Add Qwen2.5 1.5B as default extraction model

- Added qwen2.5_1.5b: 1.5B params, 128K context, non-reasoning
- Set as DEFAULT_EXTRACTION_MODEL (replaced granite_350m)
- Qwen2.5 has better instruction following for structured extraction
- Temperature 0.2 for focused but willing extraction

Granite models removed from default due to extraction failures:
- granite_350m: Only extracted from Window 1
- granite_4_0_1b_q4: Returned empty JSON for Windows 2-6

app.py

@@ -801,6 +801,22 @@ EXTRACTION_MODELS = {
             "repeat_penalty": 1.0,
         },
     },
+    "qwen2.5_1.5b": {
+        "name": "Qwen2.5 1.5B (128K Context)",
+        "repo_id": "Qwen/Qwen2.5-1.5B-Instruct-GGUF",
+        "filename": "*Q4_K_M.gguf",
+        "max_context": 131072,
+        "default_n_ctx": 4096,
+        "params_size": "1.5B",
+        "supports_reasoning": False,
+        "supports_toggle": False,
+        "inference_settings": {
+            "temperature": 0.2,
+            "top_p": 0.9,
+            "top_k": 30,
+            "repeat_penalty": 1.0,
+        },
+    },
     "granite_4_0_1b_q4": {
         "name": "Granite-4.0 1B (32K Context)",
         "repo_id": "unsloth/granite-4.0-h-1b-GGUF",
@@ -863,7 +879,7 @@ EXTRACTION_MODELS = {
     },
 }
 
-DEFAULT_EXTRACTION_MODEL = "granite_350m"
+DEFAULT_EXTRACTION_MODEL = "qwen2.5_1.5b"
 
 
 # ===== ADVANCED MODE: SYNTHESIS MODELS REGISTRY (16 models, 1B-30B) =====