magilogi commited on
Commit
9b4f201
β€’
1 Parent(s): 17a23a4
Files changed (1) hide show
  1. app.py +1 -1
app.py CHANGED
@@ -57,7 +57,7 @@ with gr.Blocks(css="custom.css") as demo:
57
  """<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>"""
58
  )
59
  with gr.Column(scale=1):
60
- gr.Image(value="workflow.svg", style={"width": "33%"})
61
 
62
  with gr.Tabs(elem_classes="tab-buttons"):
63
  with gr.TabItem("πŸ” Evaluation table"):
 
57
  """<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>"""
58
  )
59
  with gr.Column(scale=1):
60
+ gr.Image(value="workflow.svg").style(width=256, height=256)
61
 
62
  with gr.Tabs(elem_classes="tab-buttons"):
63
  with gr.TabItem("πŸ” Evaluation table"):