Swap Dolphin 8B for Dolphin 7B (8B was too slow for free CPU tier, regularly hit the 90s self-heal timeout)

app.py

@@ -34,12 +34,17 @@ CURATED_MODELS = {
     "dolphin": {
         "id":          "dolphin",
         "role":        "uncensored",
-        "label":       "Dolphin 2.9.4 Llama 3.1 8B",
+        "label":       "Dolphin 2.9.3 Mistral 7B",
         "emoji":       "🐬",
         "category":    "Uncensored",
-        "repo":        "bartowski/dolphin-2.9.4-llama3.1-8b-GGUF",
+        # Was Dolphin 2.9.4 Llama 3.1 8B — measurably too slow on the free
+        # CPU tier (regularly blew past the server's own 90s self-heal
+        # timeout even for a 10-token reply, unlike the two 7B models here
+        # which consistently finish in under 35s). This 7B Mistral-based
+        # Dolphin variant matches the size class of the working models.
+        "repo":        "bartowski/dolphin-2.9.3-mistral-7B-32k-GGUF",
         "file":        "*Q4_K_M.gguf",
-        "ram_gb":      5.5,
+        "ram_gb":      4.8,
         "n_ctx":       8192,
         "description": "Eric Hartford's Dolphin series — gold standard uncensored model. Follows all instructions without refusal.",
     },