vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

add safetensors support to convert-lora-to-ggml.py

Browse source
This commit is contained in:
kuronekosaiko 2024-01-21 18:27:18 +08:00 • committed by GitHub
parent 726c0fa9a2
commit b5bf694b23
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 8 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

9
convert-lora-to-ggml.py
View file

@ -59,7 +59,14 @@ if __name__ == '__main__':
input_model = os.path.join(sys.argv[1], "adapter_model.bin") input_model = os.path.join(sys.argv[1], "adapter_model.bin")
output_path = os.path.join(sys.argv[1], "ggml-adapter-model.bin") output_path = os.path.join(sys.argv[1], "ggml-adapter-model.bin")
model = torch.load(input_model, map_location="cpu") if os.path.exists(input_model):
model = torch.load(input_model, map_location="cpu")
else:
input_model = os.path.join(sys.argv[1], "adapter_model.safetensors")
# lazy import load_file only if lora is in safetensors format.
from safetensors.torch import load_file
model = load_file(input_model, device="cpu")
arch_name = sys.argv[2] if len(sys.argv) == 3 else "llama" arch_name = sys.argv[2] if len(sys.argv) == 3 else "llama"
if arch_name not in gguf.MODEL_ARCH_NAMES.values(): if arch_name not in gguf.MODEL_ARCH_NAMES.values():