Upd syntax

Dockerfile

@@ -17,6 +17,7 @@ RUN pip install --no-cache-dir -r requirements.txt
 # Set Hugging Face cache directory to persist model downloads
 ENV HF_HOME="/home/user/.cache/huggingface"
 ENV SENTENCE_TRANSFORMERS_HOME="/home/user/.cache/huggingface/sentence-transformers"
+ENV MEDGEMMA_HOME="/home/user/.cache/huggingface/sentence-transformers"
 
 # Create cache directories and ensure permissions
 RUN mkdir -p /app/model_cache /home/user/.cache/huggingface/sentence-transformers && \

vlm.py

@@ -2,8 +2,7 @@
 import os, logging, traceback, json, base64
 from io import BytesIO
 from PIL import Image
-# from huggingface_hub import InferenceClient # Render model on HF hub
-from transformers import pipeline             # Render model on transformers
+from huggingface_hub import InferenceClient # Render model on HF hub
 from translation import translate_query
 
 # Initialise once
@@ -19,7 +18,7 @@ def load_vlm():
     global vlm_pipe
     if vlm_pipe is None:
         logger.info("⏳ Loading MedGEMMA model via Transformers pipeline...")
-        vlm_pipe = pipeline("image-to-text", model="google/medgemma-4b", use_auth_token=HF_TOKEN, device_map="auto")
+        vlm_pipe = pipeline("image-to-text", model="google/medgemma-4b-it", use_auth_token=HF_TOKEN, device_map="auto")
         logger.info("✅ MedGEMMA model ready.")
     return vlm_pipe