Update app/sentiment.py

app/sentiment.py

@@ -1,47 +1,50 @@
 """
-Sentiment analysis module using Hugging Face Transformers with cache redirection for HF Spaces.
+Safe lazy-loading sentiment pipeline that works in Hugging Face Spaces (no /.cache error).
 """
 
 import os
-from transformers import pipeline
-from functools import lru_cache
 import hashlib
 import logging
+from functools import lru_cache
 
-# 🔧 Redirect HF cache to writable /tmp directory (important for Hugging Face Spaces)
+# Redirect the HF model cache to a writable directory
 os.environ["TRANSFORMERS_CACHE"] = "/tmp/huggingface"
 os.makedirs("/tmp/huggingface", exist_ok=True)
 
-# 🧠 Load sentiment model
-_sentiment = pipeline(
-    "sentiment-analysis",
-    model="distilbert-base-uncased-finetuned-sst-2-english"
-)
+from transformers import pipeline
 
 class SentimentCache:
-    """Handles in-memory caching and streaming of sentiment results."""
     latest_id: int = 0
     latest_result: dict = {}
+    _pipeline = None  # Lazy init
+
+    @classmethod
+    def _get_pipeline(cls):
+        if cls._pipeline is None:
+            logging.info("🔄 Loading sentiment model…")
+            cls._pipeline = pipeline(
+                "sentiment-analysis",
+                model="distilbert-base-uncased-finetuned-sst-2-english"
+            )
+        return cls._pipeline
 
     @classmethod
     def _hash(cls, text: str) -> str:
-        """Hash input text to use as a cache key."""
         return hashlib.sha256(text.encode()).hexdigest()
 
     @classmethod
     @lru_cache(maxsize=128)
     def _analyze(cls, text: str):
-        """Run inference on text, cached for performance."""
-        return _sentiment(text)[0]
+        pipe = cls._get_pipeline()
+        return pipe(text)[0]
 
     @classmethod
     def compute(cls, text: str):
-        """Trigger inference and update latest result."""
-        result = cls._analyze(text)
+        res = cls._analyze(text)
         cls.latest_id += 1
         cls.latest_result = {
             "text": text,
-            "label": result["label"],
-            "score": round(result["score"], 4)
+            "label": res["label"],
+            "score": round(res["score"], 4)
         }
         logging.info("✅ Sentiment computed: %s", cls.latest_result)