| { |
| "base_model": "Qwen/Qwen3-4B", |
| "base_model_revision": "1cfa9a7208912126459214e8b04321603b3df60c", |
| "task": "bigcodebench", |
| "method": "supervised fine-tuning (LoRA)", |
| "enable_thinking": false, |
| "training_conversations": 450, |
| "validation_conversations": 50, |
| "global_step": 24, |
| "epochs_completed": 3.0, |
| "data_source": "https://github.com/Wuyxin/collabllm/tree/d573e4949abdc301cb2b630525c593cc432e94b8/data/interactive-bigcodebench", |
| "data_preparation": "For each dialogue idx, select the row with the longest prompt and append its chosen assistant response. Hold out 50 of 500 conversations using seed 42.", |
| "adapter_sha256": "819b83c6722ab0bca4290a10a88260ac6a3b24f677f14b4720ce0e284351e437", |
| "hyperparameters": { |
| "seed": 42, |
| "learning_rate": 1e-05, |
| "warmup_ratio": 0.1, |
| "max_seq_length": 1536, |
| "num_train_epochs": 3, |
| "per_device_train_batch_size": 1, |
| "gradient_accumulation_steps": 64, |
| "assistant_only_loss": true, |
| "peft_r": 32, |
| "peft_alpha": 16, |
| "peft_dropout": 0.1, |
| "target_modules": "q_proj,k_proj,v_proj,o_proj,gate_proj,up_proj,down_proj", |
| "use_4bit": false |
| }, |
| "training_metrics": { |
| "total_flos": 4.213281476994662e+16, |
| "train_loss": 0.7953540682792664, |
| "train_runtime": 1310.0688, |
| "train_samples_per_second": 1.03, |
| "train_steps_per_second": 0.018 |
| }, |
| "validation_metrics": { |
| "eval_loss": 0.7820074558258057, |
| "eval_runtime": 16.4949, |
| "eval_samples_per_second": 3.031, |
| "eval_steps_per_second": 3.031 |
| } |
| } |
|
|
