Add application file

app.py

@@ -22,7 +22,7 @@ def compute_optimal_vocab(Nnv: float,
     # else:
     #     Vopt_app1,  Vopt_app2 = None, None
     #     Vopt_app3 = approach3_isoloss(Nnv, flops) 
-    Vopt_app1, Vopt_app2, Vopt_app3=1,2,3
+    Vopt_app1, Vopt_app2, Vopt_app3=1.,2.,3.
     
     results = f"## The optimal vocabulary size for non-vocabulary parameters {Nnv:1e} is:\nApproach 1: {Vopt_app1}\nApproach 2: {Vopt_app2}Approach 3: {Vopt_app3}"
     return results
@@ -31,8 +31,8 @@ def compute_optimal_vocab(Nnv: float,
 with gr.Blocks() as demo:
     with gr.Column():
         gr.Markdown(
-            """<img src="https://raw.githubusercontent.com/MrYxJ/calculate-flops.pytorch/main/screenshot/calflops_hf3.png?raw=true" style="float: left;" width="250" height="250"><h1> ⛽️Model(Transformers) FLOPs and Parameter Calculator</h1>
-            This tool is used to predict the optimal vocabulary size <h1> given the non-vocabulary parameters $N_{nv}$</h1>.
+            """<img src="figure_1_left_scaling_v5.png" style="float: left;" width="250" height="250"><h1>The Optimal Vocabulari Size Predictor</h1>
+            This tool is used to predict the optimal vocabulary size given the non-vocabulary parameters.
             We provide 3 ways for prediction:
 
             - Approach 1: Build the relationship between studied attributes and FLOPs: Build the relationship between the optimal data points (the points that reach the lowest loss under the same FLOPs budget) and the FLOPs.
@@ -48,7 +48,7 @@ with gr.Blocks() as demo:
         with gr.Row():
             Nnv = gr.Textbox(label="Non-vocabulary Parameters", value=7*10**9)
             flops = gr.Textbox(label="FLOPs", placeholder="Optional (e.g. 7.05*10**21)")
-            output_text = gr.Textbox()
+            output_text = gr.Textbox(label="Prediction")
         with gr.Row():
             btn = gr.Button("Compute the optimal vocabulary size")