Skip to content

Conversation

@Lyrcaxis
Copy link
Contributor

Distribution seed for sampling used to always be fixed (42) by default, causing deterministic sampling.
This PR switches it to be randomly initialized with each new sampling pipeline instance.

Also changed some of the sampling pipeline's members' names and comments to better reflect their purpose.

  • AlphaFrequency --> FrequencyPenalty
  • AlphaPresence --> PresencePenalty
  • PenalizeEOS --> PreventEOS
PenalizeEOS actually completely takes away the possibility for EOS_TOKEN to be sampled, causing infinite generation.

@Lyrcaxis Lyrcaxis force-pushed the distribution-seed-fix branch from 193ee52 to b943d1d Compare November 24, 2024 14:11
Copy link
Member

@martindevans martindevans left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good, just a couple of minor nits :)

@martindevans martindevans merged commit 5bce923 into SciSharp:master Nov 24, 2024
6 checks passed
@Lyrcaxis Lyrcaxis deleted the distribution-seed-fix branch November 24, 2024 20:34
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.

2 participants