fix: added

Dockerfile

@@ -4,7 +4,6 @@ ENV PYTHONDONTWRITEBYTECODE=1 \
     PYTHONUNBUFFERED=1 \
     PIP_NO_CACHE_DIR=1 \
     HF_HOME=/data/.huggingface \
-    TRANSFORMERS_CACHE=/data/.huggingface/hub \
     PORT=7860
 
 WORKDIR /code

app/main.py

@@ -6,12 +6,17 @@ from contextlib import asynccontextmanager
 from fastapi import FastAPI, File, Form, HTTPException, UploadFile
 from fastapi.responses import JSONResponse
 from PIL import Image, UnidentifiedImageError
-from transformers import BlipForConditionalGeneration, BlipProcessor
+from transformers import AutoTokenizer, BlipForConditionalGeneration, BlipImageProcessor, BlipProcessor
 
 Image.MAX_IMAGE_PIXELS = None
 
 MODEL_ID = os.getenv("MODEL_ID", "Salesforce/blip-image-captioning-large")
-USE_FAST_PROCESSOR = True
+USE_FAST_PROCESSOR = os.getenv("USE_FAST_PROCESSOR", "true").strip().lower() in {
+    "1",
+    "true",
+    "yes",
+    "on",
+}
 MAX_IMAGE_SIZE = (1024, 1024)
 
 processor: BlipProcessor | None = None
@@ -21,7 +26,9 @@ model: BlipForConditionalGeneration | None = None
 def load_model() -> None:
     global processor, model
     if processor is None:
-        processor = BlipProcessor.from_pretrained(MODEL_ID, use_fast=USE_FAST_PROCESSOR)
+        image_processor = BlipImageProcessor.from_pretrained(MODEL_ID)
+        tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, use_fast=USE_FAST_PROCESSOR)
+        processor = BlipProcessor(image_processor=image_processor, tokenizer=tokenizer)
     if model is None:
         model = BlipForConditionalGeneration.from_pretrained(MODEL_ID)