Mirrors/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-11-25 17:29:22 +01:00
Code Issues Packages Projects Releases Wiki Activity

Revert GPTQ_loader.py (accident)

Browse Source
This commit is contained in:
oobabooga 2023-03-24 19:57:12 -03:00 committed by GitHub
parent a6bf54739c
commit 9fa47c0eed
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 11 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
modules/GPTQ_loader.py
View File

@ -14,7 +14,7 @@ import opt
def load_quantized(model_name): def load_quantized(model_name):
if not shared.args.model_type: if not shared.args.gptq_model_type:
# Try to determine model type from model name # Try to determine model type from model name
model_type = model_name.split('-')[0].lower() model_type = model_name.split('-')[0].lower()
if model_type not in ('llama', 'opt'): if model_type not in ('llama', 'opt'):
@ -22,10 +22,10 @@ def load_quantized(model_name):
"argument") "argument")
exit() exit()
else: else:
model_type = shared.args.model_type.lower() model_type = shared.args.gptq_model_type.lower()
if model_type == 'llama': if model_type == 'llama':
if not shared.args.pre_layer: if not shared.args.gptq_pre_layer:
load_quant = llama.load_quant load_quant = llama.load_quant
else: else:
load_quant = llama_inference_offload.load_quant load_quant = llama_inference_offload.load_quant
@ -37,15 +37,15 @@ def load_quantized(model_name):
path_to_model = Path(f'models/{model_name}') path_to_model = Path(f'models/{model_name}')
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' pt_model = f'llama-7b-{shared.args.gptq_bits}bit.pt'
elif path_to_model.name.lower().startswith('llama-13b'): elif path_to_model.name.lower().startswith('llama-13b'):
pt_model = f'llama-13b-{shared.args.wbits}bit.pt' pt_model = f'llama-13b-{shared.args.gptq_bits}bit.pt'
elif path_to_model.name.lower().startswith('llama-30b'): elif path_to_model.name.lower().startswith('llama-30b'):
pt_model = f'llama-30b-{shared.args.wbits}bit.pt' pt_model = f'llama-30b-{shared.args.gptq_bits}bit.pt'
elif path_to_model.name.lower().startswith('llama-65b'): elif path_to_model.name.lower().startswith('llama-65b'):
pt_model = f'llama-65b-{shared.args.wbits}bit.pt' pt_model = f'llama-65b-{shared.args.gptq_bits}bit.pt'
else: else:
pt_model = f'{model_name}-{shared.args.wbits}bit.pt' pt_model = f'{model_name}-{shared.args.gptq_bits}bit.pt'
# Try to find the .pt both in models/ and in the subfolder # Try to find the .pt both in models/ and in the subfolder
pt_path = None pt_path = None
@ -58,10 +58,10 @@ def load_quantized(model_name):
exit() exit()
# qwopqwop200's offload # qwopqwop200's offload
if shared.args.pre_layer: if shared.args.gptq_pre_layer:
model = load_quant(str(path_to_model), str(pt_path), shared.args.wbits, shared.args.pre_layer) model = load_quant(str(path_to_model), str(pt_path), shared.args.gptq_bits, shared.args.gptq_pre_layer)
else: else:
model = load_quant(str(path_to_model), str(pt_path), shared.args.wbits) model = load_quant(str(path_to_model), str(pt_path), shared.args.gptq_bits)
# accelerate offload (doesn't work properly) # accelerate offload (doesn't work properly)
if shared.args.gpu_memory: if shared.args.gpu_memory: