mmarch/text-generation-webui
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Fix LoRA device map (attempt)

Browse source
This commit is contained in:
oobabooga 2023-03-23 16:49:41 -03:00 • committed by GitHub
parent c5ebcc5f7e
commit 9bf6ecf9e2
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 7 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

11
modules/LoRA.py
View file

@ -16,12 +16,15 @@ def add_lora_to_model(lora_name):
print(f"Adding the LoRA {lora_name} to the model...") print(f"Adding the LoRA {lora_name} to the model...")
params = {} params = {}
if shared.args.load_in_8bit: if not shared.args.cpu:
params['device_map'] = {'': 0}
elif not shared.args.cpu:
params['device_map'] = 'auto'
params['dtype'] = shared.model.dtype params['dtype'] = shared.model.dtype
if hasattr(shared.model, "hf_device_map"):
params['device_map'] = {"base_model.model."+k: v for k, v in shared.model.hf_device_map.items()}
elif shared.args.load_in_8bit:
params['device_map'] = {'': 0}
shared.model = PeftModel.from_pretrained(shared.model, Path(f"loras/{lora_name}"), **params) shared.model = PeftModel.from_pretrained(shared.model, Path(f"loras/{lora_name}"), **params)
if not shared.args.load_in_8bit and not shared.args.cpu: if not shared.args.load_in_8bit and not shared.args.cpu:
shared.model.half() shared.model.half()
if not hasattr(shared.model, "hf_device_map"):
shared.model.cuda()