Add Support for Frequency Penalties in On Device Sampling #523
+90
−61
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
Signed-off-by: quic-sanising <[email protected]>
Signed-off-by: sanising <[email protected]>
Signed-off-by: sanising <[email protected]>
Signed-off-by: sanising <[email protected]>
Signed-off-by: sanising <[email protected]>
Signed-off-by: sanising <[email protected]>
Signed-off-by: sanising <[email protected]>
Signed-off-by: sanising <[email protected]>
Signed-off-by: sanising <[email protected]>
Signed-off-by: sanising <[email protected]>
Signed-off-by: sanising <[email protected]>
Signed-off-by: sanising <[email protected]>
Signed-off-by: quic-sanising <[email protected]>
Signed-off-by: quic-sanising <[email protected]>
Signed-off-by: quic-sanising <[email protected]>
|
Depends on PR #463. |
Signed-off-by: quic-sanising <[email protected]>
quic-sanising
commented
Sep 18, 2025
| scatter_values, | ||
| torch.ones(last_accepted_output_tokens.shape, dtype=torch.bool), | ||
| ) | ||
| gather_values = past_presence_penalty_buffer[batch_index, last_accepted_output_tokens] |
There was a problem hiding this comment.
For improved performance, I intended to use CtxGatherFuncCB3D but it doesn't work as last_accepted_output_tokens is a tensor of shape (batch_size, seq_len) whereas the function expects it to be of shape (batch_size, 1). Please let me know if there is a workaround.
|
@quic-sanising is it ready for review? Can you rebase the PR? |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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.
✨ Add Frequency Penalty Support to On Device Sampling
This PR adds support for the
frequency_penaltyparameter in On Device Sampling forQEffForCausalLMmodels. This parameter adjusts token selection based on how often tokens have already appeared in the generated output:The implementation tracks token frequencies directly on the QAIC device using optimized scratch buffers, ensuring minimal overhead and maintaining high throughput. This feature integrates seamlessly with the existing
include_sampler=Trueworkflow and complements other supported strategies like repetition and presence penalties.