Update app.py

app.py

@@ -664,7 +664,18 @@ def analyze_results(stats_df):
     ####
 
 def get_llm_suggested_settings(file, num_chunks=5):
-    chunks, _, _ = process_files(file.name if file else None, 'HuggingFace', 'paraphrase-miniLM', 'recursive', 500, 50)
+    if not file:
+        return {"error": "No file uploaded"}
+
+    chunks, _, _ = process_files(
+        file.name, 
+        'HuggingFace', 
+        'paraphrase-miniLM', 
+        'recursive', 
+        500, 
+        50,
+        custom_separators=None
+    )
     
     # Select a few random chunks
     sample_chunks = random.sample(chunks, min(num_chunks, len(chunks)))
@@ -699,9 +710,9 @@ Provide your suggestions in a Python dictionary format."""
     suggested_settings = llm(prompt)
 
     # Parse the generated text to extract the dictionary
-    # Note: This assumes the LLM generates a valid Python dictionary. In practice, you might need more robust parsing.
     try:
         settings_dict = eval(suggested_settings)
+        # Convert the settings to match the interface inputs
         return {
             "embedding_models": f"{settings_dict['embedding_model_type']}:{settings_dict['embedding_model_name']}",
             "split_strategy": settings_dict["split_strategy"],
@@ -718,12 +729,9 @@ Provide your suggestions in a Python dictionary format."""
     except:
         return {"error": "Failed to parse LLM suggestions"}
 
-    return settings_dict
-
-
 def update_inputs_with_llm_suggestions(suggestions):
-    if "error" in suggestions:
-        return [gr.update() for _ in range(11)]  # Return no updates if there's an error
+    if suggestions is None or "error" in suggestions:
+        return [gr.update() for _ in range(11)]  # Return no updates if there's an error or None
     
     return [
         gr.update(value=[suggestions["embedding_models"]]),  # embedding_models_input