Make compress_pos_emb float #6276
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
Originally, compress_pos_emb had a slider gradation of 0.01. However, webui now uses gr.Number instead of gr.Slider to represent the value (to avoid having to repeatedly update max value limit) and the precision was set to 0. Setting precision to 0.01 to return to previous behaviour.
hocjordan
force-pushed
the
make_compress_pos_emb_float
branch
from
314283f to
221ec15
Compare
|
Thank you! |
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Outdated suggestions cannot be applied.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.
compress_pos_emboriginally had a slider gradation of 0.01, allowing the user to set float values, but this behaviour was changed in #6233. The UI now usesgr.Numberinstead ofgr.Sliderto represent RoPE scaling values so that devs don't have to repeatedly update the max value limit and as a side effect the gradation ofcompress_pos_embbecame integer.I find float values useful (a value of 1.3 allows Llama 3.0 models to work nicely with a 12k context, whereas 1 is insufficient and 2 causes noticeable drop in quality). This PR therefore replicates the previous behaviour by setting the precision to 2.
I have tested the change by setting
compress_pos_embto 1.3 and confirming that the value sent tollama.cppis correct, i.e.I don't believe that this will cause problems with the rest of the code because float values were originally permitted.
Checklist: