Files
llama3.1-8b-base-gsm8k-safe…/finetune_config.json
ModelHub XC afac08d916 初始化项目,由ModelHub XC社区提供模型
Model: kmseong/llama3.1-8b-base-gsm8k-safeinstr-ratio0.1-lr1e-5
Source: Original Platform
2026-05-23 05:27:15 +08:00

21 lines
635 B
JSON

{
"base_model": "kmseong/Llama-3.1-8B-base-SSFT_lr5e-5",
"fine_tuning_type": "Full Parameter Fine-tuning",
"dataset": "GSM8K",
"num_train_samples": 7473,
"batch_size": 4,
"grad_accum": 4,
"learning_rate": 1e-05,
"weight_decay": 0.01,
"warmup_ratio": 0.1,
"epochs": 3,
"max_length": 1024,
"max_grad_norm": 1.0,
"lr_scheduler_type": "cosine",
"optimizer": "AdamW (torch)",
"gradient_checkpointing": false,
"dtype": "bf16",
"trainer_type": "Trainer",
"safety_mix_ratio": 0.1,
"safety_data_path": "/NHNHOME/WORKSPACE/26msit001_A/edge_ai_lab/minseong/Safety-WaRP-LLM/data/circuit_breakers_train.json"
}