AI/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-09-20 00:07:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

More robust 4-bit model loading

Browse Source
This commit is contained in:
oobabooga 2023-04-09 23:19:28 -03:00
parent 625d81f495
commit 8c6155251a
1 changed files with 6 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
modules/GPTQ_loader.py
View File

@ -100,10 +100,10 @@ def load_quantized(model_name):
found_safetensors = list(path_to_model.glob("*.safetensors")) found_safetensors = list(path_to_model.glob("*.safetensors"))
pt_path = None pt_path = None
if len(found_pts) == 1: if len(found_pts) > 0:
pt_path = found_pts[0] pt_path = found_pts[-1]
elif len(found_safetensors) == 1: elif len(found_safetensors) > 0:
pt_path = found_safetensors[0] pt_path = found_safetensors[-1]
else: else:
if path_to_model.name.lower().startswith('llama-7b'): if path_to_model.name.lower().startswith('llama-7b'):
pt_model = f'llama-7b-{shared.args.wbits}bit' pt_model = f'llama-7b-{shared.args.wbits}bit'
@ -119,13 +119,14 @@ def load_quantized(model_name):
# Try to find the .safetensors or .pt both in the model dir and in the subfolder # Try to find the .safetensors or .pt both in the model dir and in the subfolder
for path in [Path(p + ext) for ext in ['.safetensors', '.pt'] for p in [f"{shared.args.model_dir}/{pt_model}", f"{path_to_model}/{pt_model}"]]: for path in [Path(p + ext) for ext in ['.safetensors', '.pt'] for p in [f"{shared.args.model_dir}/{pt_model}", f"{path_to_model}/{pt_model}"]]:
if path.exists(): if path.exists():
print(f"Found {path}")
pt_path = path pt_path = path
break break
if not pt_path: if not pt_path:
print("Could not find the quantized model in .pt or .safetensors format, exiting...") print("Could not find the quantized model in .pt or .safetensors format, exiting...")
exit() exit()
else:
print(f"Found the following quantized model: {pt_path}")
# qwopqwop200's offload # qwopqwop200's offload
if model_type == 'llama' and shared.args.pre_layer: if model_type == 'llama' and shared.args.pre_layer: