Speed: cap saliency at 2 models, 60s LLM timeout

app.py

@@ -195,7 +195,7 @@ def _llm_report(p_mean: float, per_model: list, net_saliency: dict | None = None
         from openai import OpenAI
         if _VLLM_URL:
             # Live AMD MI300X inference via vLLM
-            client = OpenAI(base_url=_VLLM_URL, api_key="not-required")
+            client = OpenAI(base_url=_VLLM_URL, api_key="not-required", timeout=60.0)
             model_id = _LLM_MODEL
         else:
             # Fallback: HF Inference API
@@ -243,8 +243,8 @@ def get_models(atlas: str = "cc200"):
 # ── gradient saliency ──────────────────────────────────────────────────────
 
 def _compute_saliency(bw_t, adj_t, models):
-    # Cap at 5 models to keep CPU inference fast
-    sample = models[:5] if len(models) > 5 else models
+    # Cap at 2 models — backward pass on CPU is slow
+    sample = models[:2] if len(models) > 2 else models
     maps = []
     for _, task in sample:
         try: