working E2E locally

app.py

@@ -5,9 +5,7 @@ import signal
 os.environ["GRADIO_ANALYTICS_ENABLED"] = "False"
 import gradio as gr
 
-from huggingface_hub import create_repo, HfApi
-from huggingface_hub import snapshot_download
-from huggingface_hub import whoami
+from huggingface_hub import HfApi
 from huggingface_hub import ModelCard
 
 from gradio_huggingfacehub_search import HuggingfaceHubSearch
@@ -18,22 +16,38 @@ HF_TOKEN = os.environ.get("HF_TOKEN")
 
 HF_PATH = "https://huggingface.co/"
 
-# change to include others
-CONV_TEMPLATE = "tinyllama_v1_0"
+def button_click(hf_model_id, conv_template, quantization):
+    api = HfApi()
+    model_dir_name = hf_model_id.split("/")[1]
+    mlc_model_name = model_dir_name + "-" + quantization + "-" + "MLC"
 
-def button_click(_, hf_model_id, quantization):
-    os.system("mkdir -p dist/models && cd dist/models")
+    os.system("mkdir -p dist/models")
     os.system("git lfs install")
-    os.system("git clone " + HF_PATH + hf_model_id)
-    os.system("cd ../..")
+
+    api.snapshot_download(repo_id=hf_model_id, local_dir=f"./dist/models/{model_dir_name}")
+
+    os.system("mlc_llm convert_weight ./dist/models/" + model_dir_name + "/" + \
+              " --quantization " + quantization + \
+              " -o dist/" + mlc_model_name)
     
+    os.system("mlc_llm gen_config ./dist/models/" + model_dir_name + "/" + \
+              " --quantization " + quantization + " --conv-template " + conv_template + \
+              " -o dist/" + mlc_model_name + "/")
+    
+    # push to HF
+    user_name = api.whoami()["name"]
+    api.create_repo(repo_id=f"{user_name}/{mlc_model_name}", private=True)
 
+    api.upload_large_folder(folder_path=f"./dist/{mlc_model_name}",
+                  repo_id=f"{user_name}/{mlc_model_name}",
+                  repo_type="model")
+    
     return "successful"
 
 demo = gr.Interface(
     fn=button_click,
-    inputs = [gr.Button("Click me"),
-              gr.Textbox(label="HF Model ID"), 
+    inputs = [gr.Textbox(label="HF Model ID"),
+              gr.Dropdown(["tinyllama_v1_0", "qwen2"], label="Conversation Template"),
               gr.Dropdown(["q4f16_1", "q4f32_1"], label="Quantization Method")],
     outputs = "text"
 )