bdashore3 / model-merge-tools

Originally from zaraki-tools

A simple tools for LLama

Examples

Apply Lora

python3 apply-lora.py --base_model_name_or_path "./base_model/" \
--peft_model_path "./peft_model/" \
--output_dir "/output_model/" \
--cpu_mode

Apply Lora with Weight

python3 apply-lora-weight.py --base_model_name_or_path "./base_model/" \
--peft_model_path "./peft_model/" \
--peft_model_weight 0.54 \
--output_dir "./output_model" \
--cpu_mode

Apply Lora with weight per Layer

python3 apply-lora-weight-ltl.py --base_model_name_or_path "./base_model/" \
--peft_model_path "./peft_model/" \
--peft_model_weights "0.1,0.18,0.25,0.33,0.4,0.48,0.55,0.63,0.7,0.78,0.85,0.93,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1" \
--output_dir "./output_model" \
--cpu_mode

Multiple apply lora with weight

./apply-lora-multiple.sh --base_model "./models/base_model" \
--pefts_paths "./loras/lora_1,./loras/lora2" \
--pefts_weights "0.65,0.35" \
--output_dir "./models/output_model" \
--cpu_mode

Merge Model

python3 merge-cli.py \
--first_model_path "./models/first_model" \
--second_model_path "./models/second_model" \
--merged_model_path "./models/merged_model" \
--merge_ratios "0.96,0.88,0.8,0.72,0.64,0.56,0.48,0.4,0.32,0.24,0.16,0.08,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0"