use ernie 4.5 0.3b as default model

summarize_transcript.py

@@ -18,7 +18,7 @@ def load_model(repo_id, filename):
         n_gpu_layers=-1,  # Use all layers on GPU
         seed=1337,
         n_ctx=32768,       # Context size
-        verbose=False,     # Reduced verbosity for cleaner output
+        verbose=True,     # Reduced verbosity for cleaner output
         n_batch=512,
         n_ubatch=256,
         v_type=2,
@@ -57,11 +57,9 @@ def stream_summarize_transcript(llm, transcript):
     stream = llm.create_chat_completion(
         messages=messages,
         max_tokens=512,
-        temperature=0.2,
+        temperature=0.3,
         top_p=0.9,
-        repeat_penalty=1.3,
-        frequency_penalty=1.5,
-        presence_penalty=1.0,
+        repeat_penalty=1.5,
         stop=["<|end_of_text|>", "<|eot_id|>", "<|eom_id|>"],
         stream=True
     )
@@ -88,7 +86,7 @@ def main():
                         default="./transcripts/short.txt", 
                         help="Path to the input transcript file (default: ./transcripts/short.txt)")
     parser.add_argument("-m", "--model", type=str, 
-                        default="Luigi/Falcon-H1-Tiny-Multilingual-100M-Instruct-GGUF:IQ4_NL", 
+                        default="bartowski/baidu_ERNIE-4.5-0.3B-PT-GGUF:Q6_K", 
                         help="HuggingFace model in format repo_id:quant (e.g., Luigi/Falcon-H1-Tiny-Multilingual-100M-Instruct-GGUF:IQ4_NL)")
     args = parser.parse_args()