Update app.py

app.py

@@ -15,12 +15,10 @@ import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM, StoppingCriteria, StoppingCriteriaList
 
 # Configuration
-# Use this MODEL_ID, adjust if you have a local path instead
 MODEL_ID = os.getenv("GEMMA_MODEL_PATH", "tabularisai/german-gemma-3-1b-it")
-# Hugging Face token secret (optional, for gated/private models)
-HF_TOKEN = os.getenv("Tokentest")
+HF_TOKEN = os.getenv("Tokentest")  # Optional
 
-# Load tokenizer and model
+# Load tokenizer and model using external snippet
 print(f"Loading model {MODEL_ID}...")
 tokenizer = AutoTokenizer.from_pretrained(
     MODEL_ID,
@@ -35,7 +33,7 @@ model = AutoModelForCausalLM.from_pretrained(
     device_map="auto"
 ).eval()
 
-# Optional: set up a simple stopping criteria on <end_of_turn> token
+# Stopping criteria tokens
 PAD = tokenizer.pad_token_id or tokenizer.eos_token_id
 EOT = tokenizer.convert_tokens_to_ids('<end_of_turn>')
 
@@ -68,10 +66,9 @@ class PodcastGenerator:
 
         full_prompt = system_prompt + "\n\n" + user_prompt
 
-        # sync generation in executor
+        # sync generation in executor using model.generate
         def gen_sync():
             inputs = tokenizer(full_prompt, return_tensors="pt").to(model.device)
-            # add stopping criteria
             stop_crit = StoppingCriteriaList([StoppingCriteria(max_length=512)])
             outputs = model.generate(
                 **inputs,
@@ -139,4 +136,4 @@ def run_app():
     demo.launch()
 
 if __name__ == '__main__':
-    run_app()
+    run_app()