Add input validations, proper defaults and errors for parameters

app.py

@@ -19,6 +19,12 @@ Options:
 
 API_URL = "https://hfbloom.ngrok.io/generate"
 HF_API_TOKEN = os.getenv("HF_API_TOKEN")
+MAX_TOKENS = 64
+MAX_LENGTH_ERROR = "The input max length is more than the demo supports. Please select a correct value from the slider."
+GENERATION_OPTIONS = ["Sample", "Greedy"]
+GENERATION_OPTIONS_ERROR = "Please select one option from either 'Sample' or 'Greedy'."
+SAMPLE_OPTIONS = ["Sample 1", "Sample 2", "Sample 3", "Sample 4", "Sample 5"]
+
 
 hf_writer = gr.HuggingFaceDatasetSaver(HF_API_TOKEN, "huggingface/bloom_internal_prompts", organization="huggingface")
 
@@ -47,6 +53,12 @@ def query(payload):
     return json.loads(response.content.decode("utf-8"))
     
 def inference(input_sentence, max_length, sample_or_greedy, seed=42):
+    if max_length > MAX_TOKENS:
+        return MAX_LENGTH_ERROR
+
+    if len(sample_or_greedy) == 0 or sample_or_greedy not in GENERATION_OPTIONS:
+        return GENERATION_OPTIONS_ERROR
+
     if sample_or_greedy == "Sample":
         parameters = {"max_new_tokens": max_length,
                       "top_p": 0.9,
@@ -77,9 +89,9 @@ gr.Interface(
     inference, 
     [
         gr.inputs.Textbox(label="Input"),
-        gr.inputs.Slider(1, 64, default=32, step=1, label="Tokens to generate"),
-        gr.inputs.Radio(["Sample", "Greedy"], label="Sample or greedy"),
-        gr.inputs.Radio(["Sample 1", "Sample 2", "Sample 3", "Sample 4", "Sample 5"], label="Sample other generations (only work in 'Sample' mode", type="index"),
+        gr.inputs.Slider(1, MAX_TOKENS, default=32, step=1, label="Tokens to generate"),
+        gr.inputs.Radio(GENERATION_OPTIONS, label="Sample or greedy", default="Sample"),
+        gr.inputs.Radio(SAMPLE_OPTIONS, label="Sample other generations (only work in 'Sample' mode", type="index", default="Sample 1"),
     ], 
     gr.outputs.Textbox(label="Output"),
     examples=examples,