mirror of
https://github.com/hiyouga/LLaMA-Factory.git
synced 2025-08-02 11:42:49 +08:00
13 lines
430 B
Bash
13 lines
430 B
Bash
#!/bin/bash
|
|
# DO NOT use quantized model or quantization_bit when merging lora weights
|
|
|
|
CUDA_VISIBLE_DEVICES=0 llamafactory-cli export \
|
|
--model_name_or_path meta-llama/Llama-2-7b-hf \
|
|
--adapter_name_or_path ../../saves/LLaMA2-7B/lora/sft \
|
|
--template default \
|
|
--finetuning_type lora \
|
|
--export_dir ../../models/llama2-7b-sft \
|
|
--export_size 2 \
|
|
--export_device cpu \
|
|
--export_legacy_format False
|