Salvage safe bits from reverted MoA PR (#820) - #824
Merged
Conversation
Re-applies the isolated, low-risk improvements from #820 without the MoA fanout/context changes that hung the mesh route: - catalog.json (node + client): add Gemma-4-E4B-it model entry (with vision mmproj) as a strong mini-class default. - nostr.rs (host + client): use Gemma-4-E4B-it as the 8-24GB auto-pack tier default instead of Qwen3-8B; update auto-pack tests accordingly. - mesh/mod.rs: advertised_context_length now falls back through public-model-id -> runtime-model-name resolution so context length is found when the routable model name differs from the served runtime name. - ui/tooltip.tsx: disable hover tooltips on touch/coarse-pointer devices via a matchMedia-backed useSyncExternalStore hook. Deliberately excludes the MoA gateway/context/fanout/reducer/rescue changes and openai-frontend transport/responses changes, which are the suspected cause of the mesh-route hang and need a real multi-node retest before re-merging.
Contributor
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (6)
📝 WalkthroughWalkthroughThe PR adds Gemma-4-E4B-it-Q4_K_M model support across mesh-client and mesh-llm-host-runtime, updates automatic model pack selection for 8-24GB VRAM systems to prefer this model, enhances context length resolution with descriptor-based fallback lookup, and implements touch device detection to disable tooltips on coarse-pointer interfaces. ChangesGemma-4 Model Integration
Touch Device Tooltip Handling
🎯 2 (Simple) | ⏱️ ~12 minutes 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Comment |
Contributor
There was a problem hiding this comment.
Pull request overview
Re-applies a small set of previously-reverted improvements: updates the recommended mid-tier model selection, fixes peer-advertised context length reporting when public vs runtime model names differ, and disables hover tooltips on touch/coarse-pointer devices to avoid “stuck” tooltips.
Changes:
- Add
Gemma-4-E4B-it-Q4_K_M(with mmproj sidecar) to both node + client bundled catalogs and make it the 8–24GB auto-pack default. - Fix
PeerInfo::advertised_context_length()to resolve context length for public/routable model IDs that alias a different runtime-served model name. - Disable Radix hover tooltips on touch/coarse-pointer devices using
useSyncExternalStore+matchMedia.
Reviewed changes
Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| crates/mesh-llm-ui/src/components/ui/tooltip.tsx | Disables hover tooltips on touch/coarse-pointer devices via media-query subscription. |
| crates/mesh-llm-node/src/catalog.json | Adds Gemma-4-E4B-it quant + mmproj to the node catalog. |
| crates/mesh-llm-host-runtime/src/network/nostr.rs | Switches 8–24GB auto-pack default to Gemma and updates corresponding unit tests. |
| crates/mesh-llm-host-runtime/src/mesh/mod.rs | Fixes context-length lookup to work when a public model name aliases a different runtime model name. |
| crates/mesh-client/src/network/nostr.rs | Switches 8–24GB auto-pack default to Gemma and updates corresponding unit tests. |
| crates/mesh-client/src/models/catalog.json | Adds Gemma-4-E4B-it quant + mmproj to the client catalog. |
Comment on lines
1295
to
1299
| Pack { | ||
| min_vram: 8.0, | ||
| models: vec![catalog_ref("Qwen3-8B-Q4_K_M")], | ||
| models: vec![catalog_ref("Gemma-4-E4B-it-Q4_K_M")], | ||
| }, | ||
| Pack { |
Comment on lines
1854
to
+1869
| pub fn advertised_context_length(&self, model: &str) -> Option<u32> { | ||
| self.advertised_context_length_for_runtime_model(model) | ||
| .or_else(|| { | ||
| self.served_model_descriptors | ||
| .iter() | ||
| .filter(|descriptor| { | ||
| let runtime_name = descriptor.identity.model_name.as_str(); | ||
| runtime_name != model | ||
| && self.public_model_id_for_routable_model(runtime_name) == model | ||
| }) | ||
| .find_map(|descriptor| { | ||
| self.advertised_context_length_for_runtime_model( | ||
| &descriptor.identity.model_name, | ||
| ) | ||
| }) | ||
| }) |
michaelneale
added a commit
that referenced
this pull request
Jun 11, 2026
* origin/main: Salvage safe bits from reverted MoA PR (#820) (#824) Revert "Stabilize mesh MoA context and tool loops (#820)" (#823) Stabilize mesh MoA context and tool loops (#820) chore: flip docs pages to canonical URLs (#822) chore: Console public domain (#821) Add meshllm.cloud website, catalog viewer, and onboarding docs (#806) # Conflicts: # docs/index.html
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.
Re-applies the isolated, low-risk improvements from #820 without the MoA fanout/context changes that hung the
meshroute (reverted in #823).What users get
Gemma-4-E4B-itadded to the catalog (with vision mmproj) and used as the 8–24GB auto-pack default instead of Qwen3-8B — a stronger mini-class default.What was deliberately left out
The MoA pieces suspected of causing the
mesh-route hang are not included:mesh-mixture-of-agents/src/*(context, fanout, worker, arbiter, reducer, session, normalize)network/openai/moa_gateway/*(context_budget, context_selection, progress)openai-frontend/responses.rs,network/openai/transport.rsmesh-llm-guardrails/src/rescue.rs(wired into the MoA tool-call path)These need a real multi-node retest (the hang only reproduced against the live relay-heavy mesh, not CI sim tests) before re-merging.
Validation
cargo check -p mesh-llm-host-runtime -p mesh-llm-client -p mesh-llm-node— cleancargo test --lib auto_pack(host + client) — greennpm run typecheck— cleanBackground
#820 was reverted in #823 after it hung the
mesh/MoA route in production (verified via Fly v85 works / v86 hangs); console is currently on the v85 rollback.Summary by CodeRabbit
New Features
Bug Fixes