Feat/UI math rendering#174
Merged
Merged
Conversation
… model Symptom: a user who pulled gemma-4-12b before the catalog gained mmproj_repo/mmproj_filename ended up with the GGUF on disk but no multimodal projector. Subsequent `eullm pull gemma-4-12b` short-circuited on "already downloaded" and `eullm run --image` failed with "Multimodal not configured: load the model with a valid mmproj_path". The only workaround was to delete the model and re-download ~7 GB. Two small changes, no public API change: * Extract the existing mmproj download flow from cmd_pull into a `download_mmproj(entry, model_dir)` helper, idempotent: if the file is already on disk, return its name without redownloading. Caller behaviour for the first pull is unchanged. * In the "already downloaded" branch of cmd_pull, if the catalog declares an mmproj and `store.mmproj_path` finds none, call the helper and update the manifest's mmproj_file field so `list`/`run` see the recovery on subsequent runs. clippy clean (CI flags), 25 tests pass.
Gemma 4 E4B was released in March 2026 as part of the original Gemma 4 family — its multimodal projector uses the pre-unified projector type that is already supported by llama-cpp-2 v0.1.146. Unlike the 12B Unified variant (June 3 2026, new `gemma4uv` projector that requires an upstream bump we don't have yet), E4B is the right immediate target for testing the EULLM multimodal pipeline on the current binary. Add mmproj_repo/mmproj_filename pointing at unsloth/gemma-4-E4B-it-GGUF mmproj-F16.gguf (990 MB) and update the description + tags accordingly. License remains Apache 2.0.
The 0.6.0-beta.* series was bumped to ship Gemma 4 12B Unified vision via the new mtmd path. Upstream llama-cpp-2 (v0.1.146, 30 Apr 2026) is still two months behind the llama.cpp commits that add the `gemma4uv` projector (merged 3-4 Jun 2026 as PRs #24077, #24082, #24091). Until utilityai bumps the wrapper crate, that specific model cannot load on our binary. Rather than ship 0.6.0 stable without the headline feature that motivated its existence, release the accumulated fixes as a 0.5.x patch and keep the 0.6.0 milestone for when the upstream bump arrives and Gemma 4 12B Unified actually works end-to-end. Stable contents 0.5.20: - UI: Markdown-lite (headings, lists, bold, italic, hr, blockquote) + best-effort LaTeX→MathML rendering with orphan \frac/\sqrt recovery and tightened spacing. - UI/engine: Gemma 4 channel-thought block surfaced as Reasoning section. - Engine: catalog id used as the single addressable model name across /api/tags, /v1/models, `eullm list`, and the picker. - Engine: picker tags catalog entries already pulled with `[local]`. - Engine: `eullm pull` recovers a missing mmproj for an already-downloaded catalog model. - Catalog: gemma-4-e4b marked vision-capable (pre-unified projector type, works with the current llama-cpp-2). - CI: multimodal (mtmd) enabled in the Windows CUDA release binary. Known gap (tracked, will be the 0.6.0 milestone): - Gemma 4 12B Unified vision is pending the upstream llama-cpp-2 bump to a llama.cpp commit including PR #24077+.
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.
No description provided.