changed the model

app.py

@@ -1,30 +1,35 @@
 import logging
+import io
 from fastapi import FastAPI, UploadFile, File, Form
-from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
-from qwen_vl_utils import process_vision_info   # pip install qwen-vl-utils
 from PIL import Image
-import io
+from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch
 
 logging.basicConfig(level=logging.INFO, format="%(asctime)s | %(levelname)s | %(message)s")
 logger = logging.getLogger(__name__)
 
-app = FastAPI(title="Qwen2.5-VL OCR API")
+app = FastAPI(title="Moondream2 OCR API")
 
-# ✅ Use a vision-language model, NOT Qwen3.5-0.8B (text-only)
-model_id = "Qwen/Qwen2.5-VL-3B-Instruct"
+model_id = "vikhyatk/moondream2"
+revision  = "2025-01-09"   # pin to a stable release
 
 logger.info("=== LOADING MODEL ===")
-model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
+model = AutoModelForCausalLM.from_pretrained(
     model_id,
-    torch_dtype=torch.float32,   # float16 if GPU
-    device_map="cpu"
+    revision=revision,
+    trust_remote_code=True,
+    torch_dtype=torch.float32,
 )
-processor = AutoProcessor.from_pretrained(model_id)
+tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision)
 model.eval()
 logger.info("✅ Model loaded!")
 
 
+@app.get("/")
+async def root():
+    return {"message": "Moondream2 OCR API ready! POST /ocr"}
+
+
 @app.get("/health")
 async def health():
     return {"status": "healthy", "model": model_id}
@@ -33,34 +38,16 @@ async def health():
 @app.post("/ocr")
 async def ocr(
     image: UploadFile = File(...),
-    prompt: str = Form("Extract all text from this image accurately")
+    prompt: str = Form("Extract all text from this image accurately.")
 ):
     logger.info(f"📸 OCR request | file: {image.filename}")
     try:
         img = Image.open(io.BytesIO(await image.read())).convert("RGB")
 
-        messages = [{"role": "user", "content": [
-            {"type": "image", "image": img},
-            {"type": "text",  "text": prompt}
-        ]}]
-
-        # ✅ Correct Qwen2.5-VL pipeline
-        text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
-        image_inputs, video_inputs = process_vision_info(messages)
-
-        inputs = processor(
-            text=[text],
-            images=image_inputs,
-            videos=video_inputs,
-            return_tensors="pt"
-        ).to("cpu")
+        enc  = model.encode_image(img)
 
         with torch.inference_mode():
-            outputs = model.generate(**inputs, max_new_tokens=512, do_sample=False)
-
-        # ✅ Decode only the newly generated tokens
-        input_len = inputs["input_ids"].shape[1]
-        response = processor.decode(outputs[0][input_len:], skip_special_tokens=True).strip()
+            response = model.answer_question(enc, prompt, tokenizer)
 
         logger.info("✅ OCR done")
         return {"extracted_text": response}
@@ -70,6 +57,6 @@ async def ocr(
         return {"error": str(e)}
 
 
-@app.get("/")
-async def root():
-    return {"message": "Qwen2.5-VL OCR API ready! Try POST /ocr"}
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run("app:app", host="0.0.0.0", port=7860, reload=False)

requirements.txt

@@ -2,6 +2,6 @@ fastapi
 uvicorn
 transformers
 pillow
-accelerate
 python-multipart
-qwen-vl-utils
+einops          # required by moondream2
+torch --index-url https://download.pytorch.org/whl/cpu