24 lines
861 B
JSON
24 lines
861 B
JSON
|
{
|
||
|
|
"model_id": "Qwen/Qwen3-0.6B-Base",
|
||
|
|
"run_name": "qwen3-0.6b-math-l45-qlora-sft-v2",
|
||
|
|
"hf_repo_id": "zannvznn/qwen3-0.6b-math-l45-qlora-merged-fp16-v2",
|
||
|
|
"train_file": "/kaggle/input/datasets/anurhalizah/math-he/math_level45_train.parquet",
|
||
|
|
"test_file": "/kaggle/input/datasets/anurhalizah/math-he/math_level45_test.parquet",
|
||
|
|
"train_subset_size": 3994,
|
||
|
|
"train_rows_used": 3994,
|
||
|
|
"loss_eval_rows_used": 200,
|
||
|
|
"max_seq_length": 2048,
|
||
|
|
"num_train_epochs": 3,
|
||
|
|
"learning_rate": 0.0002,
|
||
|
|
"per_device_train_batch_size": 1,
|
||
|
|
"per_device_eval_batch_size": 1,
|
||
|
|
"gradient_accumulation_steps": 8,
|
||
|
|
"fp16": true,
|
||
|
|
"bf16": false,
|
||
|
|
"qlora": true,
|
||
|
|
"shuffle_train": true,
|
||
|
|
"world_size": 1,
|
||
|
|
"device_map": "balanced",
|
||
|
|
"max_memory": "{0: 13958643712, 1: 13958643712}",
|
||
|
|
"final_eval_note": "Final 500-question evaluation is intentionally separate."
|
||
|
|
}
|