Removing fsep token from GPTRefactForCausalLM #8237
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
mofosyne
added
medium severity
compilade
approved these changes
Jul 12, 2024
convert-hf-to-gguf.py
Outdated
| special_vocab._set_special_token("prefix", 1) | ||
| special_vocab._set_special_token("suffix", 3) | ||
| special_vocab._set_special_token("middle", 2) | ||
| special_vocab._set_special_token("fsep", 4) # is this correct? |
There was a problem hiding this comment.
Token id 4 in Refact's tokenizer seems to be <fim_pad> anyway, which doesn't look like a separator, so it was not correct.
There was a problem hiding this comment.
Thanks. I've rebased the PR and resolved conflicts.
The <filename> token used by Refact doesn't serve the same purpose as the <file_separator> from CodeGemma. Signed-off-by: Jiri Podivin <jpodivin@redhat.com>
arthw
pushed a commit
to arthw/llama.cpp
that referenced
this pull request
Jul 13, 2024
The <filename> token used by Refact doesn't serve the same purpose as the <file_separator> from CodeGemma. Signed-off-by: Jiri Podivin <jpodivin@redhat.com>
arthw
pushed a commit
to arthw/llama.cpp
that referenced
this pull request
Jul 13, 2024
The <filename> token used by Refact doesn't serve the same purpose as the <file_separator> from CodeGemma. Signed-off-by: Jiri Podivin <jpodivin@redhat.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The
<filename>token used by Refact doesn't serve the same purpose as the<file_separator>from CodeGemma.I've been going through docs for both Refact and CodeGemma, and came to conclusion that
<filename>isn't a good fit for the file separator token. The biggest issue is the lack of documentation on the matter.Only mention of the
<filename>token on Huggingface and other sources comes from description of Starcoder training[0] where it is used to mark start of filename for prompt. This is substantially different from the way it is used by CodeGemma[1].Repos of Small Magellanic Cloud AI on github and HF don't provide more information.
With that in mind, I believe it's better not to set it as fsep token.
[0] https://huggingface.co/blog/starcoder#evaluation
[1] https://ai.google.dev/gemma/docs/formatting