Update app.py

app.py

@@ -3,6 +3,11 @@ from fastapi import FastAPI
 from pydantic import BaseModel
 from transformers import AutoTokenizer, AutoModelForCausalLM, TextStreamer
 from peft import PeftModel
+import os
+
+os.environ["TRANSFORMERS_CACHE"] = "/tmp/hf"
+os.environ["HF_HOME"] = "/tmp/hf"
+os.environ["HUGGINGFACE_HUB_CACHE"] = "/tmp/hf"
 
 app = FastAPI()