Update app.py

app.py

@@ -69,6 +69,7 @@ def initialize_model():
 
         tokenizer = AutoTokenizer.from_pretrained(
             MODEL_ID,
+            cache_dir="/data/models",
             trust_remote_code=True,
             use_fast=False
         )
@@ -81,6 +82,7 @@ def initialize_model():
         model = AutoModelForCausalLM.from_pretrained(
             MODEL_ID,
             torch_dtype=torch.float16,
+            cache_dir="/data/models",
             trust_remote_code=True,
             device_map={"": 0},  # <- force GPU:0
             low_cpu_mem_usage=True
@@ -314,7 +316,7 @@ class PodcastGenerator:
 
 Requirements:
 - Exactly two speakers: Speaker 1 and Speaker 2
-- The podcast should be long, focusing on the input text
+- The podcast should fill 4-5 minutes, focusing on the core context of the input text
 - DO NOT copy the example below , only use it as conversation reference 
 - The podcast should be professional, in-depth,  lively, witty and engaging, and hook the listener from the start.
 - The input text might be disorganized or unformatted. Ignore any formatting inconsistencies or irrelevant details; your task is to distill the essential points,