feat(extract): recommend Qwen3-8B as the local extraction model

A/B against the prior default (qwen2.5:7b) on CPU showed Qwen3-8B is both faster and more accurate on tricky/multilingual booking docs (correct Airbnb year+price, correct DisneySea admission date), once thinking is disabled — which the router now does. Feature it as the recommended pull, keep qwen2.5:7b as the fallback.
This commit is contained in:
Maurice
2026-06-26 14:59:38 +02:00
parent b86bdce490
commit 51e8524d5c
+5 -5
View File
@@ -318,12 +318,12 @@ export default function AddonManager({ bagTrackingEnabled, onToggleBagTracking,
const MASKED = '••••••••'
const DEFAULT_OLLAMA_URL = 'http://localhost:11434/v1'
/** Curated models the local extractor is tuned for, pullable via Ollama. The router
* uses the strong model for flights/multi-item docs and the small one (when installed)
* for simple single-item bookings — so a host only needs these two. */
/** Curated models the local extractor is tuned for, pullable via Ollama. The router drives
* one model per document via Ollama's grammar-constrained `format`; "thinking" is disabled
* automatically, so the Qwen3 family works without any tuning. A host only needs one. */
const RECOMMENDED_MODELS: { id: string; label: string; note: string; recommended: boolean; vision: boolean }[] = [
{ id: 'qwen2.5:7b', label: 'Qwen2.57B', note: 'Recommended · reliable for flights & multi-item bookings · Apache-2.0', recommended: true, vision: false },
{ id: 'qwen2.5:3b', label: 'Qwen2.5 — 3B', note: 'Optional · used automatically for simple bookings (~3× faster) · Apache-2.0', recommended: false, vision: false },
{ id: 'qwen3:8b', label: 'Qwen38B', note: 'Recommended · best extraction quality & speed on CPU (thinking auto-disabled) · Apache-2.0', recommended: true, vision: false },
{ id: 'qwen2.5:7b', label: 'Qwen2.5 — 7B', note: 'Solid fallback · a bit less accurate on tricky/multilingual docs · Apache-2.0', recommended: false, vision: false },
]
/**