576180934,
+      "clip_ratio/region_mean": 0.00014605543576180934,
+      "completion_length": 75.25,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 75.36,
+      "completions/max_terminated_length": 75.36,
+      "completions/mean_length": 75.25,
+      "completions/mean_terminated_length": 75.25,
+      "completions/min_length": 75.08,
+      "completions/min_terminated_length": 75.08,
+      "epoch": 0.0015097422781122894,
+      "frac_reward_zero_std": 0.44,
+      "grad_norm": 0.04336247965693474,
+      "kl": 0.21717634629458188,
+      "learning_rate": 9.424746342221532e-08,
+      "loss": 0.0009317247569561004,
+      "num_tokens": 802606.0,
+      "reward": 4.172325134277344e-09,
+      "reward_std": 0.5986651706695557,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 4.172325134277344e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.5986651706695557,
+      "step": 850
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 74.23,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 74.84,
+      "completions/max_terminated_length": 74.84,
+      "completions/mean_length": 74.23,
+      "completions/mean_terminated_length": 74.23,
+      "completions/min_length": 74.0,
+      "completions/min_terminated_length": 74.0,
+      "epoch": 0.001598550647413012,
+      "frac_reward_zero_std": 0.52,
+      "grad_norm": 0.4461303651332855,
+      "kl": 0.23601274881511927,
+      "learning_rate": 9.979796185697477e-08,
+      "loss": 0.0018890520930290222,
+      "num_tokens": 855396.0,
+      "reward": 0.0,
+      "reward_std": 0.5131415748596191,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.5131415796279907,
+      "step": 900
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 69.48,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 69.64,
+      "completions/max_terminated_length": 69.64,
+      "completions/mean_length": 69.48,
+      "completions/mean_terminated_length": 69.48,
+      "completions/min_length": 69.32,
+      "completions/min_terminated_length": 69.32,
+      "epoch": 0.001687359016713735,
+      "frac_reward_zero_std": 0.6,
+      "grad_norm": 0.3977927565574646,
+      "kl": 0.2380372828245163,
+      "learning_rate": 1.053484602917342e-07,
+      "loss": 0.0015320682525634766,
+      "num_tokens": 905028.0,
+      "reward": 8.940696716308593e-10,
+      "reward_std": 0.42761797904968263,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 8.940696716308593e-10,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.42761797904968263,
+      "step": 950
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 61.315,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 61.6,
+      "completions/max_terminated_length": 61.6,
+      "completions/mean_length": 61.315,
+      "completions/mean_terminated_length": 61.315,
+      "completions/min_length": 60.96,
+      "completions/min_terminated_length": 60.96,
+      "epoch": 0.001776167386014458,
+      "frac_reward_zero_std": 0.68,
+      "grad_norm": 0.0018230041023343801,
+      "kl": 0.27308079957962034,
+      "learning_rate": 1.1089895872649365e-07,
+      "loss": -0.00010064475238323211,
+      "num_tokens": 949379.0,
+      "reward": -8.940696716308593e-10,
+      "reward_std": 0.3420943880081177,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -8.940696716308593e-10,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.3420943832397461,
+      "step": 1000
+    },
+    {
+      "eval_cer_subset": 0.01309296985685671,
+      "eval_cer_subset_edit_distance": 804,
+      "eval_cer_subset_groups": 250,
+      "eval_cer_subset_items": 250,
+      "eval_cer_subset_ref_chars": 61407,
+      "eval_loss": 0.03959455178967374,
+      "step": 1013
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 61.745,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 62.04,
+      "completions/max_terminated_length": 62.04,
+      "completions/mean_length": 61.745,
+      "completions/mean_terminated_length": 61.745,
+      "completions/min_length": 61.52,
+      "completions/min_terminated_length": 61.52,
+      "epoch": 0.001864975755315181,
+      "frac_reward_zero_std": 0.6,
+      "grad_norm": 2.53622838499723e-05,
+      "kl": 0.26631607685238123,
+      "learning_rate": 1.1644945716125309e-07,
+      "loss": 0.0013614676892757416,
+      "num_tokens": 993560.0,
+      "reward": 2.9802322387695313e-10,
+      "reward_std": 0.42761797904968263,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.9802322387695313e-10,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.42761797904968263,
+      "step": 1050
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 62.33,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 62.56,
+      "completions/max_terminated_length": 62.56,
+      "completions/mean_length": 62.33,
+      "completions/mean_terminated_length": 62.33,
+      "completions/min_length": 62.16,
+      "completions/min_terminated_length": 62.16,
+      "epoch": 0.0019537841246159036,
+      "frac_reward_zero_std": 0.6,
+      "grad_norm": 0.16987353563308716,
+      "kl": 0.25355677947402,
+      "learning_rate": 1.2199995559601254e-07,
+      "loss": 0.0013103175163269043,
+      "num_tokens": 1038210.0,
+      "reward": 1.7881393432617187e-09,
+      "reward_std": 0.42761797904968263,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.7881393432617187e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.42761797904968263,
+      "step": 1100
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 67.595,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 67.72,
+      "completions/max_terminated_length": 67.72,
+      "completions/mean_length": 67.595,
+      "completions/mean_terminated_length": 67.595,
+      "completions/min_length": 67.36,
+      "completions/min_terminated_length": 67.36,
+      "epoch": 0.0020425924939166265,
+      "frac_reward_zero_std": 0.56,
+      "grad_norm": 0.21912723779678345,
+      "kl": 0.2458942100405693,
+      "learning_rate": 1.2755045403077197e-07,
+      "loss": 0.0009975716471672057,
+      "num_tokens": 1086193.0,
+      "reward": 0.0,
+      "reward_std": 0.4703797769546509,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.4703797769546509,
+      "step": 1150
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 67.81,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 67.96,
+      "completions/max_terminated_length": 67.96,
+      "completions/mean_length": 67.81,
+      "completions/mean_terminated_length": 67.81,
+      "completions/min_length": 67.68,
+      "completions/min_terminated_length": 67.68,
+      "epoch": 0.0021314008632173495,
+      "frac_reward_zero_std": 0.68,
+      "grad_norm": 0.00010790217493195087,
+      "kl": 0.24153089322149754,
+      "learning_rate": 1.331009524655314e-07,
+      "loss": 0.0017908763885498046,
+      "num_tokens": 1134467.0,
+      "reward": 3.2782554626464844e-09,
+      "reward_std": 0.3420943832397461,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 3.2782554626464844e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.3420943832397461,
+      "step": 1200
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 68.42,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 68.8,
+      "completions/max_terminated_length": 68.8,
+      "completions/mean_length": 68.42,
+      "completions/mean_terminated_length": 68.42,
+      "completions/min_length": 68.0,
+      "completions/min_terminated_length": 68.0,
+      "epoch": 0.0022202092325180724,
+      "frac_reward_zero_std": 0.6,
+      "grad_norm": 0.2312355786561966,
+      "kl": 0.23632734417915344,
+      "learning_rate": 1.3865145090029086e-07,
+      "loss": 0.0006246243417263031,
+      "num_tokens": 1183007.0,
+      "reward": 0.0,
+      "reward_std": 0.4276179885864258,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.42761797904968263,
+      "step": 1250
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 64.655,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 64.8,
+      "completions/max_terminated_length": 64.8,
+      "completions/mean_length": 64.655,
+      "completions/mean_terminated_length": 64.655,
+      "completions/min_length": 64.48,
+      "completions/min_terminated_length": 64.48,
+      "epoch": 0.0023090176018187953,
+      "frac_reward_zero_std": 0.72,
+      "grad_norm": 1.9456435893516755e-06,
+      "kl": 0.2434770704805851,
+      "learning_rate": 1.442019493350503e-07,
+      "loss": 0.00031983185559511187,
+      "num_tokens": 1229834.0,
+      "reward": -2.9802322387695313e-10,
+      "reward_std": 0.29933258056640627,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -2.9802322387695313e-10,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.29933258533477786,
+      "step": 1300
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 62.805,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 63.04,
+      "completions/max_terminated_length": 63.04,
+      "completions/mean_length": 62.805,
+      "completions/mean_terminated_length": 62.805,
+      "completions/min_length": 62.68,
+      "completions/min_terminated_length": 62.68,
+      "epoch": 0.0023978259711195183,
+      "frac_reward_zero_std": 0.56,
+      "grad_norm": 7.737488340353593e-05,
+      "kl": 0.2773256845027208,
+      "learning_rate": 1.4975244776980973e-07,
+      "loss": 0.001342010796070099,
+      "num_tokens": 1274811.0,
+      "reward": -1.1920928955078125e-09,
+      "reward_std": 0.47037978172302247,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -1.1920928955078125e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.4703797769546509,
+      "step": 1350
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0002222222276031971,
+      "clip_ratio/low_min": 0.0002222222276031971,
+      "clip_ratio/region_mean": 0.0002222222276031971,
+      "completion_length": 62.545,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 62.64,
+      "completions/max_terminated_length": 62.64,
+      "completions/mean_length": 62.545,
+      "completions/mean_terminated_length": 62.545,
+      "completions/min_length": 62.36,
+      "completions/min_terminated_length": 62.36,
+      "epoch": 0.002486634340420241,
+      "frac_reward_zero_std": 0.56,
+      "grad_norm": 9.343001693196129e-06,
+      "kl": 0.2813856675475836,
+      "learning_rate": 1.553029462045692e-07,
+      "loss": 0.0014287029206752778,
+      "num_tokens": 1320056.0,
+      "reward": 3.5762786865234374e-09,
+      "reward_std": 0.4703797769546509,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 3.5762786865234374e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.4703797769546509,
+      "step": 1400
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 63.605,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 63.68,
+      "completions/max_terminated_length": 63.68,
+      "completions/mean_length": 63.605,
+      "completions/mean_terminated_length": 63.605,
+      "completions/min_length": 63.56,
+      "completions/min_terminated_length": 63.56,
+      "epoch": 0.002575442709720964,
+      "frac_reward_zero_std": 0.88,
+      "grad_norm": 8.816698937152978e-06,
+      "kl": 0.2244873683154583,
+      "learning_rate": 1.6085344463932862e-07,
+      "loss": 0.0012283537536859513,
+      "num_tokens": 1366185.0,
+      "reward": 1.1920928955078125e-09,
+      "reward_std": 0.12828539371490477,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.1920928955078125e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.12828539371490477,
+      "step": 1450
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 7.352941203862428e-05,
+      "clip_ratio/low_min": 7.352941203862428e-05,
+      "clip_ratio/region_mean": 7.352941203862428e-05,
+      "completion_length": 68.155,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 68.48,
+      "completions/max_terminated_length": 68.48,
+      "completions/mean_length": 68.155,
+      "completions/mean_terminated_length": 68.155,
+      "completions/min_length": 67.96,
+      "completions/min_terminated_length": 67.96,
+      "epoch": 0.002664251079021687,
+      "frac_reward_zero_std": 0.56,
+      "grad_norm": 0.28626617789268494,
+      "kl": 0.2416078744456172,
+      "learning_rate": 1.6640394307408808e-07,
+      "loss": 0.001232394576072693,
+      "num_tokens": 1415224.0,
+      "reward": 2.384185791015625e-09,
+      "reward_std": 0.4703797721862793,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.384185791015625e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.4703797769546509,
+      "step": 1500
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 4.424778744578361e-05,
+      "clip_ratio/low_min": 4.424778744578361e-05,
+      "clip_ratio/region_mean": 4.424778744578361e-05,
+      "completion_length": 70.45,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 70.68,
+      "completions/max_terminated_length": 70.68,
+      "completions/mean_length": 70.45,
+      "completions/mean_terminated_length": 70.45,
+      "completions/min_length": 70.28,
+      "completions/min_terminated_length": 70.28,
+      "epoch": 0.00275305944832241,
+      "frac_reward_zero_std": 0.6,
+      "grad_norm": 0.18359848856925964,
+      "kl": 0.21603738449513912,
+      "learning_rate": 1.7195444150884752e-07,
+      "loss": 0.0005052408576011658,
+      "num_tokens": 1465586.0,
+      "reward": 2.384185791015625e-09,
+      "reward_std": 0.42761797904968263,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.384185791015625e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.42761797904968263,
+      "step": 1550
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 65.97,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 66.28,
+      "completions/max_terminated_length": 66.28,
+      "completions/mean_length": 65.97,
+      "completions/mean_terminated_length": 65.97,
+      "completions/min_length": 65.72,
+      "completions/min_terminated_length": 65.72,
+      "epoch": 0.002841867817623133,
+      "frac_reward_zero_std": 0.6,
+      "grad_norm": 0.001628563622944057,
+      "kl": 0.2370056840032339,
+      "learning_rate": 1.7750493994360695e-07,
+      "loss": 0.0010255253314971923,
+      "num_tokens": 1513108.0,
+      "reward": 5.960464477539063e-10,
+      "reward_std": 0.42761797904968263,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 5.960464477539063e-10,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.42761797904968263,
+      "step": 1600
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.00017536476254463197,
+      "clip_ratio/low_min": 0.00017536476254463197,
+      "clip_ratio/region_mean": 0.00017536476254463197,
+      "completion_length": 65.855,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 66.16,
+      "completions/max_terminated_length": 66.16,
+      "completions/mean_length": 65.855,
+      "completions/mean_terminated_length": 65.855,
+      "completions/min_length": 65.6,
+      "completions/min_terminated_length": 65.6,
+      "epoch": 0.002930676186923856,
+      "frac_reward_zero_std": 0.52,
+      "grad_norm": 0.053214434534311295,
+      "kl": 0.24284519530832768,
+      "learning_rate": 1.8305543837836638e-07,
+      "loss": 0.00212494432926178,
+      "num_tokens": 1559663.0,
+      "reward": 1.4901161193847657e-09,
+      "reward_std": 0.5131415796279907,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.4901161193847657e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.5131415748596191,
+      "step": 1650
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 67.06,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 67.2,
+      "completions/max_terminated_length": 67.2,
+      "completions/mean_length": 67.06,
+      "completions/mean_terminated_length": 67.06,
+      "completions/min_length": 66.8,
+      "completions/min_terminated_length": 66.8,
+      "epoch": 0.0030194845562245788,
+      "frac_reward_zero_std": 0.6,
+      "grad_norm": 4.732516117655905e-06,
+      "kl": 0.2397597748413682,
+      "learning_rate": 1.8860593681312582e-07,
+      "loss": 0.0009849418699741364,
+      "num_tokens": 1607275.0,
+      "reward": 8.940696716308593e-10,
+      "reward_std": 0.42761796951293946,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 8.940696716308593e-10,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.42761797904968263,
+      "step": 1700
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 59.24,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 59.44,
+      "completions/max_terminated_length": 59.44,
+      "completions/mean_length": 59.24,
+      "completions/mean_terminated_length": 59.24,
+      "completions/min_length": 58.64,
+      "completions/min_terminated_length": 58.64,
+      "epoch": 0.0031082929255253017,
+      "frac_reward_zero_std": 0.52,
+      "grad_norm": 0.0002946323947981,
+      "kl": 0.2524130155146122,
+      "learning_rate": 1.9415643524788528e-07,
+      "loss": -0.00041158072650432587,
+      "num_tokens": 1649715.0,
+      "reward": 1.7881393432617187e-09,
+      "reward_std": 0.5131415748596191,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.7881393432617187e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.5131415748596191,
+      "step": 1750
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 8.620689623057842e-05,
+      "clip_ratio/low_min": 8.620689623057842e-05,
+      "clip_ratio/region_mean": 8.620689623057842e-05,
+      "completion_length": 69.005,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 69.24,
+      "completions/max_terminated_length": 69.24,
+      "completions/mean_length": 69.005,
+      "completions/mean_terminated_length": 69.005,
+      "completions/min_length": 68.72,
+      "completions/min_terminated_length": 68.72,
+      "epoch": 0.003197101294826024,
+      "frac_reward_zero_std": 0.6,
+      "grad_norm": 1.7265272163058398e-06,
+      "kl": 0.2614383715391159,
+      "learning_rate": 1.997069336826447e-07,
+      "loss": 0.0002399120479822159,
+      "num_tokens": 1698684.0,
+      "reward": 2.9802322387695313e-10,
+      "reward_std": 0.42761797904968263,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.9802322387695313e-10,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.42761797904968263,
+      "step": 1800
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 72.4,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 72.76,
+      "completions/max_terminated_length": 72.76,
+      "completions/mean_length": 72.4,
+      "completions/mean_terminated_length": 72.4,
+      "completions/min_length": 72.08,
+      "completions/min_terminated_length": 72.08,
+      "epoch": 0.003285909664126747,
+      "frac_reward_zero_std": 0.44,
+      "grad_norm": 1.8452697077009361e-06,
+      "kl": 0.2667967968434095,
+      "learning_rate": 2.0525743211740417e-07,
+      "loss": 0.0009418275952339173,
+      "num_tokens": 1750300.0,
+      "reward": 2.9802322387695314e-09,
+      "reward_std": 0.5986651706695557,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.9802322387695314e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.5986651706695557,
+      "step": 1850
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 61.86,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 62.0,
+      "completions/max_terminated_length": 62.0,
+      "completions/mean_length": 61.86,
+      "completions/mean_terminated_length": 61.86,
+      "completions/min_length": 61.68,
+      "completions/min_terminated_length": 61.68,
+      "epoch": 0.00337471803342747,
+      "frac_reward_zero_std": 0.64,
+      "grad_norm": 9.591031812306028e-06,
+      "kl": 0.2360048858821392,
+      "learning_rate": 2.1080793055216363e-07,
+      "loss": 0.001219682991504669,
+      "num_tokens": 1794576.0,
+      "reward": 4.76837158203125e-09,
+      "reward_std": 0.38485618114471437,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 4.76837158203125e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.38485618114471437,
+      "step": 1900
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 7.24637694656849e-05,
+      "clip_ratio/low_min": 7.24637694656849e-05,
+      "clip_ratio/region_mean": 7.24637694656849e-05,
+      "completion_length": 67.59,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 67.76,
+      "completions/max_terminated_length": 67.76,
+      "completions/mean_length": 67.59,
+      "completions/mean_terminated_length": 67.59,
+      "completions/min_length": 67.48,
+      "completions/min_terminated_length": 67.48,
+      "epoch": 0.003463526402728193,
+      "frac_reward_zero_std": 0.56,
+      "grad_norm": 1.6016636436688714e-05,
+      "kl": 0.23783834256231784,
+      "learning_rate": 2.1635842898692306e-07,
+      "loss": 0.0007591144740581512,
+      "num_tokens": 1842646.0,
+      "reward": 2.9802322387695314e-09,
+      "reward_std": 0.4703797769546509,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.9802322387695314e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.4703797769546509,
+      "step": 1950
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.00012345679104328156,
+      "clip_ratio/low_min": 0.00012345679104328156,
+      "clip_ratio/region_mean": 0.00012345679104328156,
+      "completion_length": 61.395,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 61.64,
+      "completions/max_terminated_length": 61.64,
+      "completions/mean_length": 61.395,
+      "completions/mean_terminated_length": 61.395,
+      "completions/min_length": 61.16,
+      "completions/min_terminated_length": 61.16,
+      "epoch": 0.003552334772028916,
+      "frac_reward_zero_std": 0.6,
+      "grad_norm": 4.785554847330786e-05,
+      "kl": 0.25009929470717907,
+      "learning_rate": 2.219089274216825e-07,
+      "loss": 0.0014793139696121216,
+      "num_tokens": 1886781.0,
+      "reward": 0.0,
+      "reward_std": 0.4276179885864258,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.42761797904968263,
+      "step": 2000
+    },
+    {
+      "eval_cer_subset": 0.012620710993860635,
+      "eval_cer_subset_edit_distance": 775,
+      "eval_cer_subset_groups": 250,
+      "eval_cer_subset_items": 250,
+      "eval_cer_subset_ref_chars": 61407,
+      "eval_loss": 0.03968957871396896,
+      "step": 2026
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 2252040,
+  "num_input_tokens_seen": 1913893,
+  "num_train_epochs": 4,
+  "save_steps": 1013,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 0.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2026/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:142af57cbc15b9beb117b5972d297aef1f92226b896a513335c81b0f2c19ecf7
+size 6737

meta-llama__llama-3.2-1b/grpo/run_manifest.json

@@ -6,20 +6,26 @@
   "model_name": "meta-llama/Llama-3.2-1B",
   "status": "running",
   "best_metric_name": "eval_cer_subset",
-  "best_metric_value": 0.01309296985685671,
-  "best_step": 1013,
-  "best_checkpoint_step": 1013,
-  "latest_checkpoint_step": 1013,
+  "best_metric_value": 0.012620710993860635,
+  "best_step": 2026,
+  "best_checkpoint_step": 2026,
+  "latest_checkpoint_step": 2026,
   "checkpoints": [
     {
       "step": 1013,
       "epoch": 0.001799257562032646,
       "repo_path": "meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-1013",
       "timestamp": "2026-04-09T04:19:50.828733+00:00"
+    },
+    {
+      "step": 2026,
+      "epoch": 0.003598515124065292,
+      "repo_path": "meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2026",
+      "timestamp": "2026-04-09T04:32:13.100899+00:00"
     }
   ],
   "epoch_artifacts": [],
   "resume_history": [],
   "created_at": "2026-04-09T04:05:57.752019+00:00",
-  "updated_at": "2026-04-09T04:19:50.828749+00:00"
+  "updated_at": "2026-04-09T04:32:13.100926+00:00"
 }