Repository navigation
Plan 323 P2 — Qwen3.5 9B on this computer - #128
Merged
Merged
Conversation
`yarn models:verify-local qwen3.5:9b` against Ollama 0.40.5 (MLX, nvfp4): native single, parallel and streamed tool calls, vision, off/high reasoning, no `format`; 7.7 GB of weights and about 69 KB of cache per token at q8_0, so about 10 GB at the agents' 32K window — inside a 24 GB Mac's GPU where Gemma 26B spills. Assignable in the local edition like every local entry, recommended for no role until the usefulness tier exists. The probe scripts send through localRuntimeFetch, now exported: Gemma's re-verification took 8½ minutes, which Node's own fetch cut at five. Co-Authored-By: Claude Opus 5.5 <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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What it adds
ollama-qwen3.5-9b(aliasqwen-local), a second model on this computer. It comes fromyarn models:verify-local qwen3.5:9bagainst Ollama 0.40.5 (MLX runner, nvfp4):off/high;format: the MLX runner refuses it, so JSON is sent as a tool call, as for Gemma;q8_0. That's about 10 GB at the agents' 32K window, which fits a 24 GB Mac's 19.1 GB of GPU memory, where Gemma 26B spills.Like every local entry under D18, it is assignable in the local edition but recommended for no role. The usefulness tier, which isn't built yet, decides recommendations.
Also
models:verify-localandmodels:local-gatesnow send throughlocalRuntimeFetch, which@aflow/ai-clientnow exports. Gemma's re-verification took 8½ minutes, and Node's own fetch cuts a request off at 5 (the same limit Plan 323 — a long prompt on a local model no longer reads as an unreachable runtime #125 removed for the product path).OKagainst this machine's runtime.Testing
yarn models:verify-local qwen3.5:9breportsOK ollama-qwen3.5-9b;gemma4:26breportsOK ollama-gemma4-26b.yarn typecheckpasses;test:changedruns 170 tests, all passing.After merging
A tenant with its own model allowlist must add the model to it under Tenant Admin before a space can select it. The dev tenant's allowlist names Gemma explicitly.
🤖 Generated with Claude Code