Upload folder using huggingface_hub

Dockerfile

@@ -9,11 +9,7 @@ RUN pip install --no-cache-dir -r requirements.txt
 # Copy app code
 COPY . .
 
-# Pre-download model weights during build
-COPY app_preload.py /app/app_preload.py
-RUN python app_preload.py && rm app_preload.py
-
 # Remove build-time files
-RUN rm -f requirements.txt Dockerfile
+RUN rm -f requirements.txt Dockerfile app_preload.py
 
 CMD ["python", "app.py"]

app.py

@@ -7,14 +7,11 @@ from tinydoc_vlm import TinyDocVLMForConditionalGeneration, TinyDocVLMProcessor
 
 MODEL_ID = "eulogik/TinyDoc-VLM-256M"
 device = "cuda" if torch.cuda.is_available() else "cpu"
+dtype = torch.float16 if torch.cuda.is_available() else torch.float32
 
-print(f"Loading {MODEL_ID} on {device} with float16...")
-model = TinyDocVLMForConditionalGeneration.from_pretrained(
-    MODEL_ID,
-    torch_dtype=torch.float16,
-    low_cpu_mem_usage=True,
-)
-model.to(device).half().eval()
+print(f"Loading {MODEL_ID} on {device}...")
+model = TinyDocVLMForConditionalGeneration.from_pretrained(MODEL_ID)
+model.to(device).eval()
 processor = TinyDocVLMProcessor()
 print("Model loaded!")