|
|
|
import streamlit as st |
|
import glob |
|
import os |
|
from datetime import datetime |
|
|
|
|
|
st.title('Meta Open LLM leaderboard') |
|
|
|
directories = os.listdir("./data") |
|
|
|
|
|
|
|
def format_dir_date(data_dir): |
|
|
|
parsed_date = datetime.strptime(data_dir, "%Y%m%d_%H%M") |
|
|
|
|
|
return parsed_date.strftime("%b %d, %Y %H:%M") |
|
|
|
data_dir = st.selectbox( |
|
'Select different Date', |
|
directories, |
|
format_func=format_dir_date |
|
) |
|
|
|
captions_map = { |
|
"hg_average_to_agentbench_compare.png": "HF to AgentBench compare", |
|
"hg_average_to_opencompass_compare.png": "HF to OpenCompass</a> compare", |
|
"hg_average_to_mt_bench_compare.png": "HF to MT-Bench compare", |
|
"hg_average_to_mosaic_compare.png": "HF to MosaicML compare", |
|
"hg_average_to_alpacaeval_compare.png": "HF to AlpacaEval compare" |
|
} |
|
|
|
st.write("Generated on: <b>" + format_dir_date(data_dir) + "</b>", unsafe_allow_html=True) |
|
st.divider() |
|
|
|
data_path = './data/' + data_dir |
|
|
|
imgs = glob.glob(os.path.join(data_path, '*.png')) |
|
|
|
|
|
hf_llm_diagrams = [img for img in imgs if 'hf_llm_diagram' in os.path.basename(img)] |
|
|
|
|
|
remaining_imgs = [img for img in imgs if 'hf_llm_diagram' not in os.path.basename(img)] |
|
|
|
st.write("HuggingFace Open LLM leaderboard by Model Size") |
|
st.image(hf_llm_diagrams,use_column_width="auto") |
|
|
|
st.divider() |
|
st.write("HuggingFace and Other Leaderboards: A Comparative Model Evaluation") |
|
st.caption("Only models evaluated on both leaderboards are included.") |
|
cols = st.columns(2) |
|
|
|
for i, img in enumerate(remaining_imgs): |
|
|
|
filename = os.path.basename(img) |
|
|
|
|
|
caption = captions_map.get(filename, "") |
|
|
|
|
|
cols[i % 2].image(img, caption=caption, width=None) |
|
|
|
st.write( |
|
""" |
|
<ul><li> |
|
<a href="https://opencompass.org.cn/leaderboard-llm">OpenCompass</a> |
|
</li> |
|
<li><a href="https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboard">MT-Bench</a></li> |
|
|
|
</ul> |
|
""", unsafe_allow_html=True |
|
) |
|
st.divider() |
|
st.subheader('About') |
|
st.write('This meta leaderboard is built and maintained by Felix Zaslavskiy. For feedback, correction, suggestions please reach out on X at <a href="https://twitter.com/FZaslavskiy" >@FZaslavskiy</a> or here via community discussions.', unsafe_allow_html=True) |