llm-checker 3.5.15 → 3.7.0

package/README.md

@@ -573,6 +573,19 @@ This makes integrated GPUs visible even when the selected runtime backend is sti
 llm-checker recommend
 ```
 
+As of the scoring unification (#96), `check`, `recommend`, and `smart-recommend`
+all derive their ranking from **one canonical scoring core**
+(`DeterministicModelSelector` via `src/models/scoring-core.js`), so identical
+`(model, hardware)` inputs score identically across all three and the
+high-capacity right-sizing floor applies everywhere. They differ only in their
+model **source** and **presentation**, not in how a given model is ranked:
+
+| Command | Role | Ranking core |
+|---------|------|--------------|
+| `recommend` | Canonical model recommendations by category | Shared core (reference output) |
+| `check` | Full hardware-compatibility report with a recommendation card | Shared core (consistent ranking, fit-oriented report) |
+| `smart-recommend` | Catalog/DB-backed recommendations with a detailed score breakdown | Shared core (same ordering + scores) |
+
 Use optimization profiles to steer ranking by intent:
 
 ```bash
@@ -628,30 +641,36 @@ llm-checker search qwen --quant Q4_K_M --max-size 8
 
 ## Model Catalog
 
-LLM Checker ships with a pre-synced SQLite snapshot of the Ollama catalog. On first run, that snapshot is copied to `~/.llm-checker/models.db`, so recommendations and catalog search work immediately after npm install.
+LLM Checker ships with a pre-synced SQLite snapshot of the Ollama catalog plus a multi-source registry of exact downloadable/installable model artifacts. On first run, that snapshot is copied to `~/.llm-checker/models.db`, so recommendations and catalog search work immediately after npm install.
 
 The packaged snapshot currently includes:
 
 - 229 Ollama models
 - 7176 variants
+- 3259 multi-source registry repositories
+- 33729 exact model artifacts from Hugging Face, Ollama, and GPT4All
+- Hugging Face top 3000 repositories by downloads, fetched with API pagination
 - pull counts
 - tag counts
 - last-updated metadata
-- variant params, quantization, size, context, and input type fields when available
+- variant params, quantization, size, context, runtime, install commands, download URLs, license/gated flags, tasks, and modalities when available
 
 Refresh it any time:
 
 ```bash
 llm-checker sync
+llm-checker registry-sync --sources ollama,huggingface,gpt4all
+llm-checker registry-search qwen --runtime auto --max-size 8
+llm-checker registry-recommend --category coding --runtime auto --max-size 8
 ```
 
-For release maintainers, the packaged seed can be regenerated from the synced local DB:
+For release maintainers, the packaged seed can be regenerated from the synced local DB and registry APIs:
 
 ```bash
 npm run sync:seed
 ```
 
-`recommend`, `list-models`, `ai-run`, and `ai-check` prefer the synced SQLite catalog. If the SQLite catalog is unavailable, LLM Checker falls back to the scraped cache and then to the curated catalog.
+`recommend`, `list-models`, `ai-run`, and `ai-check` prefer the synced SQLite catalog. `registry-search` queries exact artifacts across sources, and `registry-recommend` ranks exact artifacts from the registry with the deterministic hardware-aware selector. If the SQLite catalog is unavailable, LLM Checker falls back to the scraped cache and then to the curated catalog.
 
 The curated fallback catalog includes 35+ models from the most popular Ollama families:
 
@@ -695,7 +714,7 @@ Three scoring systems are available, each optimized for different workflows:
 | `reasoning` | 60% | 10% | 20% | 10% |
 | `multimodal` | 50% | 15% | 20% | 15% |
 
-**Scoring Engine** (used by `smart-recommend` and `search`):
+**Scoring Engine** (used by `search` for catalog scoring; `smart-recommend`'s final ranking is produced by the shared scoring core — see #96):
 
 | Use Case | Quality | Speed | Fit | Context |
 |----------|:-------:|:-----:|:---:|:-------:|
@@ -823,7 +842,7 @@ LLM Checker uses a deterministic pipeline so the same inputs produce the same ra
 flowchart LR
   subgraph Inputs
     HW["Hardware detector<br/>CPU/GPU/RAM/backend"]
-    REG["Synced SQLite Ollama catalog<br/>(packaged seed + live sync)"]
+    REG["Synced SQLite model catalog<br/>(Ollama seed + multi-source registry)"]
     LOCAL["Installed local models"]
     FLAGS["CLI options<br/>use-case/runtime/limits/policy"]
   end
@@ -939,8 +958,9 @@ src/
     detector.js                # Hardware detection
     unified-detector.js        # Cross-platform detection
   data/
-    model-database.js          # SQLite storage and packaged seed loading
-    seed/models.db             # npm-packaged Ollama catalog snapshot
+    model-database.js          # SQLite storage, registry tables, and packaged seed loading
+    registry-ingestors.js      # Ollama/Hugging Face/GPT4All artifact normalization
+    seed/models.db             # npm-packaged Ollama + multi-source registry snapshot
     sync-manager.js            # Database sync from Ollama registry
 bin/
   enhanced_cli.js              # CLI entry point

package/analyzer/compatibility.js

@@ -428,6 +428,11 @@ class CompatibilityAnalyzer {
     }
 
     parseModelSize(sizeString) {
+        // Guard non-string input (undefined / a numeric size) — this runs for every
+        // model in calculateModelCompatibility, so one bad entry must not crash the
+        // whole analysis. Matches the guard in analyzer/performance.js.
+        if (typeof sizeString !== 'string' || !sizeString.trim()) return 1;
+
         const match = sizeString.match(/(\d+\.?\d*)[BM]/i);
         if (!match) return 1;
 

package/analyzer/performance.js

@@ -363,12 +363,13 @@ class PerformanceAnalyzer {
     }
 
     estimateLoadTime(model, hardware) {
+        // ~2 GB per 1B params (fp16-ish) on-disk approximation.
         const modelSizeGB = this.parseModelSize(model.size) * 2;
 
-        let loadTimeSeconds = modelSizeGB * 2;
-
-
-        loadTimeSeconds *= 0.7;
+        // Fold the previous `* 2` then `* 0.7` two-step (a leftover from an
+        // incomplete edit, with a dead blank line) into one documented factor:
+        // ~1.4 s of load time per GB before hardware adjustments.
+        let loadTimeSeconds = modelSizeGB * 1.4;
 
         const cpuSpeedFactor = Math.max(0.5, Math.min(1.5, (hardware.cpu.speed || 2.5) / 2.5));
         loadTimeSeconds /= cpuSpeedFactor;

package/bin/cli.js

@@ -11,44 +11,10 @@ if (!Number.isFinite(majorNodeVersion) || majorNodeVersion < 16) {
     process.exit(1);
 }
 
-function preprocessAiCheckModelsArg(argv) {
-    const normalizedArgs = [];
-    let modelsFilter = null;
-    let sawAiCheck = false;
-
-    for (let index = 0; index < argv.length; index += 1) {
-        const token = argv[index];
-
-        if (token === 'ai-check') {
-            sawAiCheck = true;
-        }
-
-        if (sawAiCheck && token === '--models') {
-            const nextToken = argv[index + 1];
-            if (nextToken && !nextToken.startsWith('-')) {
-                modelsFilter = nextToken;
-                index += 1;
-            }
-            continue;
-        }
-
-        if (sawAiCheck && token.startsWith('--models=')) {
-            modelsFilter = token.slice('--models='.length);
-            continue;
-        }
-
-        normalizedArgs.push(token);
-    }
-
-    return { args: normalizedArgs, modelsFilter };
-}
-
-const preprocessedArgs = preprocessAiCheckModelsArg(process.argv.slice(2));
-
-if (typeof preprocessedArgs.modelsFilter === 'string' && preprocessedArgs.modelsFilter.trim()) {
-    process.env.LLM_CHECKER_AI_CHECK_MODELS = preprocessedArgs.modelsFilter.trim();
-}
-
-process.argv = [process.argv[0], process.argv[1], ...preprocessedArgs.args];
+// `ai-check --models <list>` is now a real commander option handled in
+// enhanced_cli.js (and the AICheckSelector applies it as a candidate filter), so
+// the previous argv-rewriting shim — which stripped the flag and stashed it in an
+// env var that nothing read — is gone. LLM_CHECKER_AI_CHECK_MODELS still works as
+// an explicit fallback for the same filter.
 
 require('./enhanced_cli');