Update main.py

main.py

@@ -20,7 +20,7 @@ PORT              = int(os.getenv("PORT", 7860))
 app = FastAPI(
     title       = "AI-Powered Web-App API",
     description = "Backend for summarisation, captioning & QA",
-    version     = "1.2.3",               # <-- bumped
+    version     = "1.2.4",               # <-- bumped
 )
 
 app.add_middleware(
@@ -47,15 +47,12 @@ summary_client        = InferenceClient(
     timeout = 120,
 )
 
-# ➜ Upgraded QA model (higher accuracy than roberta-base)
+# ➜ Upgraded QA model to a BigBird-RoBERTa checkpoint (better for documents)
 qa_client             = InferenceClient(
-    "deepset/roberta-large-squad2",
+    "google/bigbird-roberta-base-squad2",
     token   = HUGGINGFACE_TOKEN,
-    timeout = 120,
+    timeout = 180,
 )
-# If you need multilingual support, swap for:
-# qa_client = InferenceClient("deepset/xlm-roberta-large-squad2",
-#                             token=HUGGINGFACE_TOKEN, timeout=120)
 
 image_caption_client  = InferenceClient(
     "nlpconnect/vit-gpt2-image-captioning",
@@ -114,7 +111,6 @@ async def summarize_document(file: UploadFile = File(...)):
 # -------------------- Image Caption ------------------------------------------
 @app.post("/api/caption")
 async def caption_image(image: UploadFile = File(...)):
-    """`image` field name matches frontend (was `file` before)."""
     try:
         img_bytes = await image.read()
         img       = Image.open(io.BytesIO(img_bytes)).convert("RGB")
@@ -149,7 +145,7 @@ async def question_answering(file: UploadFile = File(...),
             context  = (res.get("generated_text") if isinstance(res, dict)
                         else str(res))
         else:
-            context = process_uploaded_file(file)[:3000]
+            context = process_uploaded_file(file)[:4096]   # BigBird handles longer
 
         if not context:
             return {"result": "No context – cannot answer."}