add workflow

app.py

@@ -47,12 +47,15 @@ def create_scatter_plot(df, x_col, y_col, title, x_title, y_title):
 
 with gr.Blocks(css="custom.css") as demo:
     with gr.Row():
-        gr.Markdown(
-            """<div style="text-align: center;"><h1> <span style='color: #6aa84f;'>🐰 RABBITS:</span> <span style='color: #6aa84f;'>R</span>obust <span style='color: #6aa84f;'>A</span>ssessment of <span style='color: #6aa84f;'>B</span>iomedical <span style='color: #6aa84f;'>B</span>enchmarks <span style='color: #6aa84f;'>I</span>nvolving drug
+        with gr.Column(scale=1):
+            gr.Markdown(
+                """<div style="text-align: left;"><h1> <span style='color: #6aa84f;'>🐰 RABBITS:</span> <span style='color: #6aa84f;'>R</span>obust <span style='color: #6aa84f;'>A</span>ssessment of <span style='color: #6aa84f;'>B</span>iomedical <span style='color: #6aa84f;'>B</span>enchmarks <span style='color: #6aa84f;'>I</span>nvolving drug
 <span style='color: #6aa84f;'>T</span>erm <span style='color: #6aa84f;'>S</span>ubstitutions for Language Models <span style='color: #6aa84f;'></span></h1></div>\
-            <br>\
-            <p class='markdown-text'>Robust language models are crucial in the medical domain and the RABBITS project tests the robustness of LLMs by evaluating their handling of synonyms, specifically brand and generic drug names. We assessed 16 open-source language models from Hugging Face using systematic synonym substitution on MedQA and MedMCQA tasks. Our results show a consistent decline in performance across all model sizes, highlighting challenges in synonym comprehension. Additionally, we discovered significant dataset contamination by identifying overlaps between MedQA, MedMCQA test sets, and the Dolma 1.6 dataset using an 8-gram analysis. This highlights the need to improve model robustness and address contamination in open-source datasets</p>"""
-        )
+                <br>\
+                <p class='markdown-text'>Robust language models are crucial in the medical domain and the RABBITS project tests the robustness of LLMs by evaluating their handling of synonyms, specifically brand and generic drug names. We assessed 16 open-source language models from Hugging Face using systematic synonym substitution on MedQA and MedMCQA tasks. Our results show a consistent decline in performance across all model sizes, highlighting challenges in synonym comprehension. Additionally, we discovered significant dataset contamination by identifying overlaps between MedQA, MedMCQA test sets, and the Dolma 1.6 dataset using an 8-gram analysis. This highlights the need to improve model robustness and address contamination in open-source datasets</p>"""
+            )
+        with gr.Column(scale=1):
+            gr.Image(value="workflow.svg", tool=False)
 
     with gr.Tabs(elem_classes="tab-buttons"):
         with gr.TabItem("🔍 Evaluation table"):