leaderboard / app.py
clefourrier's picture
clefourrier HF staff
Update app.py
a163e5c
raw
history blame
17.4 kB
import json
import os
import gradio as gr
import pandas as pd
from apscheduler.schedulers.background import BackgroundScheduler
from huggingface_hub import snapshot_download
from src.display.about import (
CITATION_BUTTON_LABEL,
CITATION_BUTTON_TEXT,
EVALUATION_QUEUE_TEXT,
INTRODUCTION_TEXT,
LLM_BENCHMARKS_TEXT,
TITLE,
)
from src.display.css_html_js import custom_css, get_window_url_params
from src.display.utils import (
BENCHMARK_COLS,
COLS,
EVAL_COLS,
EVAL_TYPES,
NUMERIC_INTERVALS,
TYPES,
AutoEvalColumn,
ModelType,
fields,
)
from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, H4_TOKEN, IS_PUBLIC, QUEUE_REPO, REPO_ID, RESULTS_REPO
from src.populate import get_evaluation_queue_df, get_leaderboard_df
from src.submission.submit import add_new_eval
from src.tools.collections import update_collections
from src.tools.plots import (
HUMAN_BASELINES,
create_metric_plot_obj,
create_plot_df,
create_scores_df,
join_model_info_with_results,
)
def restart_space():
API.restart_space(repo_id=REPO_ID, token=H4_TOKEN)
try:
snapshot_download(
repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30
)
except Exception:
restart_space()
try:
snapshot_download(
repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30
)
except Exception:
restart_space()
original_df = get_leaderboard_df(EVAL_RESULTS_PATH, COLS, BENCHMARK_COLS)
update_collections(original_df.copy())
leaderboard_df = original_df.copy()
# models = original_df["model_name_for_query"].tolist() # needed for model backlinks in their to the leaderboard
# plot_df = create_plot_df(create_scores_df(join_model_info_with_results(original_df)))
# to_be_dumped = f"models = {repr(models)}\n"
(
finished_eval_queue_df,
running_eval_queue_df,
pending_eval_queue_df,
) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
# Basics
#def change_tab(query_param: str):
# query_param = query_param.replace("'", '"')
# query_param = json.loads(query_param)
# if isinstance(query_param, dict) and "tab" in query_param and query_param["tab"] == "evaluation":
# return gr.Tabs.update(selected=1)
# else:
# return gr.Tabs.update(selected=0)
# Searching and filtering
def update_table(
hidden_df: pd.DataFrame,
columns: list,
type_query: list,
precision_query: str,
size_query: list,
show_deleted: bool,
query: str,
):
filtered_df = filter_models(hidden_df, type_query, size_query, precision_query, show_deleted)
filtered_df = filter_queries(query, filtered_df)
df = select_columns(filtered_df, columns)
return df
def search_table(df: pd.DataFrame, query: str) -> pd.DataFrame:
return df[(df[AutoEvalColumn.dummy.name].str.contains(query, case=False))]
def select_columns(df: pd.DataFrame, columns: list) -> pd.DataFrame:
always_here_cols = [
AutoEvalColumn.model_type_symbol.name,
AutoEvalColumn.model.name,
]
# We use COLS to maintain sorting
filtered_df = df[
always_here_cols + [c for c in COLS if c in df.columns and c in columns] + [AutoEvalColumn.dummy.name]
]
return filtered_df
def filter_queries(query: str, filtered_df: pd.DataFrame):
"""Added by Abishek"""
final_df = []
if query != "":
queries = [q.strip() for q in query.split(";")]
for _q in queries:
_q = _q.strip()
if _q != "":
temp_filtered_df = search_table(filtered_df, _q)
if len(temp_filtered_df) > 0:
final_df.append(temp_filtered_df)
if len(final_df) > 0:
filtered_df = pd.concat(final_df)
filtered_df = filtered_df.drop_duplicates(
subset=[AutoEvalColumn.model.name, AutoEvalColumn.precision.name, AutoEvalColumn.revision.name]
)
return filtered_df
def filter_models(
df: pd.DataFrame, type_query: list, size_query: list, precision_query: list, show_deleted: bool
) -> pd.DataFrame:
# Show all models
if show_deleted:
filtered_df = df
else: # Show only still on the hub models
filtered_df = df[df[AutoEvalColumn.still_on_hub.name] == True]
type_emoji = [t[0] for t in type_query]
filtered_df = filtered_df[df[AutoEvalColumn.model_type_symbol.name].isin(type_emoji)]
filtered_df = filtered_df[df[AutoEvalColumn.precision.name].isin(precision_query + ["None"])]
numeric_interval = pd.IntervalIndex(sorted([NUMERIC_INTERVALS[s] for s in size_query]))
params_column = pd.to_numeric(df[AutoEvalColumn.params.name], errors="coerce")
mask = params_column.apply(lambda x: any(numeric_interval.contains(x)))
filtered_df = filtered_df.loc[mask]
return filtered_df
demo = gr.Blocks(css=custom_css)
with demo:
gr.HTML(TITLE)
gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
with gr.Tabs(elem_classes="tab-buttons") as tabs:
with gr.TabItem("πŸ… LLM Benchmark", elem_id="llm-benchmark-tab-table", id=0):
with gr.Row():
with gr.Column():
with gr.Row():
search_bar = gr.Textbox(
placeholder=" πŸ” Search for your model (separate multiple queries with `;`) and press ENTER...",
show_label=False,
elem_id="search-bar",
)
with gr.Row():
shown_columns = gr.CheckboxGroup(
choices=[
c.name
for c in fields(AutoEvalColumn)
if not c.hidden and not c.never_hidden and not c.dummy
],
value=[
c.name
for c in fields(AutoEvalColumn)
if c.displayed_by_default and not c.hidden and not c.never_hidden
],
label="Select columns to show",
elem_id="column-select",
interactive=True,
)
with gr.Row():
deleted_models_visibility = gr.Checkbox(
value=False, label="Show gated/private/deleted models", interactive=True
)
with gr.Column(min_width=320):
#with gr.Box(elem_id="box-filter"):
filter_columns_type = gr.CheckboxGroup(
label="Model types",
choices=[t.to_str() for t in ModelType],
value=[t.to_str() for t in ModelType],
interactive=True,
elem_id="filter-columns-type",
)
filter_columns_precision = gr.CheckboxGroup(
label="Precision",
choices=["torch.float16", "torch.bfloat16", "torch.float32", "8bit", "4bit", "GPTQ"],
value=["torch.float16", "torch.bfloat16", "torch.float32", "8bit", "4bit", "GPTQ"],
interactive=True,
elem_id="filter-columns-precision",
)
filter_columns_size = gr.CheckboxGroup(
label="Model sizes (in billions of parameters)",
choices=list(NUMERIC_INTERVALS.keys()),
value=list(NUMERIC_INTERVALS.keys()),
interactive=True,
elem_id="filter-columns-size",
)
leaderboard_table = gr.components.Dataframe(
value=leaderboard_df[
[c.name for c in fields(AutoEvalColumn) if c.never_hidden]
+ shown_columns.value
+ [AutoEvalColumn.dummy.name]
],
headers=[c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value,
datatype=TYPES,
elem_id="leaderboard-table",
interactive=False,
visible=True,
column_widths=["2%", "33%"]
)
# Dummy leaderboard for handling the case when the user uses backspace key
hidden_leaderboard_table_for_search = gr.components.Dataframe(
value=original_df[COLS],
headers=COLS,
datatype=TYPES,
visible=False,
)
search_bar.submit(
update_table,
[
hidden_leaderboard_table_for_search,
shown_columns,
filter_columns_type,
filter_columns_precision,
filter_columns_size,
deleted_models_visibility,
search_bar,
],
leaderboard_table,
)
shown_columns.change(
update_table,
[
hidden_leaderboard_table_for_search,
shown_columns,
filter_columns_type,
filter_columns_precision,
filter_columns_size,
deleted_models_visibility,
search_bar,
],
leaderboard_table,
queue=True,
)
filter_columns_type.change(
update_table,
[
hidden_leaderboard_table_for_search,
shown_columns,
filter_columns_type,
filter_columns_precision,
filter_columns_size,
deleted_models_visibility,
search_bar,
],
leaderboard_table,
queue=True,
)
filter_columns_precision.change(
update_table,
[
hidden_leaderboard_table_for_search,
shown_columns,
filter_columns_type,
filter_columns_precision,
filter_columns_size,
deleted_models_visibility,
search_bar,
],
leaderboard_table,
queue=True,
)
filter_columns_size.change(
update_table,
[
hidden_leaderboard_table_for_search,
shown_columns,
filter_columns_type,
filter_columns_precision,
filter_columns_size,
deleted_models_visibility,
search_bar,
],
leaderboard_table,
queue=True,
)
deleted_models_visibility.change(
update_table,
[
hidden_leaderboard_table_for_search,
shown_columns,
filter_columns_type,
filter_columns_precision,
filter_columns_size,
deleted_models_visibility,
search_bar,
],
leaderboard_table,
queue=True,
)
# with gr.TabItem("πŸ“ˆ
# evolution through time", elem_id="llm-benchmark-tab-table", id=4):
# with gr.Row():
# with gr.Column():
# chart = create_metric_plot_obj(
# plot_df,
# ["Average ⬆️"],
# HUMAN_BASELINES,
# title="Average of Top Scores and Human Baseline Over Time",
# )
# gr.Plot(value=chart, interactive=False, width=500, height=500)
# with gr.Column():
# chart = create_metric_plot_obj(
# plot_df,
# ["ARC", "HellaSwag", "MMLU", "TruthfulQA", "Winogrande", "GSM8K", "DROP"],
# HUMAN_BASELINES,
# title="Top Scores and Human Baseline Over Time",
# )
# gr.Plot(value=chart, interactive=False, width=500, height=500)
with gr.TabItem("πŸ“ About", elem_id="llm-benchmark-tab-table", id=2):
gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
with gr.TabItem("πŸš€ Submit here! ", elem_id="llm-benchmark-tab-table", id=3):
with gr.Column():
with gr.Row():
gr.Markdown(EVALUATION_QUEUE_TEXT, elem_classes="markdown-text")
with gr.Column():
with gr.Accordion(
f"βœ… Finished Evaluations ({len(finished_eval_queue_df)})",
open=False,
):
with gr.Row():
finished_eval_table = gr.components.Dataframe(
value=finished_eval_queue_df,
headers=EVAL_COLS,
datatype=EVAL_TYPES,
row_count=5,
)
with gr.Accordion(
f"πŸ”„ Running Evaluation Queue ({len(running_eval_queue_df)})",
open=False,
):
with gr.Row():
running_eval_table = gr.components.Dataframe(
value=running_eval_queue_df,
headers=EVAL_COLS,
datatype=EVAL_TYPES,
row_count=5,
)
with gr.Accordion(
f"⏳ Pending Evaluation Queue ({len(pending_eval_queue_df)})",
open=False,
):
with gr.Row():
pending_eval_table = gr.components.Dataframe(
value=pending_eval_queue_df,
headers=EVAL_COLS,
datatype=EVAL_TYPES,
row_count=5,
)
with gr.Row():
gr.Markdown("# βœ‰οΈβœ¨ Submit your model here!", elem_classes="markdown-text")
with gr.Row():
with gr.Column():
model_name_textbox = gr.Textbox(label="Model name")
revision_name_textbox = gr.Textbox(label="Revision commit", placeholder="main")
private = gr.Checkbox(False, label="Private", visible=not IS_PUBLIC)
model_type = gr.Dropdown(
choices=[t.to_str(" : ") for t in ModelType],
label="Model type",
multiselect=False,
value=None,
interactive=True,
)
with gr.Column():
precision = gr.Dropdown(
choices=["float16", "bfloat16", "8bit (LLM.int8)", "4bit (QLoRA / FP4)", "GPTQ"],
label="Precision",
multiselect=False,
value="float16",
interactive=True,
)
weight_type = gr.Dropdown(
choices=["Original", "Delta", "Adapter"],
label="Weights type",
multiselect=False,
value="Original",
interactive=True,
)
base_model_name_textbox = gr.Textbox(label="Base model (for delta or adapter weights)")
submit_button = gr.Button("Submit Eval")
submission_result = gr.Markdown()
submit_button.click(
add_new_eval,
[
model_name_textbox,
base_model_name_textbox,
revision_name_textbox,
precision,
private,
weight_type,
model_type,
],
submission_result,
)
with gr.Row():
with gr.Accordion("πŸ“™ Citation", open=False):
citation_button = gr.Textbox(
value=CITATION_BUTTON_TEXT,
label=CITATION_BUTTON_LABEL,
lines=20,
elem_id="citation-button",
show_copy_button=True,
)
#dummy = gr.Textbox(visible=False)
#demo.load(
# change_tab,
# dummy,
# tabs,
# js=get_window_url_params,
#)
scheduler = BackgroundScheduler()
scheduler.add_job(restart_space, "interval", seconds=1800)
scheduler.start()
demo.queue().launch()