Skip to content

Conversation

@KerfuffleV2
Copy link
Contributor

@KerfuffleV2 KerfuffleV2 commented Aug 21, 2023

When converting without metadata, the hex values for bytes entries weren't 0 padded to 2 digits. (The llama.cpp changes in #2668 made handling more strict, though my output was always wrong.)

Closes #2697

When converting without metadata, the hex value for bytes entries weren't 0 padded to 2 digits.
@KerfuffleV2
Copy link
Contributor Author

KerfuffleV2 commented Aug 21, 2023

@SlyEcho Could you please give this fix a try and approve if it works for you?

Any model converted without the metadata should be affected. I'd like to get the fix merged as soon as possible before more people have issues.

edit: Confirmed to work, just needs approval.

@KerfuffleV2 KerfuffleV2 requested a review from ggerganov August 21, 2023 23:53
@klosax
Copy link
Contributor

klosax commented Aug 22, 2023

Have not tested myself, but seems to work?

@KerfuffleV2 KerfuffleV2 merged commit 930523c into ggml-org:master Aug 22, 2023
@KerfuffleV2 KerfuffleV2 removed the request for review from ggerganov August 22, 2023 00:01
@KerfuffleV2 KerfuffleV2 mentioned this pull request Aug 22, 2023
34 tasks
@SlyEcho
Copy link
Contributor

SlyEcho commented Aug 22, 2023

Yes, it works. FYI it was one of the files from OpenLLaMA 3B v2.

@KerfuffleV2 KerfuffleV2 deleted the fix-convert-ggml-to-gguf-byte-vocab branch August 27, 2023 07:17
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Silently failing ggml to gguf conversion

3 participants