style

app.py

@@ -97,7 +97,9 @@ with gr.Blocks() as demo:
     For more details on the TorchServe implementation and to reproduce, see [this reference code](https://github.com/fxmarty/bettertransformer_demo). For more details on BetterTransformer, check out the [blog post on PyTorch's Medium](https://medium.com/pytorch/bettertransformer-out-of-the-box-performance-for-huggingface-transformers-3fbe27d50ab2), and [the Optimum documentation](https://huggingface.co/docs/optimum/bettertransformer/overview)!"""
     )
 
-    gr.Markdown("## Single input scenario")
+    gr.Markdown("""## Single input scenario
+    Note: BetterTransformer normally shines with batch size > 1 and some padding. So this is not the best case here. Check out the heavy workload case below as well.
+    """)
 
     address_input_vanilla = gr.Textbox(
         max_lines=1, label="ip vanilla", value=ADDRESS_VANILLA, visible=False
@@ -149,7 +151,7 @@ with gr.Blocks() as demo:
     )
 
     input_n_spam_artif = gr.Number(
-        label="Number of inputs to send",
+        label="Number of sequences to send",
         value=80,
     )
     sequence_length = gr.Number(
@@ -157,7 +159,7 @@ with gr.Blocks() as demo:
         value=128,
     )
     padding_ratio = gr.Number(
-        label="Padding ratio",
+        label="Padding ratio (i.e. how much of the input is padding. In the real world when batch size > 1, the token sequence is padded with 0 to have all inputs with the same length.)",
         value=0.7,
     )
     btn_spam_artif = gr.Button("Spam text requests (using artificial data)")