Fix: Remove Cached Models (#16)

* fix: Remove Cached Models

* refactor: default repo id, remove_dir function

core.py

@@ -1,5 +1,9 @@
 import os, shutil
 
+default_repo_id = "openai-community/gpt2"
+default_repo_id_parts = default_repo_id.split("/")
+default_model_folder = f"models--{'--'.join(default_repo_id_parts)}"
+
 def format_model_name(directory_name):
     parts = directory_name.split("--")
     return "/".join(parts[1:])
@@ -11,7 +15,11 @@ def list_download_models(cache_dir):
 
 def remove_dir(path):
     try:
-        shutil.rmtree(os.path.join(path, "/*"))
-        print(f"Directory '{path}' successfully removed.")
+        for model in os.listdir(path):
+            if model != default_model_folder:
+                model_path = os.path.join(path, model)
+                if os.path.isdir(model_path):
+                    shutil.rmtree(model_path)
+        print("successfully removed cached models!")
     except OSError as e:
         print(f"Error: {e.strerror}")

webui.py

@@ -7,10 +7,10 @@
 from langchain.llms.base import LLM
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer, AutoConfig
 
-from core import list_download_models, format_model_name, remove_dir
+from core import list_download_models, format_model_name, remove_dir, default_repo_id
 
 cache_dir = os.path.join(os.getcwd(), "models")
-saved_models = list_download_models(cache_dir)
+saved_models_list = list_download_models(cache_dir)
 
 #check if cuda is available
 device = 'cuda' if torch.cuda.is_available() else 'cpu'
@@ -59,7 +59,7 @@ def _llm_type(self) -> str:
     llm_chain = LLMChain(prompt=prompt, llm=llm)
     return llm_chain, llm
 
-model, tokenizer = initialize_model_and_tokenizer("openai-community/gpt2")
+model, tokenizer = initialize_model_and_tokenizer(default_repo_id)
 
 with gr.Blocks(fill_height=True) as demo:
     with gr.Row():
@@ -71,7 +71,7 @@ def _llm_type(self) -> str:
                 interactive=True)
             with gr.Group():
                 repo_id = gr.Textbox(
-                    value="openai-community/gpt2",
+                    value=default_repo_id,
                     label="Hugging Face Repo",
                     info="Default: openai-community/gpt2")
                 load_model_btn = gr.Button(
@@ -88,7 +88,7 @@ def _llm_type(self) -> str:
 
             with gr.Group():
                 saved_models = gr.Dropdown(
-                    choices=saved_models,
+                    choices=saved_models_list,
                     max_choices=5, 
                     filterable=True, 
                     label="Saved Models",
@@ -112,6 +112,7 @@ def user(user_message, history):
 
     def removeModelCache():
         remove_dir(cache_dir)
+        return gr.update(value=default_repo_id), gr.update(choices=[default_repo_id])
 
     def updateExecutionProvider(provider):
         if provider == "cuda":
@@ -147,7 +148,7 @@ def bot(history):
     load_model_btn.click(loadModel, repo_id, repo_id, queue=False, show_progress="full")
     execution_provider.change(fn=updateExecutionProvider, inputs=execution_provider, queue=False, show_progress="full")
     saved_models.change(loadModel, saved_models, repo_id, queue=False, show_progress="full")
-    offload_models.click(removeModelCache, None, saved_models, queue=False, show_progress="full")
+    offload_models.click(removeModelCache, None, [repo_id, saved_models], queue=False, show_progress="full")
 
 demo.queue()
 demo.launch(server_name="0.0.0.0")