============================================================
Refinement started at 2026-06-14 06:18:05
============================================================
Logging to ./refined_adapters/python/refine.log
Args: Namespace(language='python', results_dir='ankhanhtran02/executed_calibration_results', results_source='hf_hub', results_repo_type='model', adapter_path='ankhanhtran02/lora-per-task-executable-start-4', model_name_or_path='Qwen/Qwen2.5-Coder-1.5B', output_dir='./refined_adapters/python', per_device_train_batch_size=4, num_train_epochs=3, learning_rate=5e-05, weight_decay=0.01, gradient_accumulation_steps=4, num_warmup_steps=0, max_prompt_len=1024, max_ans_len=1024, seed=42, logging_steps=5, local_rank=0, offload=False, zero_stage=0, gradient_checkpointing=False, disable_dropout=False, run_name='refine_python', group_name='refine_adapter', enable_wandb=False, enable_tensorboard=False, tensorboard_path='refine_tensorboard', deepspeed=True, deepspeed_config=None, deepscale=False, deepscale_config=None, global_rank=0)
[refine] Loaded adapter from HF Hub: ankhanhtran02/lora-per-task-executable-start-4/python/0
trainable params: 1,867,776 || all params: 1,545,176,576 || trainable%: 0.1208778355180036
[repair] extracted 195 pairs (skipped: 231 no-fail-traceback, 0 no-label)
[refine] 195 repair pairs for language=python
Time to load fused_adam op: 1.9950942993164062 seconds
***** Running adapter refinement *****
  language      = python
  train pairs   = 195
  epochs        = 3
  steps/epoch   = 17
  total steps   = 51
  batch/device  = 4
  grad accum    = 4
  lr            = 5e-05
[epoch 1/3] step 5/17 global_step 5 loss 0.5386
[epoch 1/3] step 10/17 global_step 10 loss 0.4107
[epoch 1/3] step 15/17 global_step 15 loss 0.5793
[refine] Epoch 1/3 complete.
[epoch 2/3] step 3/17 global_step 20 loss 0.4969
[epoch 2/3] step 8/17 global_step 25 loss 0.5477
[epoch 2/3] step 13/17 global_step 30 loss 0.7573
[refine] Epoch 2/3 complete.
[epoch 3/3] step 1/17 global_step 35 loss 0.5654
[epoch 3/3] step 6/17 global_step 40 loss 0.4474
[epoch 3/3] step 11/17 global_step 45 loss 0.5370
[epoch 3/3] step 16/17 global_step 50 loss 0.3835
[refine] Epoch 3/3 complete.
[refine] Saved fine-tuned adapter to ./refined_adapters/python

============================================================
Refinement started at 2026-06-14 23:53:30
============================================================
Logging to ./refined_adapters/python/refine.log
Args: Namespace(language='python', results_dir='ankhanhtran02/executed_calibration_results', results_source='hf_hub', results_repo_type='model', adapter_path='ankhanhtran02/lora-per-task-executable-start-4', model_name_or_path='Qwen/Qwen2.5-Coder-1.5B', output_dir='./refined_adapters/python', per_device_train_batch_size=4, num_train_epochs=3, learning_rate=5e-05, weight_decay=0.01, gradient_accumulation_steps=4, num_warmup_steps=0, max_prompt_len=1024, max_ans_len=2048, seed=42, logging_steps=5, local_rank=0, offload=False, zero_stage=0, gradient_checkpointing=False, disable_dropout=False, run_name='refine_python', group_name='refine_adapter', enable_wandb=False, enable_tensorboard=False, tensorboard_path='refine_tensorboard', deepspeed=True, deepspeed_config=None, deepscale=False, deepscale_config=None, global_rank=0)
[refine] Loaded adapter from HF Hub: ankhanhtran02/lora-per-task-executable-start-4/python/0
trainable params: 1,867,776 || all params: 1,545,176,576 || trainable%: 0.1208778355180036
[repair] extracted 195 pairs (skipped: 231 no-fail-traceback, 0 no-label)
[refine] 195 repair pairs for language=python
Time to load fused_adam op: 1.7055249214172363 seconds
***** Running adapter refinement *****
  language      = python
  train pairs   = 195
  epochs        = 3
  steps/epoch   = 25
  total steps   = 75
  batch/device  = 4
  grad accum    = 4
  lr            = 5e-05
[epoch 1/3] step 5/25 global_step 5 loss 0.5697
[epoch 1/3] step 10/25 global_step 10 loss 0.3202
[epoch 1/3] step 15/25 global_step 15 loss 0.3142
[epoch 1/3] step 20/25 global_step 20 loss 0.6230
[epoch 1/3] step 25/25 global_step 25 loss 0.3026
[refine] Epoch 1/3 complete.
[epoch 2/3] step 5/25 global_step 30 loss 0.5361
[epoch 2/3] step 10/25 global_step 35 loss 0.5786
[epoch 2/3] step 15/25 global_step 40 loss 0.3814
[epoch 2/3] step 20/25 global_step 45 loss 0.6810
[epoch 2/3] step 25/25 global_step 50 loss 0.2489
[refine] Epoch 2/3 complete.
[epoch 3/3] step 5/25 global_step 55 loss 0.2804
[epoch 3/3] step 10/25 global_step 60 loss 0.3424
[epoch 3/3] step 15/25 global_step 65 loss 0.7264
[epoch 3/3] step 20/25 global_step 70 loss 0.3545
[epoch 3/3] step 25/25 global_step 75 loss 0.4598
[refine] Epoch 3/3 complete.
[refine] Saved fine-tuned adapter to ./refined_adapters/python