42 lines
1.3 KiB
JSON
42 lines
1.3 KiB
JSON
|
{
|
||
|
|
"model_name": "A.X-3.1-Light-sft_v3_0",
|
||
|
|
"base_model": "skt/A.X-3.1-Light",
|
||
|
|
"base_revision": "9b41bb2406472634d8812c0b8931fa40fa9a6c3a",
|
||
|
|
"finetuning_method": "LoRA SFT followed by merge_and_unload",
|
||
|
|
"merge_output": "standalone BF16 safetensors",
|
||
|
|
"custom_code_required": false,
|
||
|
|
"training": {
|
||
|
|
"train_examples": 36000,
|
||
|
|
"validation_examples": 3000,
|
||
|
|
"epochs": 1,
|
||
|
|
"learning_rate": 0.00005,
|
||
|
|
"scheduler": "cosine",
|
||
|
|
"warmup_ratio": 0.03,
|
||
|
|
"weight_decay": 0.01,
|
||
|
|
"effective_batch_size": 32,
|
||
|
|
"max_sequence_length": 2048,
|
||
|
|
"lora_rank": 16,
|
||
|
|
"lora_alpha": 32,
|
||
|
|
"lora_dropout": 0.05,
|
||
|
|
"target_modules": ["q_proj", "k_proj", "v_proj", "o_proj"],
|
||
|
|
"precision": "bfloat16",
|
||
|
|
"packing": false,
|
||
|
|
"seed": 20260825,
|
||
|
|
"loss": "standard assistant-token causal LM cross entropy"
|
||
|
|
},
|
||
|
|
"training_data": {
|
||
|
|
"source": "AI Hub domain-specific Korean datasets",
|
||
|
|
"public_benchmark_data_included": false,
|
||
|
|
"train_validation_source_overlap": 0,
|
||
|
|
"train_examples_by_output_contract": {
|
||
|
|
"EXPLANATION": 19643,
|
||
|
|
"SHORT_ANSWER": 7583,
|
||
|
|
"NUMBER_ONLY": 5176,
|
||
|
|
"LABEL_ONLY": 2798,
|
||
|
|
"REASONING_FINAL": 800
|
||
|
|
},
|
||
|
|
"reasoning_final_output_token_share": 0.243691
|
||
|
|
},
|
||
|
|
"created_at_utc": "2026-08-26T10:46:37Z"
|
||
|
|
}
|