D-7: reconcile Ollama model baseline — qwen2.5-coder:32b not installed; qwen3.6 + q8_0 in use #58

New Issue

2026-05-12T19:09:57Z

KofTwentyTwo commented

2026-05-12 19:09:57 +00:00

(Migrated from github.com)

Summary

Pull local Ollama models. Current state on user's machine (verified ollama list 2026-05-12):

nomic-embed-text:latest ✓ INSTALLED (274 MB, pulled 4h ago)
qwen2.5-coder:32b ✗ NOT installed; only qwen2.5-coder:32b-instruct-q8_0 (34 GB) is present
qwen3.6:latest ✓ installed (23 GB, 11 days old) — used as memory extractor per 8d0a959 commit

Evidence

docs/TODO.md:86 — "D-7 (deferred — needs user environment work): pull local Ollama models — ollama pull nomic-embed-text + ollama pull qwen2.5-coder:32b"
Live ollama list output (2026-05-12)

Recommended fix

User decision: is qwen2.5-coder:32b (vanilla) still the canonical baseline, or has the codebase moved to qwen3.6:latest / qwen2.5-coder:32b-instruct-q8_0? Per commit 8d0a959, qwen3.6 is the memory extractor and q8_0 is the agent-loop fallback. If so:

nomic-embed-text ✓ done
qwen3.6:latest ✓ done
qwen2.5-coder:32b may not be needed at all

Reconcile CLAUDE.md (line about qwen2.5-coder:32b baseline) against current model usage. Close this issue if the codebase no longer wants vanilla qwen2.5-coder:32b.

Source

docs/TODO.md:86
git log --oneline -1 8d0a959 — "feat(model): qwen3.6 for memory extractor; q8_0 for agent-loop fallback"

Cross-references

Related: #8 (M-4 Memory surface) — vector search activation path

## Summary Pull local Ollama models. Current state on user's machine (verified `ollama list` 2026-05-12): - `nomic-embed-text:latest` ✓ INSTALLED (274 MB, pulled 4h ago) - `qwen2.5-coder:32b` ✗ NOT installed; only `qwen2.5-coder:32b-instruct-q8_0` (34 GB) is present - `qwen3.6:latest` ✓ installed (23 GB, 11 days old) — used as memory extractor per `8d0a959` commit ## Evidence - `docs/TODO.md:86` — "D-7 (deferred — needs user environment work): pull local Ollama models — `ollama pull nomic-embed-text` + `ollama pull qwen2.5-coder:32b`" - Live `ollama list` output (2026-05-12) ## Recommended fix User decision: is `qwen2.5-coder:32b` (vanilla) still the canonical baseline, or has the codebase moved to `qwen3.6:latest` / `qwen2.5-coder:32b-instruct-q8_0`? Per commit `8d0a959`, qwen3.6 is the memory extractor and q8_0 is the agent-loop fallback. If so: - `nomic-embed-text` ✓ done - `qwen3.6:latest` ✓ done - `qwen2.5-coder:32b` may not be needed at all Reconcile `CLAUDE.md` (line about `qwen2.5-coder:32b` baseline) against current model usage. Close this issue if the codebase no longer wants vanilla `qwen2.5-coder:32b`. ## Source - `docs/TODO.md:86` - `git log --oneline -1 8d0a959` — "feat(model): qwen3.6 for memory extractor; q8_0 for agent-loop fallback" ## Cross-references - Related: #8 (M-4 Memory surface) — vector search activation path

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: KofTwentyTwo/Jarvis#58