added cache

app.py

@@ -6,6 +6,8 @@ llm = Llama.from_pretrained(
     repo_id="amir22010/fine_tuned_product_marketing_email_gemma_2_9b_q4_k_m",
     filename="unsloth.Q4_K_M.gguf",
     cache_dir=os.path.abspath(os.getcwd()),
+    n_ctx=2048, 
+    n_batch=126,
     verbose=False
 )
 
@@ -37,7 +39,7 @@ def greet(product,description):
     # response_format={
     #     "type": "json_object",
     # },
-    max_tokens=8192,
+    max_tokens=4096,
     temperature=0.7,
     stream=True
     )