Update app.py

app.py

@@ -38,8 +38,8 @@ def get_system_tokens(model):
     return get_message_tokens(model, **system_message)
 
 
-repo_name = "IlyaGusev/saiga2_13b_ggml"
-model_name = "ggml-model-q4_1.bin"
+repo_name = "IlyaGusev/saiga2_13b_gguf"
+model_name = "ggml-model-q4_K.gguf"
 
 snapshot_download(repo_id=repo_name, local_dir=".", allow_patterns=model_name)
 
@@ -100,7 +100,7 @@ with gr.Blocks(
 ) as demo:
     favicon = '<img src="https://cdn.midjourney.com/b88e5beb-6324-4820-8504-a1a37a9ba36d/0_1.png" width="48px" style="display: inline">'
     gr.Markdown(
-        f"""<h1><center>{favicon}Saiga2 13B GGML Q4_1</center></h1>
+        f"""<h1><center>{favicon}Saiga2 13B GGUF Q4_K</center></h1>
 
         This is a demo of a **Russian**-speaking LLaMA2-based model. If you are interested in other languages, please check other models, such as [MPT-7B-Chat](https://huggingface.co/spaces/mosaicml/mpt-7b-chat).