chore(engine): 0.6.1 — multimodal vision fix (BOS) promoted to stable#184
Merged
Conversation
Vision is now correct on all images (portraits, vertical/dark shots, not just easy landscapes). Root cause was the missing <bos> token in the shared mtmd tokenize path (fixed in 0698e2c): Gemma requires BOS, and generate_multimodal sent the hand-built prompt without it, degrading the model into 'wall of text / line art' confabulation on harder subjects. Confirmed by an A/B of released binaries — beta.3 (no BOS) hallucinates, beta.4 (BOS) sees — on both the web/API and CLI one-shot paths. Also refresh two stale docs: the CLI --image help (the HTTP API does route media now) and the E4B catalog blurb (the 12B gemma4uv projector is served by the vendored llama-cpp-rs, no upstream bump pending).
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Vision is now correct on all images (portraits, vertical/dark shots, not just easy landscapes). Root cause was the missing token in the shared mtmd tokenize path (fixed in 0698e2c): Gemma requires BOS, and generate_multimodal sent the hand-built prompt without it, degrading the model into 'wall of text / line art' confabulation on harder subjects. Confirmed by an A/B of released binaries — beta.3 (no BOS) hallucinates, beta.4 (BOS) sees — on both the web/API and CLI one-shot paths.
Also refresh two stale docs: the CLI --image help (the HTTP API does route media now) and the E4B catalog blurb (the 12B gemma4uv projector is served by the vendored llama-cpp-rs, no upstream bump pending).