promtp17

app.py

@@ -61,11 +61,10 @@ compression_retriever = ContextualCompressionRetriever(
 # Configure FLAN-T5 (better task)
 # -----------------------------
 generator = pipeline(
-    repo_id="google/flan-t5-base",  # también puedes usar flan-t5-small
-    task="text-generation",
-    provider="auto",
-    temperature=0.1,
-    max_new_tokens=512,
+   "text-generation",       # 👈 Use this for T5 models
+   model="google/flan-t5-base",
+   max_new_tokens=128,           # shorter answers
+   temperature=0.2,
    repetition_penalty=1.1
 )
 llm = HuggingFacePipeline(pipeline=generator)