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

Fix LlamaTokenizer eos_token (attempt)

Browse source
This commit is contained in:
oobabooga 2023-04-09 21:19:39 -03:00
parent 120f5662cf
commit a3085dba07
1 changed files with 3 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

3
modules/models.py
View file

@ -174,6 +174,9 @@ def load_model(model_name):
tokenizer = AutoTokenizer.from_pretrained(Path(f"{shared.args.model_dir}/gpt-j-6B/")) tokenizer = AutoTokenizer.from_pretrained(Path(f"{shared.args.model_dir}/gpt-j-6B/"))
elif type(model) is transformers.LlamaForCausalLM: elif type(model) is transformers.LlamaForCausalLM:
tokenizer = LlamaTokenizer.from_pretrained(Path(f"{shared.args.model_dir}/{shared.model_name}/"), clean_up_tokenization_spaces=True) tokenizer = LlamaTokenizer.from_pretrained(Path(f"{shared.args.model_dir}/{shared.model_name}/"), clean_up_tokenization_spaces=True)
tokenizer.eos_token_id = 2
tokenizer.bos_token_id = 1
tokenizer.pad_token_id = 0
else: else:
tokenizer = AutoTokenizer.from_pretrained(Path(f"{shared.args.model_dir}/{shared.model_name}/")) tokenizer = AutoTokenizer.from_pretrained(Path(f"{shared.args.model_dir}/{shared.model_name}/"))
tokenizer.truncation_side = 'left' tokenizer.truncation_side = 'left'