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

Some minor fixes to the GPTQ loader

Browse source
This commit is contained in:
oobabooga 2023-03-13 16:45:08 -03:00
parent 8778b756e6
commit 518e5c4244
1 changed files with 5 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

8
modules/quant_loader.py
View file

@ -7,6 +7,8 @@ import torch
import modules.shared as shared import modules.shared as shared
sys.path.insert(0, str(Path("repositories/GPTQ-for-LLaMa"))) sys.path.insert(0, str(Path("repositories/GPTQ-for-LLaMa")))
import llama
import opt
def load_quantized(model_name): def load_quantized(model_name):
@ -21,9 +23,9 @@ def load_quantized(model_name):
model_type = shared.args.gptq_model_type.lower() model_type = shared.args.gptq_model_type.lower()
if model_type == 'llama': if model_type == 'llama':
from llama import load_quant load_quant = llama.load_quant
elif model_type == 'opt': elif model_type == 'opt':
from opt import load_quant load_quant = opt.load_quant
else: else:
print("Unknown pre-quantized model type specified. Only 'llama' and 'opt' are supported") print("Unknown pre-quantized model type specified. Only 'llama' and 'opt' are supported")
exit() exit()
@ -50,7 +52,7 @@ def load_quantized(model_name):
print(f"Could not find {pt_model}, exiting...") print(f"Could not find {pt_model}, exiting...")
exit() exit()
model = load_quant(path_to_model, str(pt_path), shared.args.gptq_bits) model = load_quant(str(path_to_model), str(pt_path), shared.args.gptq_bits)
# Multiple GPUs or GPU+CPU # Multiple GPUs or GPU+CPU
if shared.args.gpu_memory: if shared.args.gpu_memory: