train method
supervised fine-tuning
Reward Modeling
PPO training
DPO training
full-parameter
partial-parameter
LoRA
QLoRA
command parameter
fp16
gradient_accumulation_steps
lr_scheduler_type
lora_target
overwrite_cache
stage
llama-factory fine-tuning-3 (conception and technologies explanation)
发布时间 2023-11-29 15:45:50作者: Daze_Lu