Mirrors/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-11-23 00:18:20 +01:00
Code Issues Packages Projects Releases Wiki Activity

Load any 13b/20b/30b model in 8-bit mode when no flags are supplied

Browse Source
This commit is contained in:
oobabooga 2023-02-20 15:44:10 -03:00
parent c1de491c63
commit d7a738fb7a
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
server.py
View File

@ -108,7 +108,7 @@ def load_model(model_name):
# Default settings # Default settings
if not (args.cpu or args.load_in_8bit or args.auto_devices or args.disk or args.gpu_memory is not None or args.cpu_memory is not None or args.deepspeed): if not (args.cpu or args.load_in_8bit or args.auto_devices or args.disk or args.gpu_memory is not None or args.cpu_memory is not None or args.deepspeed):
if model_name.lower().startswith(('gpt-neo', 'opt-', 'galactica')) and any(size in model_name.lower() for size in ('13b', '20b', '30b')): if any(size in model_name.lower() for size in ('13b', '20b', '30b')):
model = AutoModelForCausalLM.from_pretrained(Path(f"models/{model_name}"), device_map='auto', load_in_8bit=True) model = AutoModelForCausalLM.from_pretrained(Path(f"models/{model_name}"), device_map='auto', load_in_8bit=True)
else: else:
model = AutoModelForCausalLM.from_pretrained(Path(f"models/{model_name}"), low_cpu_mem_usage=True, torch_dtype=torch.bfloat16 if args.bf16 else torch.float16).cuda() model = AutoModelForCausalLM.from_pretrained(Path(f"models/{model_name}"), low_cpu_mem_usage=True, torch_dtype=torch.bfloat16 if args.bf16 else torch.float16).cuda()