zhengjl/text-generation-webui
Public Access
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Make LoRAs work in 16-bit mode

Browse Source
This commit is contained in:
oobabooga
2023-03-23 00:55:33 -03:00
committed by GitHub
parent bfa81e105e
commit eac27f4f55
1 changed files with 9 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
modules/LoRA.py
View File

@@ -13,10 +13,15 @@ def add_lora_to_model(lora_name):
print("Reloading the model to remove the LoRA...") print("Reloading the model to remove the LoRA...")
shared.model, shared.tokenizer = load_model(shared.model_name) shared.model, shared.tokenizer = load_model(shared.model_name)
else: else:
# Why doesn't this work in 16-bit mode?
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:
params['device_map'] = {'': 0} params['device_map'] = {'': 0}
#params['dtype'] = shared.model.dtype else:
params['device_map'] = 'auto'
params['dtype'] = shared.model.dtype
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:
shared.model.half()