tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

fix issue with completed model being empty

Browse Source 
see https://github.com/huggingface/peft/issues/286#issuecomment-1501617281
This commit is contained in:
Wing Lian
2023-04-14 23:57:55 -04:00
parent 80b2ed29d8
commit 902dd0ab47
1 changed files with 3 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
scripts/finetune.py
View File

@@ -369,22 +369,18 @@ def train(
) )
model.config.use_cache = False model.config.use_cache = False
old_state_dict = model.state_dict
model.state_dict = (
lambda self, *_, **__: get_peft_model_state_dict(self, old_state_dict())
).__get__(model, type(model))
if torch.__version__ >= "2" and sys.platform != "win32": if torch.__version__ >= "2" and sys.platform != "win32":
model = torch.compile(model) model = torch.compile(model)
# go ahead and presave, so we have the adapter config available to inspect
lora_config.save_pretrained(cfg.output_dir)
# In case we want to stop early with ctrl+c, this is a nice to have to save the pretrained model # In case we want to stop early with ctrl+c, this is a nice to have to save the pretrained model
signal.signal( signal.signal(
signal.SIGINT, signal.SIGINT,
lambda signal, frame: (model.save_pretrained(cfg.output_dir), exit(0)), lambda signal, frame: (model.save_pretrained(cfg.output_dir), exit(0)),
) )
# go ahead and presave the adapter config
lora_config.save_pretrained(cfg.output_dir)
trainer.train(resume_from_checkpoint=cfg.resume_from_checkpoint) trainer.train(resume_from_checkpoint=cfg.resume_from_checkpoint)
model.save_pretrained(cfg.output_dir) model.save_pretrained(cfg.output_dir)