Mirrors/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-12-23 21:18:00 +01:00
Code Issues Packages Projects Releases Wiki Activity

Use oasst tokenizer instead of universal tokenizer

Browse Source
This commit is contained in:
oobabooga 2023-05-04 15:55:39 -03:00
parent b6ff138084
commit 97a6a50d98
2 changed files with 3 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
models/config.yaml
View File

@ -24,7 +24,6 @@ llama-[0-9]*b-4bit$:
.*(oasst-sft-1-pythia-12b|oasst-sft-6-llama-30b):
mode: 'instruct'
instruction_template: 'Open Assistant'
custom_stopping_strings: '"<|endoftext|>"'
.*vicuna:
mode: 'instruct'
instruction_template: 'Vicuna-v0'

4
modules/models.py
View File

@ -54,6 +54,8 @@ def find_model_type(model_name):
return 'galactica'
elif 'llava' in model_name_lower:
return 'llava'
elif 'oasst' in model_name_lower:
return 'oasst'
elif any((k in model_name_lower for k in ['gpt4chan', 'gpt-4chan'])):
return 'gpt4chan'
else:
@ -227,7 +229,7 @@ def load_model(model_name):
tokenizer = None
# Try to load an universal LLaMA tokenizer
if shared.model_type != 'llava':
if shared.model_type not in ['llava', 'oasst']:
for p in [Path(f"{shared.args.model_dir}/llama-tokenizer/"), Path(f"{shared.args.model_dir}/oobabooga_llama-tokenizer/")]:
if p.exists():
logging.info(f"Loading the universal LLaMA tokenizer from {p}...")