Skip to content

chore: benchmark flags for dataset config and implementation subset#64

Merged
AmitMY merged 1 commit into
mainfrom
chore/bench-dataset-flags
Jul 24, 2026
Merged

chore: benchmark flags for dataset config and implementation subset#64
AmitMY merged 1 commit into
mainfrom
chore/bench-dataset-flags

Conversation

@AmitMY

@AmitMY AmitMY commented Jul 24, 2026

Copy link
Copy Markdown
Contributor

Two small flags for scaling the benchmark beyond wikitext-2:

  • --dataset-config (default unchanged: wikitext-2-raw-v1) — e.g. wikitext-103-raw-v1 for the ~50× corpus; combined with --samples this gives any size slice.
  • --impls — comma-separated subset (e.g. "HuggingFace,fast (Rust)") to skip the reference at scales where its connected cases would take many minutes.

Default invocation behavior is unchanged. First use: the 10× comparison (fast vs HuggingFace on a ~109M-char wikitext-103 slice, 500 merges) — results to follow as a comment.

🤖 Generated with Claude Code

--dataset-config selects e.g. wikitext-103-raw-v1 for larger-corpus runs;
--impls runs a subset (e.g. skip the reference at scales where its
connected cases would take many minutes).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
@AmitMY
AmitMY merged commit 90ae664 into main Jul 24, 2026
3 checks passed
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.

1 participant