simplify script

summarize_transcript.py

@@ -8,13 +8,13 @@ import argparse
 from llama_cpp import Llama
 from huggingface_hub import hf_hub_download
 
-def load_model():
+def load_model(repo_id, filename):
     """Load the model from Hugging Face Hub."""
 
     # Initialize the model with SYCL support
     llm = Llama.from_pretrained(
-        repo_id="unsloth/ERNIE-4.5-21B-A3B-PT-GGUF",
-        filename="*TQ1_0.gguf",
+        repo_id=repo_id,
+        filename=filename,
         n_gpu_layers=-1,  # Use all layers on GPU
         seed=1337,
         n_ctx=32768,       # Context size
@@ -84,14 +84,27 @@ def stream_summarize_transcript(llm, transcript):
 
 
 def main():
-    parser = argparse.ArgumentParser(description="Summarize transcript using ERNIE-4.5-21B-A3B-PT-GGUF model.")
-    parser.add_argument("-i", "--input", type=str, default="./transcripts/short.txt", help="Path to the input transcript file (default: ./transcripts/short.txt)")
+    parser = argparse.ArgumentParser(description="Summarize transcript in zh-TW using a GGUF model.")
+    parser.add_argument("-i", "--input", type=str, 
+                        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", 
+                        help="HuggingFace model in format repo_id:quant (e.g., Luigi/Falcon-H1-Tiny-Multilingual-100M-Instruct-GGUF:IQ4_NL)")
     args = parser.parse_args()
 
-    print("Loading ERNIE-4.5-21B-A3B-PT-GGUF model with SYCL acceleration...")
+    # Parse model argument if provided
+    if ":" in args.model:
+        repo_id, quant = args.model.rsplit(":", 1)
+        filename = f"*{quant}.gguf"
+    else:
+        print(f"Error: Invalid model format '{args.model}'. Expected format: repo_id:quant")
+        return
+
+    print(f"Loading model: {repo_id} ({filename}) with SYCL acceleration...")
 
     # Load the model
-    llm = load_model()
+    llm = load_model(repo_id, filename)
 
     # Read the transcript
     transcript_path = args.input