app.py

import json
import pandas as pd
import gradio as gr
from content import *
from css import *

NONE_COL = "Ranking"
    
AGENT_COLS = ["Method", "Model" , "SS Easy", "SS Medium", "SS Hard", "MS Easy", "MS Meduium", "MS Hard", "Overall", NONE_COL]
AGENT_TYPES = ["str", "str", "number", "number", "number", "number", "number", "number", "number", "number" , "number"]
model_name_adic = {
    "qwen-plus": "Qwen-Plus",
    "qwen2.5-72b-instruct": "Qwen2.5-72B",
    "qwen2.5-7b-instruct": "Qwen2.5-7B",
    "qwen2.5-14b-instruct": "Qwen2.5-14B",
    "qwen2.5-32b-instruct": "Qwen2.5-32B",
    "gpt-4o": "GPT-4o",
}
method_name_adic = {
    "reflexion": "Relfexion",
    "react": "React",
    "seeker": "WebWalker",
}

rag_name_adic = {
    "kimi": "Kimi",
    "mindsearch": "MindSearch",
    "navie": "Navie RAG",
    "o1": "o1",
    "tongyi": "Tongyi",
    "wenxin": "ERNIE",
    "gemini": "Gemini",
    "gemini_search": "Gemini w/ Search",
    "doubao": "Doubao",
}
agent_ranking = []
with open("agents_result.jsonl", "r") as f:
    for line in f:
        item = json.loads(line)
        agent_ranking.append([method_name_adic[item["method"]], model_name_adic[item["model"]], item["overall"]])
agent_ranking = sorted(agent_ranking, key=lambda x: x[2], reverse=False)
ranking_dict = {}
for i, (method, model, score) in enumerate(agent_ranking):
    ranking_dict[method+model] = i

agent_df = []
with open("agents_result.jsonl", "r") as f:
    for line in f:
        item = json.loads(line)
        agent_df.append([method_name_adic[item["method"]], model_name_adic[item["model"]], 
                         f"{item['ss_easy'] * 100:.2f}",
                       f"{item['ss_medium'] * 100:.2f}",
                       f"{item['ss_hard'] * 100:.2f}",
                       f"{item['ms_easy'] * 100:.2f}",
                       f"{item['ms_medium'] * 100:.2f}",
                       f"{item['ms_hard'] * 100:.2f}",
                       f"{item['overall'] * 100:.2f}",
                       ranking_dict[method_name_adic[item["method"]] + model_name_adic[item["model"]]]])
agent_df = pd.DataFrame.from_records(agent_df, columns=AGENT_COLS)
agent_df = agent_df.sort_values(by=["Ranking"], ascending=False)
agent_df = agent_df[AGENT_COLS]

RAG_COLS = ["System", "SS Easy", "SS Medium", "SS Hard", "MS Easy", "MS Meduium", "MS Hard", "Overall", NONE_COL]
RAG_TYPES = ["str", "number", "number", "number", "number", "number", "number", "number", "number" , "number"]

rag_ranking = []
with open("rag_result.jsonl", "r") as f:
    for line in f:
        item = json.loads(line)
        rag_ranking.append([rag_name_adic[item["system"]], item["overall"]])
rag_ranking = sorted(rag_ranking, key=lambda x: x[1], reverse=False)
ranking_dict = {}
for i, (system, score) in enumerate(rag_ranking):
    ranking_dict[system] = i
rag_df = []
with open("rag_result.jsonl", "r") as f:
    for line in f:
        item = json.loads(line)
        rag_df.append([rag_name_adic[item["system"]],
                       f"{item['ss_easy'] * 100:.2f}",
                       f"{item['ss_medium'] * 100:.2f}",
                       f"{item['ss_hard'] * 100:.2f}",
                       f"{item['ms_easy'] * 100:.2f}",
                       f"{item['ms_medium'] * 100:.2f}",
                       f"{item['ms_hard'] * 100:.2f}",
                       f"{item['overall'] * 100:.2f}",
                       ranking_dict[rag_name_adic[item["system"]]]])

rag_df = pd.DataFrame.from_records(rag_df, columns=RAG_COLS)
rag_df = rag_df.sort_values(by=["Ranking"], ascending=False)
rag_df = rag_df[RAG_COLS]

demo = gr.Blocks(css=CUSTOM_CSS)
with demo:
    gr.HTML(TITLE)
    gr.Markdown(INTRO_TEXT, elem_classes="markdown-text")
    gr.Markdown(HOW_TO, elem_classes="markdown-text")
    gr.Markdown("## Leaderboard")
    with gr.Group():
        with gr.Tab("Results: Agent 🤖️"):
            leaderboard_table_test = gr.components.Dataframe(
                value=agent_df, datatype=AGENT_TYPES, interactive=False,
                column_widths = ["20%"] * len(agent_df.columns)
            )
        with gr.Tab("Results: RAG-system 🔍"):
            leaderboard_table_val = gr.components.Dataframe(
                value=rag_df, datatype=RAG_TYPES, interactive=False,
                column_widths=["20%"] 
        )
    gr.Markdown("SS denotes single-source, and MS denotes multi-source. Easy, Medium, and Hard denote the difficulty level of the question.")

    gr.Markdown(CREDIT, elem_classes="markdown-text")
    gr.Markdown(CITATION, elem_classes="markdown-text")

demo.launch(share=True)