phase7: verification table fully green — 6 models 36/36, catalog 30/30 · 5c5ed69 · zyquo-local

phase7: verification table fully green — 6 models 36/36, catalog 30/30

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

simon-pierre boucher committed 12 days ago (Jul 30, 2026) parent 705294b

Showing 2 changed files with +54 and −1

modified docs/PLAN.md +9 −1

		`@@ -119,5 +119,13 @@ full chat round-trip through the real UI — user bubble, thinking disclosure,`
119	119	“⚡ 107.0 tok/s · 467 tokens · 2.0s to first token” stats line, auto-title;
120	120	sidebar search filters live; Installed + Discover screens render per spec.
121	121	Zero warnings. Dark-mode + full design gate re-run before Definition of Done.
122		−## Phase 7 — Local model verification — pending
	122	+## Phase 7 — Local model verification — ✅ DONE (2026-07-30)
	123	+
	124	+- [x] `--verify` harness: download via the app's own DownloadManager, validate, load, deterministic gen, multi-turn carry-over, streaming cancellation, unload + memory-release check, tok/s + TTFT recorded
	125	+- [x] 6 models across architectures (qwen3/llama/smollm3/gemma3/qwen2-coder/R1-distill): 36/36 checks green
	126	+- [x] Entire Featured catalog (30 repos) dry-verified live: exists + files + sizes within 10 % — 30/30, zero corrections needed (MODELS.md ↔ ModelCatalog in sync)
	127	+- [x] Test models deleted, smallest kept for dev; results table in docs/VERIFICATION.md
	128	+
	129	+Checkpoint: VERIFY: ALL GREEN, exit 0. Best perf: Llama-3.2-1B at 222.8 tok/s;
	130	+unloads verified down to KBs of MLX active memory.
123	131	## Phase 8 — Signing & notarization — pending

added docs/VERIFICATION.md +45 −0

		`@@ -0,0 +1,45 @@`
	1	+<!--
	2	+ VERIFICATION.md
	3	+ Zyquo Local
	4	+
	5	+ Author: Simon-Pierre Boucher
	6	+ Mail: contact@spboucher.ai
	7	+-->
	8	+
	9	+# Phase 7 — Local Model Verification
	10	+
	11	+Run on 2026-07-30 with `ZyquoLocal --verify` on the build machine
	12	+(M5 Max, 48 GB, macOS 27.0 beta). Per model: download through the app's own
	13	+`DownloadManager`, directory validation, load, deterministic generation
	14	+(temperature 0, "Reply with exactly: OK" → replied "OK"), multi-turn context
	15	+carry-over ("vermilion" recall), streaming-cancellation test (stream stops in
	16	+≤0.4 s after cancel), unload with verified memory release (MLX active memory
	17	+drops to KBs).
	18	+
	19	+## End-to-end results — 6 models across architectures
	20	+
	21	+\| Model \| Download \| Load \| Generate \| Multi-turn \| Cancel \| Unload \| tok/s \| TTFT \|
	22	+\|---\|---\|---\|---\|---\|---\|---\|---\|---\|
	23	+\| Qwen3-0.6B-4bit \| ✅ \| ✅ \| ✅ \| ✅ \| ✅ \| ✅ \| 109.4 \| 0.08s \|
	24	+\| Llama-3.2-1B-Instruct-4bit \| ✅ \| ✅ \| ✅ \| ✅ \| ✅ \| ✅ \| 222.8 \| 0.74s \|
	25	+\| SmolLM3-3B-4bit \| ✅ \| ✅ \| ✅ \| ✅ \| ✅ \| ✅ \| 97.4 \| 0.11s \|
	26	+\| gemma-3-4b-it-qat-4bit \| ✅ \| ✅ \| ✅ \| ✅ \| ✅ \| ✅ \| 58.3 \| 1.08s \|
	27	+\| Qwen2.5-Coder-7B-Instruct-4bit \| ✅ \| ✅ \| ✅ \| ✅ \| ✅ \| ✅ \| 112.5 \| 0.42s \|
	28	+\| DeepSeek-R1-0528-Qwen3-8B-4bit \| ✅ \| ✅ \| ✅ \| ✅ \| ✅ \| ✅ \| 61.3 \| 0.08s \|
	29	+
	30	+Architecture coverage: qwen3, llama, smollm3, gemma3, qwen2 (coder), R1
	31	+distill (reasoning, `<think>` streaming). Memory release examples:
	32	+4.65 GB → 20 KB (R1-8B), 2.63 GB → 13 KB (gemma-3-4b).
	33	+
	34	+Per the protocol, test models except the smallest (Qwen3-0.6B-4bit, kept for
	35	+ongoing dev) were deleted afterward to reclaim disk.
	36	+
	37	+## Featured catalog dry-verification — 30/30 ✅
	38	+
	39	+Every repo in `docs/MODELS.md` / `ModelCatalog` verified against the live
	40	+Hub: repo exists, required files listed, total size within 10 % of the
	41	+catalog value (all matched to the hundredth of a GB; no gated repos, no
	42	+missing files, no corrections required — `MODELS.md` and `ModelCatalog`
	43	+remain in sync).
	44	+
	45	+VERDICT: ALL GREEN (`--verify` exit 0)
	46