Refactor Dockerfile and encoder.py to improve environment variable handling and model initialization

Dockerfile

@@ -3,7 +3,6 @@ FROM python:3.9
 WORKDIR /app
 
 ENV HF_HOME=/app/hf_cache
-ENV HF_TOKEN=${HF_TOKEN}
 RUN mkdir -p /app/hf_cache && chmod 777 /app/hf_cache
 
 COPY requirements.txt .

src/encoder.py

@@ -1,23 +1,38 @@
 from typing import List, Dict
 from PIL.Image import Image
+import os
 
 import torch
 from transformers import AutoModel, AutoProcessor
 
 
 MODEL_NAME = "Marqo/marqo-fashionCLIP"
+HF_TOKEN = os.environ.get("HF_TOKEN")
 
 
 class FashionCLIPEncoder:
     def __init__(self):
+        self.device = torch.device("cpu")
+        
         self.processor = AutoProcessor.from_pretrained(
-            MODEL_NAME, trust_remote_code=True
+            MODEL_NAME, 
+            trust_remote_code=True,
+            token=HF_TOKEN
         )
-        self.model = AutoModel.from_pretrained(MODEL_NAME, trust_remote_code=True)
-
-        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-        self.model = self.model.to(self.device)
-        self.model.eval()
+        
+        try:
+            self.model = AutoModel.from_pretrained(
+                MODEL_NAME,
+                trust_remote_code=True,
+                device_map=None,
+                token=HF_TOKEN
+            )
+            self.model = self.model.to(self.device)
+            self.model.eval()
+            
+        except Exception as e:
+            print(f"Error initializing model: {str(e)}")
+            raise
 
     def encode_text(self, texts: List[str]) -> List[List[float]]:
         kwargs = {
@@ -45,4 +60,4 @@ class FashionCLIPEncoder:
         return self.model.get_text_features(**batch).detach().cpu().numpy().tolist()
 
     def _encode_images(self, batch: Dict) -> List[List[float]]:
-        return self.model.get_image_features(**batch).detach().cpu().numpy().tolist()
+        return self.model.get_image_features(**batch).detach().cpu().numpy().tolist()