decrease batch number to save mem.

summarize_transcript.py

@@ -19,11 +19,10 @@ def load_model(repo_id, filename):
         seed=1337,
         n_ctx=32768,       # Context size
         verbose=False,     # Reduced verbosity for cleaner output
-        n_batch=2048,
-        n_ubatch=1024,
+        n_batch=512,
+        n_ubatch=256,
         v_type=2,
         k_type=2,
-        n_threads=6
     )
 
     return llm