Llama-2-13B-fp16-french / QLoRA /finetuning_args.json
Nekochu's picture
QLoRA merge
719771a
raw
history blame
233 Bytes
{
"finetuning_type": "lora",
"lora_alpha": 32.0,
"lora_dropout": 0,
"lora_rank": 8,
"lora_target": [
"q_proj",
"v_proj"
],
"name_module_trainable": "mlp",
"num_hidden_layers": 32,
"num_layer_trainable": 3
}