Update app.py

app.py

@@ -30,7 +30,7 @@ def plot_throughput(bs=1):
             name=df.loc[i, 'Models'],
             hovertemplate =
             '<b>%{text}</b><br><br>' +
-            'throughput_column: %{x}<br>' +
+            f'{throughput_column}: %{{x}}<br>'+
             'Average Score: %{y}<br>' +
             'Peak Memory (MB): ' + str(df.loc[i, 'Peak Memory (MB)']) + '<br>' +
             'Human Eval (Python): ' + str(df.loc[i, 'humaneval-python']),
@@ -40,10 +40,10 @@ def plot_throughput(bs=1):
 
     fig.update_layout(
         autosize=False,
-        width=1000,
-        height=800,
+        width=700,
+        height=600,
         title=f'Average Score Vs Throughput (A100-80GB, Batch Size {bs}, Float16)',
-        xaxis_title='throughput_column',
+        xaxis_title=f'{throughput_column}',
         yaxis_title='Average Code Score',
     )
     return fig
@@ -53,7 +53,7 @@ demo = gr.Blocks()
 with demo:
     with gr.Row():
         gr.Markdown(
-            """<div style="text-align: center;"><h1> ⭐ Base <span style='color: #e6b800;'>Code</span> Models <span style='color: #e6b800;'>Evaluation</span></h1></div>\
+            """<div style="text-align: center;"><h1> ⭐ Multilingual <span style='color: #e6b800;'>Code</span> Models <span style='color: #e6b800;'>Evaluation</span></h1></div>\
             <br>\
             <p>We compare base multilingual code generation models on <a href="https://huggingface.co/datasets/openai_humaneval">HumanEval</a> benchmark and <a href="https://huggingface.co/datasets/nuprl/MultiPL-E">MultiPL-E</a>, in addition to throughput measurment\
             and information about the model. We only compare pre-trained models without instruction tuning.</p>"""
@@ -82,11 +82,11 @@ with demo:
         gr.Markdown(
             """Notes:
             <ul>
-            <li> Average score is the average pass@1 over all languages, for each model we exclude languages with a pass@1 score lower than 1 for the averaging.</li>
-            <li> Throughputs are measured using <a href="https://github.com/huggingface/optimum-benchmark/tree/main">Optimum-Benchmark</a> which powers <a href="https://huggingface.co/spaces/optimum/llm-perf-leaderboard">LLM Perf LeaderBoard</a>.</li>
+            <li> Throughputs and peak memory usage are measured using <a href="https://github.com/huggingface/optimum-benchmark/tree/main">Optimum-Benchmark</a> which powers <a href="https://huggingface.co/spaces/optimum/llm-perf-leaderboard">🤗 Open LLM-Perf Leaderboard 🏋️</a>. (0 throughput corresponds to OOM).</li>
+            <li> All models were evaluated with the <a href="https://github.com/bigcode-project/bigcode-evaluation-harness/tree/main">🔍 bigcode-evaluation-harness</a> with top-p=0.95, temperature=0.2 and n_samples=50.</li>
             <li> HumanEval-Python, reports the pass@1 on HumanEval, the rest is from MultiPL-E benchmark.</li>
-            <li> All models were evaluated with the <a href="https://github.com/bigcode-project/bigcode-evaluation-harness/tree/main">bigcode-evaluation-harness</a> with top-p=0.95, temperature=0.2 and n_samples=50.</li>
-            <li> #languages column represents the number of programming languages included during the pretraining.
+            <li> Average score is the average pass@1 over all languages. During the averaging, we exclude languages with a pass@1 score lower than 1 for each model.</li>
+            <li> #Languages column represents the number of programming languages included during the pretraining.
             </ul>"""
         )
 demo.launch()