Repository navigation
Make gemma4:e2b the default chat model - #250
Merged
Merged
Conversation
llama3.2 missed 2 of 8 grounded questions on `make eval-chat` (one false refusal, one miscite). gemma4:e2b cleared all 8 across three runs at about 2.5 s to first token, the fastest of every model that cleared the set (qwen3:4b, qwen3:8b, gemma4:e4b, lfm3, muse-glimmer:30b were also run; rows are in crates/b2-llm/evals/results.jsonl). The picker's 32 GB rung goes: the only model that earned it, muse-glimmer 30B, takes 26 s to first token, too slow to recommend. Two rungs remain, gemma4:e2b and gemma4:e4b. Thinking on/off is the lever behind the gap and is tracked as #249. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
Note Currently processing new changes in this PR. This may take a few minutes, please wait...
 |
…rify, undici Rust: candle 0.9 pinned zip ^7.2, which carried the critical alert (fixed in 9.0.5); candle 0.11.1 moves it to zip 9.4.1. rustls goes to 0.23.46 for its two advisories. `make eval` on both corpora prints the same rows as the last run on main, so the bump does not change the embedding space (ADR-0007). `make ci` is green. npm: vite 6.4.5, dompurify 3.4.17 and undici 7.30.2 (via jsdom), all within the ranges package.json already allows. Left open: glib 0.18 on Linux, pulled in by gtk 0.18 under Tauri 2. It cannot move until Tauri does. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Why
Research scan on 2026-11-07: llama3.2 (2024) is behind current small models. Measured with
make eval-chaton the 11-question groundedness set rather than by leaderboard.What changed
DEFAULT_MODELisgemma4:e2b. Docs, CLI help and the Settings placeholder follow.MODEL_TIERSdrops the 32 GB rung (muse-glimmer at 26 s is too slow to recommend) and reads gemma4:e2b / gemma4:e4b.Checked
make checkgreen. Eval rows appended tocrates/b2-llm/evals/results.jsonl(untracked, per the harness convention).🤖 Generated with Claude Code