Zhongjing-LLaMA-reward-v1 / finetuning_args.json
Suprit's picture
Upload 10 files
8fa1e6e
raw
history blame contribute delete
235 Bytes
{
"finetuning_type": "lora",
"lora_alpha": 32.0,
"lora_dropout": 0.1,
"lora_rank": 8,
"lora_target": [
"q_proj",
"v_proj"
],
"name_module_trainable": "mlp",
"num_hidden_layers": 32,
"num_layer_trainable": 3
}