update

.gitattributes

@@ -334,3 +334,70 @@ heatmaps/Yi-34B-Chat_CoT.jpg filter=lfs diff=lfs merge=lfs -text
 heatmaps/claude-3-opus-20240229_vision.jpg filter=lfs diff=lfs merge=lfs -text
 heatmaps/claude-3-sonnet-20240229_CoT.jpg filter=lfs diff=lfs merge=lfs -text
 heatmaps/deepseek-llm-67b-chat_Textonly.jpg filter=lfs diff=lfs merge=lfs -text
+heatmaps/CodeLlama-70b-Instruct-hf_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/Mixtral-8x7B-Instruct-v0.1_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/claude-3-haiku-20240307_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/Yi-34B-Chat_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/claude-3-sonnet-20240229_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/claude-3-sonnet-20240229_vision.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gemini-pro_vision-CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-35-turbo_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/Llama-3-70b-chat-hf_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/Mistral-7B-Instruct-v0.2_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/claude-3-haiku-20240307_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/deepseek-llm-67b-chat_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4-turbo-2024-04-09_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/Qwen1.5-72B-Chat_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/dbrx-instruct_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/claude-3-haiku-20240307_vision-CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gemini-pro-vision_vision.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-3.5-0613_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-3.5-turbo-0125_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-35-turbo_1shot.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4_vision-CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/claude-3-opus-20240229_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-35-turbo_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4-1106_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4-vision-preview_vision.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/claude-3-haiku-20240307_vision.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gemini-pro_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4-turbo-2024-04-09_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/Mixtral-8x7B-Instruct-v0.1_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gemma-7b-it_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4-1106_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/CodeLlama-70b-Instruct-hf_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/Mistral-7B-Instruct-v0.2_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/claude-3-sonnet-20240229_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/dbrx-instruct_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gemini-pro-vision_vision-CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gemma-7b-it_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/Qwen1.5-72B-Chat_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/claude-3-opus-20240229_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4-0125-preview_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/claude-3-opus-20240229_vision-CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/claude-3-opus-20240229_vision.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/claude-3-sonnet-20240229_vision-CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gemini-pro_vision.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-3.5-turbo-0125_1shot.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4-vision-preview_vision-CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/Llama-2-70b-chat-hf_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/Llama-2-70b-chat-hf_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/Yi-34B-Chat_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-3.5-turbo-0125_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/Llama-3-70b-chat-hf_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/deepseek-llm-67b-chat_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-3.5-0613_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4-0125-preview_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4_vision.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/claude-3-haiku-20240307_1shot.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4_CoT.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4_Textonly.png filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4_CoT.jpg filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4_vision.jpg filter=lfs diff=lfs merge=lfs -text
+heatmaps/gemini-pro-vision_vision.jpg filter=lfs diff=lfs merge=lfs -text
+heatmaps/gemini-pro-vision_vision-CoT.jpg filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-35-turbo_Textonly.jpg filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-35-turbo_1shot.jpg filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-35-turbo_CoT.jpg filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4_Textonly.jpg filter=lfs diff=lfs merge=lfs -text
+heatmaps/gpt-4_vision-CoT.jpg filter=lfs diff=lfs merge=lfs -text

app.py

@@ -34,6 +34,8 @@ def get_accuracy_dataframe(df_mother, category):
         columns={"is_answer_correct": "Overall Accuracy"}, inplace=True
     )
 
+    model_accuracy_df['model'] = model_accuracy_df['model'].apply(lambda x: x.split('/')[-1])
+
     # Ensure all expected difficulty levels are present
     expected_levels = [1, 2, 3, 4]  # Adjust based on your data
     for level in expected_levels: