Upload merges.txt
#1
by
mike-ravkine
- opened
Loading this model with autogptq fails:
│ ❱ 194 with open(merges_file, encoding="utf-8") as merges_handle: │
This text file is part of the weird tokenizer setup this model requires
Oh, thank you. I tested this with AutoGPTQ and didn't get this error.
TheBloke
changed pull request status to
merged
When you tested did you use use_fast=False tokenizer or use_fast=True tokenizer?
This file is only required for use_fast=False as far as I can tell. I can confirm adding it makes everything happy.
I honestly can't remember - does the issue only occur with one or the other?
Yes, it seem this file is only read by use_fast=False
Both should now work.
Great, thanks for the PR. I will remember to add this file for future models of this type