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

fix lora training with alpaca_lora_4bit (#3853)

Browse source
This commit is contained in:
John Smith 2023-09-11 12:22:20 +08:00 • committed by GitHub
parent 15e9b8c915
commit cc7b7ba153
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 22 additions and 30 deletions
Download patch file Download diff file Expand all files Collapse all files

20
modules/monkey_patch_gptq_lora.py
View file

@ -1,39 +1,35 @@
# Copied from https://github.com/johnsmith0031/alpaca_lora_4bit
import sys
from pathlib import Path
sys.path.insert(0, str(Path("repositories/alpaca_lora_4bit")))
import autograd_4bit
from amp_wrapper import AMPWrapper
from autograd_4bit import (
import alpaca_lora_4bit.autograd_4bit as autograd_4bit
from alpaca_lora_4bit.amp_wrapper import AMPWrapper
from alpaca_lora_4bit.autograd_4bit import (
Autograd4bitQuantLinear,
load_llama_model_4bit_low_ram
)
from monkeypatch.peft_tuners_lora_monkey_patch import (
Linear4bitLt,
replace_peft_model_with_gptq_lora_model
from alpaca_lora_4bit.models import Linear4bitLt
from alpaca_lora_4bit.monkeypatch.peft_tuners_lora_monkey_patch import (
replace_peft_model_with_int4_lora_model
)
from modules import shared
from modules.GPTQ_loader import find_quantized_model_file
replace_peft_model_with_gptq_lora_model()
replace_peft_model_with_int4_lora_model()
def load_model_llama(model_name):
config_path = str(Path(f'{shared.args.model_dir}/{model_name}'))
model_path = str(find_quantized_model_file(model_name))
model, tokenizer = load_llama_model_4bit_low_ram(config_path, model_path, groupsize=shared.args.groupsize, is_v1_model=False)
for n, m in model.named_modules():
for _, m in model.named_modules():
if isinstance(m, Autograd4bitQuantLinear) or isinstance(m, Linear4bitLt):
if m.is_v1_model:
m.zeros = m.zeros.half()
m.scales = m.scales.half()
m.bias = m.bias.half()
autograd_4bit.use_new = True
autograd_4bit.auto_switch = True
model.half()