import os
import requests
import huggingface_hub
import gradio as gr
webhook_url = os.environ.get("WEBHOOK_URL")
def submit_model(name):
try:
huggingface_hub.hf_hub_download(repo_id=name, filename="config.json") # sanity check input
except huggingface_hub.utils._errors.EntryNotFoundError:
return "# ERROR: Model does not have a config.json file!"
except huggingface_hub.utils._errors.RepositoryNotFoundError:
return "# ERROR: Model could not be found on the Hugging Face Hub!"
except requests.exceptions.HTTPError:
return "# ERROR: Network error while validating model. Please try again later."
except Exception as e:
print(e)
return "ERROR: Unexpected error. Please try again later."
try:
result = requests.post(webhook_url, json={"content":name})
except requests.exceptions.HTTPError:
return "# ERROR: Network error while contacting queue. Please try again in a few minutes."
except Exception as e:
print(e)
return "ERROR: Unexpected error. Please try again later."
return "# SUCCESS: Please wait up to 24 hours for your model to be added to the queue."
with gr.Blocks() as demo:
gr.HTML('
Subquadratic LLM Leaderboard
')
with gr.Tabs(elem_classes="tab-buttons") as tabs:
with gr.Tab("🏅 LLM Benchmark"):
gr.Markdown("Table filters under construction")
gr.Dataframe("data.csv")
with gr.Tab("📝 About"):
gr.Markdown("""
The **Subquadratic LLM Leaderboard** evaluates LLMs with subquadratic architectures (ie RWKV & Mamba) with the goal of providing open evaluation results while the architectures themselves are pending inclusion in 🤗 Transformers.
The metrics are the same as the Open LLM Leaderboard: ARC 25-shot, HellaSwag 10-shot, MMLU 5-shot, TruthfulQA zeroshot, Winogrande 5-shot, and GSM8K 5-shot.
This leaderboard is maintained by Devin Gulliver and is still under construction, check back regularly for further improvements!
""")
with gr.Tab("🚀 Submit here!"):
with gr.Group():
with gr.Row():
model_name = gr.Textbox(max_lines=1, label="Model Name", scale=4)
submit = gr.Button("Submit", variant="primary", scale=0)
output = gr.Markdown("Enter a public HF repo id, then hit Submit to add it to the evaluation queue.")
submit.click(fn=submit_model, inputs=model_name, outputs=output)
demo.launch()