Update

README.md

@@ -16,6 +16,7 @@ This is a space to try out the
 attempts to infer a function name, comment/description, and suitable variable
 names, when given the output of Hex-Rays decompiler output of a function.  More information is available in this [blog post](https://www.atredis.com/blog/2024/6/3/how-to-train-your-large-language-model).
 
-## TODO
+## TODO / Issues
 
 * We currently use `transformers` which de-quantizes the gguf.  This is easy but inefficient.  Can we use llama.cpp or Ollama with zerogpu?
+* Model returns the markdown json prefix often.  Is this something I am doing wrong?

app.py

@@ -72,7 +72,7 @@ def predict(code):
 
     output = pipe_out[0]["generated_text"]
 
-    json_output = json.dumps("Failed")
+    json_output = json.dumps([])
     try:
         json.loads(output)
         json_output = output
@@ -87,7 +87,7 @@ def predict(code):
 demo = gr.Interface(
     fn=predict,
     inputs=gr.Text(label="Hex-Rays decompiler output"),
-    outputs=[gr.Text(label="Aidapal Output as JSON"), gr.Text(label="Raw Aidapal Output")],
+    outputs=[gr.JSON(label="Aidapal Output as JSON"), gr.Text(label="Raw Aidapal Output")],
     description=frontmatter.load("README.md").content,
     examples=examples
 )