File size: 11,905 Bytes
73adc36 b6acaed 73adc36 b6acaed 73adc36 23f0448 0a35c11 23f0448 73adc36 0a35c11 23f0448 73adc36 23f0448 73adc36 23f0448 5484928 73adc36 0a35c11 73adc36 95be10d 73adc36 23f0448 73adc36 23f0448 73adc36 55fbb52 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 |
import gradio as gr
import pandas as pd
from pathlib import Path
import plotly.express as px
from gradio_leaderboard import Leaderboard, ColumnFilter, SelectColumns
from apscheduler.schedulers.background import BackgroundScheduler
from huggingface_hub import snapshot_download
from src.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
from src.display.utils import (
BENCHMARK_COLS,
COLS,
EVAL_COLS,
EVAL_TYPES,
AutoEvalColumn,
ModelType,
fields,
WeightType,
Precision
)
from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN
from src.populate import get_evaluation_queue_df, get_leaderboard_df
from src.submission.submit import add_new_eval
import base64
def restart_space():
API.restart_space(repo_id=REPO_ID)
def make_rate_chart(df: pd.DataFrame):
"""Return a Plotly bar chart of hallucination rates."""
# long-form dataframe for grouped bars
df_long = df.melt(
id_vars="Models",
value_vars=["RAG Hallucination Rate (%)", "Non-RAG Hallucination Rate (%)"],
var_name="Benchmark",
value_name="Rate",
)
fig = px.bar(
df_long,
x="Models",
y="Rate",
color="Benchmark",
barmode="group",
title="Hallucination Rates by Model",
height=400,
)
fig.update_layout(xaxis_title="", yaxis_title="%")
return fig
def make_leaderboard_plot(df: pd.DataFrame, col: str, title: str, bar_color: str):
"""
Return a horizontal bar chart sorted ascending by `col`.
Lowest value (best) at the top.
"""
df_sorted = df.sort_values(col, ascending=False) # best → worst
fig = px.bar(
df_sorted,
x=col,
y="Models",
orientation="h",
title=title,
text_auto=".2f",
height=400,
color_discrete_sequence=[bar_color],
)
fig.update_traces(textposition="outside", cliponaxis=False)
fig.update_layout(
xaxis_title="Hallucination Rate (%)",
yaxis_title="",
yaxis=dict(dtick=1), # ensure every model shown
margin=dict(l=140, r=60, t=60, b=40)
)
fig.update_traces(textposition="outside")
return fig
def color_scale(s, cmap):
"""
Return background-colour styles for a numeric Series (lower = greener,
higher = redder). Works with any palette length.
"""
colours = px.colors.sequential.__dict__[cmap]
n = len(colours) - 1 # max valid index
rng = s.max() - s.min()
norm = (s - s.min()) / (rng if rng else 1)
return [f"background-color:{colours[int(v * n)]}" for v in 1 - norm]
### Space initialisation
try:
print(EVAL_REQUESTS_PATH)
snapshot_download(
repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
)
except Exception:
# restart_space()
print(f"[WARN] Skipping RESULTS sync: {Exception}")
try:
print(EVAL_RESULTS_PATH)
snapshot_download(
repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
)
except Exception:
# restart_space()
print(f"[WARN] Skipping RESULTS sync: {Exception}")
# LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)
LEADERBOARD_DF = get_leaderboard_df("leaderboard/data/leaderboard.csv")
# (
# finished_eval_queue_df,
# running_eval_queue_df,
# pending_eval_queue_df,
# ) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
def init_leaderboard(df: pd.DataFrame):
if df is None or df.empty:
raise ValueError("Leaderboard DataFrame is empty or None.")
return Leaderboard(
value=df,
datatype=["markdown", "markdown", "number", "number", "number"],
select_columns=SelectColumns(
default_selection=[
"Rank", "Models",
"Average Hallucination Rate (%)",
"RAG Hallucination Rate (%)",
"Non-RAG Hallucination Rate (%)"
],
cant_deselect=["Models", "Rank"],
label="Select Columns to Display:",
),
search_columns=["Models"],
# column_widths=["3%"],
bool_checkboxgroup_label=None,
interactive=False,
)
image_path = "static/kluster-color.png"
with open(image_path, "rb") as img_file:
b64_string = base64.b64encode(img_file.read()).decode("utf-8")
# print("CUSTOM CSS\n", custom_css[-1000:], "\n---------")
demo = gr.Blocks(css=custom_css)
with demo:
gr.HTML(f"""
<div style="text-align: center; margin-top: 2em; margin-bottom: 1em;">
<img src="data:image/png;base64,{b64_string}" alt="kluster.ai logo"
style="height: 80px; display: block; margin-left: auto; margin-right: auto;" />
<div style="font-size: 2.5em; font-weight: bold; margin-top: 0.4em; color: var(--text-color);">
LLM Hallucination Detection Leaderboard
</div>
<div style="font-size: 1.5em; margin-top: 0.5em;">
Evaluating factual accuracy and faithfulness of LLMs in both RAG and real-world knowledge settings with
<a href="https://platform.kluster.ai/verify" target="_blank">
Verify
</a> by
<a href="https://platform.kluster.ai/" target="_blank">
kluster.ai
</a>
</div>
</div>
""")
# gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
with gr.Tabs(elem_classes="tab-buttons") as tabs:
with gr.TabItem("🏅 Hallucination Leaderboard", elem_id="llm-benchmark-tab-table", id=0):
# ---------- Chart ----------
with gr.Row():
gr.Plot(
make_leaderboard_plot(
LEADERBOARD_DF,
"RAG Hallucination Rate (%)",
"RAG Hallucination Rate (lower is better)",
bar_color="#4CAF50",
),
show_label=False,
)
gr.Plot(
make_leaderboard_plot(
LEADERBOARD_DF,
"Non-RAG Hallucination Rate (%)",
"Non-RAG Hallucination Rate (lower is better)",
bar_color="#FF7043",
),
show_label=False,
)
# ---------- Leaderboard ----------
leaderboard = init_leaderboard(LEADERBOARD_DF)
with gr.TabItem("📝 Details", elem_id="llm-benchmark-tab-table", id=2):
gr.Markdown((Path(__file__).parent / "docs.md").read_text())
with gr.TabItem("🚀 Submit Here! ", elem_id="llm-benchmark-tab-table", id=3):
gr.Markdown((Path(__file__).parent / "submit.md").read_text())
# with gr.Column():
# with gr.Row():
# gr.Markdown(EVALUATION_QUEUE_TEXT, elem_classes="markdown-text")
# with gr.Column():
# with gr.Accordion(
# f"✅ 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"🔄 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("# ✉️✨ 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")
# model_type = gr.Dropdown(
# choices=[t.to_str(" : ") for t in ModelType if t != ModelType.Unknown],
# label="Model type",
# multiselect=False,
# value=None,
# interactive=True,
# )
# with gr.Column():
# precision = gr.Dropdown(
# choices=[i.value.name for i in Precision if i != Precision.Unknown],
# label="Precision",
# multiselect=False,
# value="float16",
# interactive=True,
# )
# weight_type = gr.Dropdown(
# choices=[i.value.name for i in WeightType],
# 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,
# weight_type,
# model_type,
# ],
# submission_result,
# )
# with gr.Row():
# with gr.Accordion("📙 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,
# )
scheduler = BackgroundScheduler()
scheduler.add_job(restart_space, "interval", seconds=1800)
scheduler.start()
demo.queue(default_concurrency_limit=40).launch() |