Upload app.py with huggingface_hub

app.py

@@ -36,7 +36,9 @@ class WebAIConfig:
     storage_bucket: str = os.getenv("SUPABASE_STORAGE_BUCKET", "tender-documents")
     
     # WebAI Model Configuration
+    # Using 4b model weights with 9b processor (shared Qwen 3.5 architecture)
     model_name: str = "webAI-Official/webAI-ColVec1-4b"
+    processor_name: str = "webAI-Official/webAI-ColVec1-9b"  # Has working config
     use_8bit_quantization: bool = True
     use_4bit_quantization: bool = False
     use_flash_attention_2: bool = True
@@ -93,12 +95,19 @@ def setup_model():
                 bnb_4bit_use_double_quant=True,
             )
         
-        # Load processor first (lighter)
+        # Load processor from 9b repo (has working config) + model from 4b repo
+        log_event("model.setup.using_processor_workaround", {
+            "model": CONFIG.model_name,
+            "processor_source": CONFIG.processor_name
+        })
+        
         processor = AutoProcessor.from_pretrained(
-            CONFIG.model_name,
+            CONFIG.processor_name,  # Use 9b config
             trust_remote_code=True,
         )
         
+        # Note: processor is tied to model architecture, both use Qwen 3.5-VL base
+        
         # Configure model kwargs for FlashAttention-2
         model_kwargs = {
             "quantization_config": quantization_config,