Update with Leaderboard V2

app.py

@@ -627,6 +627,9 @@ COLUMNS = [
     "Parallel Functions Exec",
     "Parallel Multiple Exec",
     "Relevance Detection",
+    "Cost ($ Per 1k Function Calls)",
+    "Latency Mean (s)",
+    "Latency Standard Deviation (s)",
 ]
 
 def parse_csv(text):
@@ -639,6 +642,12 @@ def parse_csv(text):
         row.pop(3)
         row.pop(5)
         row.pop(5)
+        row.pop(6)
+        row.pop(6)
+        row.pop(6)
+        row.pop(10)
+        row.pop(10)
+        
         result.append(row)
     return result
 
@@ -646,7 +655,7 @@ def parse_value(value):
     if value.endswith('%'):
         return float(value[:-1])
     try:
-        return int(value)
+        return float(value)
     except:
         return value
 

data.csv

@@ -1,23 +1,27 @@
-Rank,Overall Acc,Model,Model Link,Organization,License,AST Summary,Exec Summary,Simple Function AST,Multiple Functions AST,Parallel Functions AST,Parallel Multiple AST,Simple Function Exec,Multiple Functions Exec,Parallel Functions Exec,Parallel Multiple Exec,Relevance Detection
-1,84.28%,GPT-4-1106-Preview (FC),https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo,OpenAI,Proprietary,86.06%,65.53%,80.73%,88.50%,90.50%,84.50%,74.12%,70.00%,68.00%,50.00%,88.75%
-2,84.16%,GPT-4-0125-Preview (FC),https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo,OpenAI,Proprietary,85.61%,67.24%,81.45%,89.00%,88.50%,83.50%,72.94%,78.00%,68.00%,50.00%,87.50%
-3,84.16%,Gorilla-OpenFunctions-v2 (FC),https://gorilla.cs.berkeley.edu/blogs/7_open_functions_v2.html,Gorilla LLM,Apache 2.0,84.33%,72.72%,87.82%,89.00%,82.50%,78.00%,85.88%,82.00%,68.00%,55.00%,71.67%
-4,83.67%,Claude-3-Opus-20240229 (Prompt),https://www.anthropic.com/news/claude-3-family,Anthropic,Proprietary,79.82%,73.73%,85.27%,83.00%,79.00%,72.00%,89.41%,80.00%,68.00%,57.50%,84.58%
-5,81.75%,Mistral-Medium-2312 (Prompt),https://docs.mistral.ai/guides/model-selection/,Mistral AI,Proprietary,78.67%,66.93%,80.18%,84.50%,76.50%,73.50%,84.71%,76.00%,62.00%,45.00%,90.00%
-6,80.30%,Claude-3-Sonnet-20240229 (Prompt),https://www.anthropic.com/news/claude-3-family,Anthropic,Proprietary,84.91%,76.15%,85.64%,87.50%,83.50%,83.00%,90.59%,82.00%,72.00%,60.00%,41.25%
-7,80.30%,GPT-3.5-Turbo-0125 (FC),https://platform.openai.com/docs/models/gpt-3-5-turbo,OpenAI,Proprietary,81.55%,69.43%,80.18%,84.50%,82.50%,79.00%,84.71%,80.00%,68.00%,45.00%,68.33%
-8,79.07%,Functionary-Small (FC),https://huggingface.co/meetkai/functionary-small-v2.2,MeetKai,N/A,82.31%,64.40%,75.75%,89.50%,82.50%,81.50%,64.12%,78.00%,68.00%,47.50%,78.33%
-9,79.03%,Functionary-Medium-v2.2 (FC),https://huggingface.co/meetkai/functionary-medium-v2.2,MeetKai,N/A,82.25%,61.97%,76.00%,90.00%,85.00%,77.99%,65.88%,62.00%,70.00%,50.00%,79.17%
-10,77.41%,Claude-2.1 (Prompt),https://www.anthropic.com/news/claude-2-1,Anthropic,Proprietary,76.53%,53.93%,85.64%,83.00%,77.00%,60.50%,68.23%,48.00%,52.00%,47.50%,78.33%
-11,61.75%,Mistral-tiny-2312 (Prompt),https://docs.mistral.ai/guides/model-selection/,Mistral AI,Proprietary,55.28%,53.42%,59.64%,62.50%,56.00%,43.00%,71.17%,74.00%,36.00%,32.50%,77.08%
-12,61.02%,Claude-instant-1.2 (Prompt),https://www.anthropic.com/news/releasing-claude-instant-1-2,Anthropic,Proprietary,57.06%,49.88%,68.73%,59.00%,56.50%,44.00%,60.00%,52.00%,50.00%,37.50%,61.67%
-13,56.87%,Mistral-small-2312 (Prompt),https://docs.mistral.ai/guides/model-selection/,Mistral AI,Proprietary,57.01%,36.18%,46.55%,68.00%,50.50%,63.00%,34.71%,32.00%,38.00%,40.00%,89.58%
-14,56.81%,Mistral-large-2402 (FC),https://docs.mistral.ai/guides/model-selection/,Mistral AI,Proprietary,40.58%,38.49%,71.82%,90.50%,0.00%,0.00%,72.94%,76.00%,0.00%,5.00%,84.58%
-15,55.90%,Nexusflow-Raven-v2 (FC),https://huggingface.co/Nexusflow/NexusRaven-V2-13B,Nexusflow,Apache 2.0,58.01%,63.67%,76.55%,83.50%,39.50%,32.50%,61.18%,84.00%,62.00%,47.50%,0.00%
-16,55.87%,FireFunction-v1 (FC),https://huggingface.co/fireworks-ai/firefunction-v1,Fireworks,Apache 2.0,40.05%,37.31%,73.19%,87.00%,0.00%,0.00%,68.23%,76.00%,0.00%,5.00%,81.25%
-17,55.68%,Gemini-1.0-Pro (FC),https://deepmind.google/technologies/gemini/#introduction,Google,Proprietary,42.18%,29.30%,79.71%,89.00%,0.00%,0.00%,51.19%,66.00%,0.00%,0.00%,78.30%
-18,54.52%,GPT-4-0613 (FC),https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo,OpenAI,Proprietary,40.14%,27.12%,74.55%,86.00%,0.00%,0.00%,50.00%,56.00%,0.00%,2.50%,87.08%
-19,45.96%,Deepseek-v1.5 (Prompt),https://huggingface.co/deepseek-ai/deepseek-coder-7b-instruct-v1.5,Deepseek,Deepseek License,48.59%,8.55%,48.36%,61.00%,37.50%,47.50%,24.70%,2.00%,0.00%,7.50%,66.25%
-20,44.40%,Gemma-7b-it (Prompt),https://blog.google/technology/developers/gemma-open-models/,Google,gemma-terms-of-use,48.61%,40.43%,61.45%,60.00%,41.00%,32.00%,44.71%,48.00%,44.00%,25.00%,0.42%
-21,33.37%,Gorilla-OpenFunctions-v0 (FC),https://gorilla.cs.berkeley.edu/blogs/4_open_functions.html,Gorilla LLM,Apache 2.0,29.88%,24.06%,60.00%,56.00%,0.00%,3.50%,38.24%,58.00%,0.00%,0.00%,4.58%
-22,24.58%,Glaive-v1 (FC),https://huggingface.co/glaiveai/glaive-function-calling-v1,Glaive,cc-by-sa-4.0,15.14%,14.92%,34.55%,26.00%,0.00%,0.00%,21.18%,36.00%,0.00%,2.50%,46.25%
+Rank,Overall Acc,Model,Model Link,Organization,License,AST Summary,Exec Summary,Simple Function AST,Python Simple Function AST,Java Simple Function AST,JavaScript Simple Function AST,Multiple Functions AST,Parallel Functions AST,Parallel Multiple AST,Simple Function Exec,Python Simple Function Exec,REST Simple Function Exec,Multiple Functions Exec,Parallel Functions Exec,Parallel Multiple Exec,Relevance Detection,Cost ($ Per 1k Function Calls),Latency Mean (s),Latency Standard Deviation (s)
+1,79.35%,Claude-3-Opus-20240229 (Prompt),https://www.anthropic.com/news/claude-3-family,Anthropic,Proprietary,79.02%,68.13%,83.09%,89.25%,64.00%,72.00%,90.50%,80.00%,62.50%,83.53%,84.00%,82.86%,74.00%,70.00%,45.00%,80.83%,10.81,5.67,1.43
+2,78.71%,GPT-4-0125-Preview (Prompt),https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo,OpenAI,Proprietary,81.16%,67.36%,83.64%,89.00%,65.00%,78.00%,90.00%,86.50%,64.50%,82.94%,83.00%,82.86%,76.00%,68.00%,42.50%,69.17%,5.22,2.10,1.4
+3,78.35%,Gorilla-OpenFunctions-v2 (FC),https://gorilla.cs.berkeley.edu/blogs/7_open_functions_v2.html,Gorilla LLM,Apache 2.0,81.94%,69.64%,83.27%,93.00%,53.00%,66.00%,93.00%,85.50%,66.00%,87.06%,83.00%,92.86%,76.00%,68.00%,47.50%,60.83%,1.53,2.38,2.05
+4,78.18%,GPT-4-0125-Preview (FC),https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo,OpenAI,Proprietary,81.23%,63.88%,78.91%,88.75%,55.00%,48.00%,91.00%,88.50%,66.50%,70.00%,82.00%,52.86%,68.00%,70.00%,47.50%,81.67%,4.76,4.37,5.31
+5,77.71%,GPT-4-1106-Preview (FC),https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo,OpenAI,Proprietary,78.86%,67.10%,77.45%,87.25%,48.00%,58.00%,88.00%,88.00%,62.00%,79.41%,81.00%,77.14%,74.00%,70.00%,45.00%,80.83%,4.95,6.45,6.45
+6,73.82%,Claude-3-Sonnet-20240229 (Prompt),https://www.anthropic.com/news/claude-3-family,Anthropic,Proprietary,78.86%,66.07%,79.45%,88.25%,52.00%,64.00%,85.50%,85.00%,65.50%,75.29%,74.00%,77.14%,74.00%,70.00%,45.00%,53.33%,2.13,2.38,0.66
+7,72.24%,Mistral-Medium-2312 (Prompt),https://docs.mistral.ai/guides/model-selection/,Mistral AI,Proprietary,70.31%,56.79%,74.73%,82.75%,54.00%,52.00%,73.50%,79.50%,53.50%,67.65%,77.00%,54.29%,68.00%,64.00%,27.50%,88.33%,1.75,3.41,4.65
+8,69.18%,Claude-instant-1.2 (Prompt),https://www.anthropic.com/news/releasing-claude-instant-1-2,Anthropic,Proprietary,71.17%,61.37%,76.18%,83.75%,53.00%,62.00%,84.00%,78.50%,46.00%,76.47%,81.00%,70.00%,66.00%,58.00%,45.00%,54.17%,0.95,1.75,1.0
+9,68.24%,Claude-3-Opus-20240229 (FC),https://www.anthropic.com/news/claude-3-family,Anthropic,Proprietary,69.17%,61.85%,70.18%,81.25%,38.00%,46.00%,74.50%,75.00%,57.00%,75.88%,81.00%,68.57%,66.00%,68.00%,37.50%,62.50%,26.58,14.82,6.0
+10,64.06%,Claude-3-Sonnet-20240229 (FC),https://www.anthropic.com/news/claude-3-family,Anthropic,Proprietary,66.52%,59.26%,69.09%,78.50%,38.00%,56.00%,70.50%,72.50%,54.00%,63.53%,83.00%,35.71%,68.00%,68.00%,37.50%,51.67%,5.08,6.70,3.07
+11,63.65%,Claude-2.1 (Prompt),https://www.anthropic.com/news/claude-2-1,Anthropic,Proprietary,59.47%,45.68%,74.36%,80.00%,59.00%,60.00%,72.50%,55.00%,36.00%,48.24%,61.00%,30.00%,66.00%,36.00%,32.50%,83.33%,6.62,3.20,1.42
+12,62.53%,Mistral-large-2402 (FC Auto),https://docs.mistral.ai/guides/model-selection/,Mistral AI,Proprietary,57.60%,47.33%,64.91%,88.00%,3.00%,4.00%,88.50%,22.50%,54.50%,68.82%,75.00%,60.00%,72.00%,6.00%,42.50%,84.17%,4.74,2.91,2.73
+13,62.00%,Claude-3-Haiku-20240307 (FC),https://www.anthropic.com/news/claude-3-family,Anthropic,Proprietary,71.22%,57.65%,74.36%,84.25%,49.00%,46.00%,76.00%,76.00%,58.50%,50.59%,82.00%,5.71%,72.00%,68.00%,40.00%,21.67%,0.44,3.81,2.32
+14,60.65%,DBRX-Instruct (Prompt),https://www.databricks.com/blog/introducing-dbrx-new-state-art-open-llm,Databricks,Databricks Open Model,62.20%,59.25%,61.82%,72.75%,26.00%,46.00%,73.00%,70.00%,44.00%,60.00%,77.00%,35.71%,72.00%,60.00%,45.00%,54.58%,1.24,0.62,0.41
+15,57.53%,Mistral-large-2402 (FC Any),https://docs.mistral.ai/guides/model-selection/,Mistral AI,Proprietary,63.89%,50.25%,78.55%,86.75%,56.00%,58.00%,89.50%,32.50%,55.00%,80.00%,83.00%,75.71%,74.00%,2.00%,45.00%,0.00%,3.86,1.99,2.53
+16,56.65%,GPT-3.5-Turbo-0125 (FC),https://platform.openai.com/docs/models/gpt-3-5-turbo,OpenAI,Proprietary,66.31%,66.38%,56.73%,57.75%,51.00%,60.00%,65.50%,86.00%,57.00%,83.53%,78.00%,91.43%,70.00%,72.00%,40.00%,2.08%,0.42,1.53,1.25
+17,55.82%,Gemini-1.0-Pro (FC),https://deepmind.google/technologies/gemini/#introduction,Google,Proprietary,40.92%,38.68%,76.18%,87.50%,43.00%,52.00%,87.50%,0.00%,0.00%,78.24%,83.00%,71.43%,74.00%,0.00%,2.50%,76.67%,0.18,1.05,0.23
+18,54.82%,Mistral-small-2402 (FC Any),https://docs.mistral.ai/guides/model-selection/,Mistral AI,Proprietary,58.92%,46.38%,78.18%,86.00%,56.00%,60.00%,90.00%,36.50%,31.00%,80.00%,84.00%,74.29%,76.00%,12.00%,17.50%,0.00%,0.95,1.14,1.15
+19,50.18%,Nexusflow-Raven-v2 (FC),https://huggingface.co/Nexusflow/NexusRaven-V2-13B,Nexusflow,Apache 2.0,55.36%,56.95%,65.45%,70.50%,49.00%,58.00%,81.00%,35.00%,40.00%,55.29%,78.00%,22.86%,72.00%,58.00%,42.50%,2.08%,N/A,1.90,1.57
+20,49.88%,FireFunction-v1 (FC),https://huggingface.co/fireworks-ai/firefunction-v1,Fireworks,Apache 2.0,37.56%,32.19%,66.73%,85.75%,13.00%,22.00%,83.50%,0.00%,0.00%,61.76%,73.00%,45.71%,62.00%,0.00%,5.00%,73.33%,N/A,0.92,0.8
+21,47.94%,GPT-4-0613 (FC),https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo,OpenAI,Proprietary,36.57%,25.10%,59.27%,80.50%,3.00%,2.00%,87.00%,0.00%,0.00%,39.41%,55.00%,17.14%,56.00%,0.00%,5.00%,90.83%,10.27,3.64,3.99
+22,46.35%,Mistral-tiny-2312 (Prompt),https://docs.mistral.ai/guides/model-selection/,Mistral AI,Proprietary,41.98%,27.32%,48.91%,60.25%,28.00%,0.00%,48.50%,44.50%,26.00%,21.76%,36.00%,1.43%,28.00%,42.00%,17.50%,84.17%,0.13,1.14,1.19
+23,39.29%,Gemma-7b-it (Prompt),https://blog.google/technology/developers/gemma-open-models/,Google,gemma-terms-of-use,38.10%,27.93%,40.91%,46.25%,28.00%,24.00%,49.00%,29.50%,33.00%,24.71%,38.00%,5.71%,32.00%,30.00%,25.00%,57.08%,0.04,0.09,N/A
+24,37.12%,Deepseek-v1.5 (Prompt),https://huggingface.co/deepseek-ai/deepseek-coder-7b-instruct-v1.5,Deepseek,Deepseek License,33.51%,30.05%,36.55%,45.25%,7.00%,26.00%,41.00%,33.00%,23.50%,34.71%,39.00%,28.57%,38.00%,30.00%,17.50%,56.25%,0.45,1.20,N/A
+25,29.76%,Glaive-v1 (FC),https://huggingface.co/glaiveai/glaive-function-calling-v1,Glaive,cc-by-sa-4.0,28.84%,0.00%,52.36%,70.25%,3.00%,8.00%,63.00%,0.00%,0.00%,0.00%,0.00%,0.00%,0.00%,0.00%,0.00%,38.33%,0.02,0.06,N/A
+26,17.82%,Mistral-small-2402 (FC Auto),https://docs.mistral.ai/guides/model-selection/,Mistral AI,Proprietary,2.52%,7.65%,1.09%,1.50%,0.00%,0.00%,3.00%,4.00%,2.00%,20.59%,18.00%,24.29%,10.00%,0.00%,0.00%,99.58%,1.99,2.82,1.43