14july

app.py

@@ -48,8 +48,9 @@ llm = HuggingFacePipeline(pipeline=pipeline(
     tokenizer=tokenizer,
     max_new_tokens=512,  # Adjust as needed for desired response length
     return_full_text=False, # Crucial for getting only the AI's response, esp when ans is small
-    temperature=0.3,      # Controls randomness (0.0 for deterministic, 1.0 for very creative)
+    temperature=0.2,      # Controls randomness (0.0 for deterministic, 1.0 for very creative)
     do_sample=True        # Enable sampling for more varied outputs
+    stop_sequence=["Human:", "AI:", "\nHuman:", "\nAI:"] #to prevent gen unwanted conversations
     ))
 
 template = """The following is a concise and direct conversation between a human and an AI.
@@ -113,7 +114,7 @@ async def generate_text(request: QuestionRequest):
     # Return a StreamingResponse, which will send chunks as they are yielded by generate_stream()
     # media_type can be "text/event-stream" for SSE, or "text/plain" for simple newline-delimited text.
     # For simplicity, we'll start with "text/plain" for easier initial client parsing.
-    return StreamingResponse(generate_stream(), media_type="text/plain")
+    return StreamingResponse(generate_stream(), media_type="application/json")