llm-checker 3.2.7 → 3.2.8

package/README.md

@@ -97,6 +97,12 @@ LLM Checker is published in all primary channels:
 - GitHub Releases: [Release history](https://github.com/Pavelevich/llm-checker/releases)
 - GitHub Packages: [`@pavelevich/llm-checker`](https://github.com/users/Pavelevich/packages/npm/package/llm-checker)
 
+### v3.2.8 Highlights
+
+- Fixed multimodal recommendation false positives from noisy metadata.
+- Coding-only models with incidental `input_types: image` flags are no longer treated as vision models.
+- Added regression tests to keep multimodal category picks aligned with true vision-capable models.
+
 ### v3.2.7 Highlights
 
 - License updated to **NPDL-1.0**: paid redistribution and paid hosted/API delivery now require a separate commercial license.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "llm-checker",
-  "version": "3.2.7",
+  "version": "3.2.8",
   "description": "Intelligent CLI tool with AI-powered model selection that analyzes your hardware and recommends optimal LLM models for your system",
   "bin": {
     "llm-checker": "bin/cli.js",

package/src/models/deterministic-selector.js

@@ -1042,6 +1042,9 @@ class DeterministicModelSelector {
 
     inferModalities(model, variantTag = '') {
         const inputTypes = Array.isArray(model.input_types) ? model.input_types.map((x) => String(x).toLowerCase()) : [];
+        const primaryCategory = String(model.primary_category || '').toLowerCase();
+        const categories = Array.isArray(model.categories) ? model.categories.map((x) => String(x).toLowerCase()) : [];
+        const useCases = Array.isArray(model.use_cases) ? model.use_cases.map((x) => String(x).toLowerCase()) : [];
         const text = [
             model.model_identifier,
             model.model_name,
@@ -1050,9 +1053,19 @@ class DeterministicModelSelector {
             variantTag
         ].filter(Boolean).join(' ').toLowerCase();
 
-        const hasVision = inputTypes.includes('image') ||
-            inputTypes.includes('vision') ||
-            /vision|vl\b|llava|pixtral|moondream|image|multimodal/.test(text);
+        const hasVisionInputFlag = inputTypes.includes('image') || inputTypes.includes('vision');
+        const hasVisionMetadataHint =
+            primaryCategory === 'multimodal' ||
+            categories.some((cat) => cat.includes('multimodal') || cat.includes('vision')) ||
+            useCases.some((useCase) => useCase.includes('multimodal') || useCase.includes('vision'));
+        const hasVisionTextHint =
+            /(?:\bmultimodal\b|\bvision\b|\bllava\b|\bbakllava\b|\bmoondream\b|\bpixtral\b|\bidefics\b|\bpaligemma\b|\bminicpm-?v\b|\bqwen[\w.-]*vl\b|\bllama3\.2[-_ ]?vision\b|\bdeepseek-ocr\b)/.test(text);
+        const hasVisionContextHint =
+            /\b(image[- ]?(understanding|caption|analysis)|vision[- ]?language|vlm)\b/.test(text);
+
+        // Some upstream scrapers may over-report `image` support by scanning generic page text.
+        // Trust image input flags only when accompanied by multimodal metadata or explicit vision naming.
+        const hasVision = hasVisionTextHint || hasVisionMetadataHint || (hasVisionInputFlag && hasVisionContextHint);
 
         return hasVision ? ['text', 'vision'] : ['text'];
     }