logit_bias: apply configurable escalating EOG bias at low n_remain #14229
Conversation
This file contains hidden or 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
graehl
requested review from
JohannesGaessler,
am17an,
ggerganov and
slaren
as code owners
give eog an increasing (with length - per token, could be per
codepoint in future) bias, only after a configured amount generated
add to `sample_apply` an `n_remain` param, which is safer than having
logit_bias maintain state for how many times it's called (which
would lead to wrong assumptions e.g. when calling multiple times per
token).
see new command line options (incl a request 'after' instead of
'remain'):
-eog, --eog-bias-per-tok N when fewer than -start-eog-at-remain tokens are left to generate after
-n, add this bias eog for each subsequent token (default: 0.0)
-remain, --start-eog-at-remain N start applying -eog bias when this many tokens remain of the -n max
(default: 0.0)
-after, --start-eog-after N start applying -eog bias after this many tokens generated (default:
1000000000.0); whichever happens first between -remain and -after
applies
Verified that eog bias was effective at avoiding
overgeneration and is a reasonable supplement or alternative
to editing the prompt; a *constant* eog bias, already supported
in samplers, is likely to allow pathologically short outputs.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.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.
give eog an increasing (with length - per token, could be per codepoint in future) bias, only after a configured amount generated
add to
sample_applyann_remainparam, which is safer than having logit_bias maintain state for how many times it's called (which would lead to wrong assumptions e.g. when calling multiple times per token).see new command line options (incl a request 'after' instead of 'remain'):
-eog, --eog-bias-per-tok N when fewer than -start-eog-at-remain tokens are left to generate after
-n, add this bias eog for each subsequent token (default: 0.0)
-remain, --start-eog-at-remain N start applying -eog bias when this many tokens remain of the -n max
(default: 0.0)
-after, --start-eog-after N start applying -eog bias after this many tokens generated (default:
1000000000.0); whichever happens first between -remain and -after
applies
Verified that eog bias was effective at avoiding
overgeneration and is a reasonable supplement or alternative to editing the prompt; a constant eog bias, already supported in samplers, is likely to allow pathologically short outputs.
Make sure to read the contributing guidelines before submitting a PR