ggml-org/llama.cpp

llama : combined beam search + grammar sampling strategy

Closed

#2,923 opened on Aug 31, 2023

 (14 comments) (20 reactions) (0 assignees)C++ (21,737 forks)batch import
generation qualitygood first issueresearch 🔬roadmap

Repository metrics

Stars
 (124,043 stars)
PR merge metrics
 (Avg merge 6d 8h) (389 merged PRs in 30d)

Description

This feature was proposed by @spion in https://github.com/ggerganov/llama.cpp/issues/2813#issuecomment-1694390583

In some cases, its useful to do constrained evaluation of logits based on a union of possible text values, then pick the sum { logits } (i.e. product(probabilities)) that gives the most probable outcome overall.

E.g. template (using MS guidance)

{{#select 'armor'}}leather{{or}}chainmail{{or}}plate{{/select}}

To definitely make the best choice, we'd need to calculate the probability of all 3 token sequences. Its easy if all the choices map to a single token, but with multiple tokens we'd need not just parallel generation but parallel logit evaluation of multiple possible paths.

If we go greedy, we might get suboptimal results in cases multiple choices start with the same logit.

It should be possible to implement this by combining the existing beam search and grammar sampling features. See the discussion in the referenced comment for more info

Contributor guide