Spaces:
Running
Running
File size: 10,569 Bytes
01f830e 87a0436 e6cfa34 87a0436 2a72d1d 87a0436 01f830e ac62be1 01f830e ac62be1 87a0436 01f830e ac62be1 87a0436 ac62be1 72b6fa6 ac62be1 06c4694 01f830e e6cfa34 06c4694 ac62be1 e6cfa34 01f830e 06c4694 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 |
import gradio as gr
import plotly.graph_objects as go
from datasets import load_dataset
import plotly.graph_objects as go
from plotly.subplots import make_subplots
import os
# ==
import json
import pandas as pd
import plotly.express as px
import plotly.graph_objects as go
# CATEGORIES = ["Writing", "Roleplay", "Reasoning", "Math", "Coding", "Extraction", "STEM", "Humanities"]
CATEGORIES = ["task-solving", "math-reasoning", "general-instruction", "natural-question", "safety"]
LANGS = ['en', 'vi', 'th', 'id', 'km', 'lo', 'ms', 'my', 'tl']
force_download = bool(int(os.environ.get("force_download", "1")))
HF_TOKEN = str(os.environ.get("HF_TOKEN", ""))
DATA_SET_REPO_PATH = str(os.environ.get("DATA_SET_REPO_PATH", ""))
PERFORMANCE_FILENAME = str(os.environ.get("PERFORMANCE_FILENAME", "gpt4_single_json.csv"))
MODEL_DFRAME = None
CATEGORIES = ["task-solving", "math-reasoning", "general-instruction", "natural-question", "safety"]
LANGS = ['en', 'vi', 'th', 'id', 'km', 'lo', 'ms', 'my', 'tl']
FORCE_DOWNLOAD = bool(int(os.environ.get("FORCE_DOWNLOAD", "0")))
HF_TOKEN = str(os.environ.get("HF_TOKEN", ""))
DATA_SET_REPO_PATH = str(os.environ.get("DATA_SET_REPO_PATH", "SeaLLMs/Sea-bench"))
PERFORMANCE_FILENAME = str(os.environ.get("PERFORMANCE_FILENAME", "model_judgment/gpt-4_single.jsonl"))
QUESTION_FILE_NAME = str(os.environ.get("QUESTION_FILE_NAME", "question.jsonl"))
rename_map = {
"seallm-13b-chat": "SeaLLM-13b",
# "polylm-13b": "PolyLM-13b",
"qwen-14b": "Qwen-14b",
"gpt-3.5-turbo": "GPT-3.5-turbo",
}
rename_map = {
"SeaLLM-7B-v2.5": "SeaLLM-7B-v2.5",
# "SeaLLM-7B-v2.5-CUT3": "SeaLLM-7B-v2.5-CUT",
"SeaLLM-7B-v2": "SeaLLM-7B-v2",
"Sailor-7B-Chat": "Sailor-7B-Chat",
# "SeaLLM-13B-v1": "SeaLLM-13B-v1",
# "llama2-13b": "LLaMA2-13B",
# "sealion7b_instr": "SeaLion-7b",
# "polylm": "PolyLM-13b",
# "qwen": "Qwen-14b",
"qwen_15_7b_chat": "Qwen1.5-7B-chat",
"gpt-3.5-turbo": "GPT-3.5-turbo",
# "gpt-4-1106-preview": "GPT-4-turbo",
}
CATEGORIES = [ "task-solving", "math-reasoning", "general-instruction", "natural-question", "safety", ]
CATEGORIES_NAMES = {
"task-solving": 'Task-solving',
"math-reasoning": 'Math',
"general-instruction": 'General-instruction',
"natural-question": 'NaturalQA',
"safety": 'Safety',
}
LANGS = ['en', 'vi', 'id', 'ms', 'tl', 'th', 'km', 'lo', 'my']
LANG_NAMES = {
'en': 'eng',
'vi': 'vie',
'th': 'tha',
'id': 'ind',
'km': 'khm',
'lo': 'lao',
'ms': 'msa',
'my': 'mya',
'tl': 'tgl',
}
MODEL_DFRAME = None
def read_jsonl_report(question_path, file_path):
with open(question_path, 'r') as f:
questions = [
json.loads(x)
for x in f
]
questions = {
q['question_id']: q
for q in questions
}
q2result = []
fin = open(file_path, "r")
for line in fin:
obj = json.loads(line)
obj["category"] = questions[obj['question_id']]['category']
obj["lang"] = questions[obj['question_id']]['lang']
q2result.append(obj)
df = pd.DataFrame(q2result)
return df
def get_report_df_from_jsonl():
from huggingface_hub import hf_hub_download
assert DATA_SET_REPO_PATH != ''
assert HF_TOKEN != ''
repo_id = DATA_SET_REPO_PATH
question_path = hf_hub_download(
repo_id=repo_id,
filename=QUESTION_FILE_NAME,
force_download=FORCE_DOWNLOAD,
local_dir='./hf_cache',
repo_type="dataset",
token=HF_TOKEN
)
file_path = hf_hub_download(
repo_id=repo_id,
filename=PERFORMANCE_FILENAME,
force_download=FORCE_DOWNLOAD,
local_dir='./hf_cache',
repo_type="dataset",
token=HF_TOKEN
)
print(f'Downloaded file at {question_path}/ {file_path} from {DATA_SET_REPO_PATH} / {PERFORMANCE_FILENAME}')
return read_jsonl_report(question_path, file_path)
def get_model_df():
global MODEL_DFRAME
if isinstance(MODEL_DFRAME, pd.DataFrame):
print(f'Load cache data frame')
return MODEL_DFRAME
from huggingface_hub import hf_hub_download
assert DATA_SET_REPO_PATH != ''
assert HF_TOKEN != ''
repo_id = DATA_SET_REPO_PATH
filename = PERFORMANCE_FILENAME
# data_path = f"{DATA_SET_REPO_PATH}/{PERFORMANCE_FILENAME}"
file_path = hf_hub_download(
repo_id=repo_id,
filename=filename,
force_download=force_download,
local_dir='./hf_cache',
repo_type="dataset",
token=HF_TOKEN
)
print(f'Downloaded file at {file_path} from {DATA_SET_REPO_PATH} / {PERFORMANCE_FILENAME}')
MODEL_DFRAME = pd.read_csv(file_path)
return MODEL_DFRAME
def aggregate_df(df, model_dict, category_name, categories):
scores_all = []
all_models = df["model"].unique()
for model in all_models:
for i, cat in enumerate(categories):
# filter category/model, and score format error (<1% case)
res = df[(df[category_name]==cat) & (df["model"]==model) & (df["score"] >= 0)]
score = res["score"].mean()
cat_name = cat
scores_all.append({"model": model, category_name: cat_name, "score": score})
target_models = list(model_dict.keys())
scores_target = [scores_all[i] for i in range(len(scores_all)) if scores_all[i]["model"] in target_models]
scores_target = sorted(scores_target, key=lambda x: target_models.index(x["model"]), reverse=True)
df_score = pd.DataFrame(scores_target)
df_score = df_score[df_score["model"].isin(target_models)]
rename_map = model_dict
for k, v in rename_map.items():
df_score.replace(k, v, inplace=True)
return df_score
def polar_subplot(fig, dframe, model_names, category_label, category_names, row, col, showlegend=True):
# cat category
colors = px.colors.qualitative.Plotly
for i, (model, model_name) in enumerate(model_names):
cat_list = dframe[dframe['model'] == model_name][category_label].tolist()
score_list = dframe[dframe['model'] == model_name]['score'].tolist()
cat_list += [cat_list[0]]
cat_list = [category_names[x] for x in cat_list]
score_list += [score_list[0]]
polar = go.Scatterpolar(
name = model_name,
r = score_list,
theta = cat_list,
legendgroup=f'{i}',
marker=dict(color=colors[i]),
hovertemplate="""Score: %{r:.2f}""",
showlegend=showlegend,
)
fig.add_trace(polar, row, col)
def plot_agg_fn():
df = get_report_df_from_jsonl()
all_models = df["model"].unique()
model_names = list(rename_map.items())
colors = px.colors.qualitative.Plotly
cat_df = aggregate_df(df, rename_map, "category", CATEGORIES, )
lang_df = aggregate_df(df, rename_map, "lang", LANGS, )
fig = make_subplots(
rows=1, cols=2,
specs=[[{'type': 'polar'}]*2],
subplot_titles=("By Category", "By Language"),
)
fig.layout.annotations[0].y = 1.05
fig.layout.annotations[1].y = 1.05
# cat category
for i, (model, model_name) in enumerate(model_names):
cat_list = cat_df[cat_df['model'] == model_name]['category'].tolist()
score_list = cat_df[cat_df['model'] == model_name]['score'].tolist()
cat_list += [cat_list[0]]
cat_list = [CATEGORIES_NAMES[x] for x in cat_list]
score_list += [score_list[0]]
polar = go.Scatterpolar(
name = model_name,
r = score_list,
theta = cat_list,
legendgroup=f'{i}',
marker=dict(color=colors[i]),
hovertemplate="""Score: %{r:.2f}""",
)
fig.add_trace(polar, 1, 1)
# cat langs
for i, (model, model_name) in enumerate(model_names):
cat_list = lang_df[lang_df['model'] == model_name]['lang'].tolist()
score_list = lang_df[lang_df['model'] == model_name]['score'].tolist()
cat_list += [cat_list[0]]
score_list += [score_list[0]]
cat_list = [LANG_NAMES[x] for x in cat_list]
polar = go.Scatterpolar(
name = model_name,
r = score_list,
theta = cat_list,
legendgroup=f'{i}',
marker=dict(color=colors[i]),
hovertemplate="""Score: %{r:.2f}""",
showlegend=False,
)
fig.add_trace(polar, 1, 2)
polar_config = dict(
angularaxis = dict(
rotation=90, # start position of angular axis
),
radialaxis = dict(
range=[0, 10],
),
)
fig.update_layout(
polar = polar_config,
polar2 = polar_config,
title='Sea-Bench (rated by GPT-4)',
)
return fig
def plot_by_lang_fn():
df = get_report_df_from_jsonl()
model_names = list(rename_map.items())
fig = make_subplots(
rows=3, cols=3,
specs=[[{'type': 'polar'}]*3] * 3,
# subplot_titles=("By Category", "By Language"),
subplot_titles=[LANG_NAMES[x] for x in LANGS],
# vertical_spacing=1
)
# print(fig.layout.annotations)
for ano in fig.layout.annotations:
ano.y = ano.y + 0.02
has_safety = ['vi', 'id', 'th']
for lang_id, lang in enumerate(LANGS):
cat_names = CATEGORIES if lang in has_safety else [x for x in CATEGORIES if x != 'safety']
cat_lang_df = aggregate_df(df[df['lang'] == lang], rename_map, "category", cat_names, )
row = lang_id // 3 + 1
col = lang_id % 3 + 1
polar_subplot(fig, cat_lang_df, model_names, 'category', CATEGORIES_NAMES, row, col, showlegend=lang_id == 0)
polar_config = dict(
angularaxis = dict(
rotation=90, # start position of angular axis
),
radialaxis = dict(
range=[0, 10],
),
)
layer_kwargs = {f"polar{i}": polar_config for i in range(1, 10)}
fig.update_layout(
title='Sea-Bench - By language (rated by GPT-4)',
height=1000,
width=1500,
**layer_kwargs
)
return fig
description = """
<h1>Sea-bench benchmark leaderboard</h1>
"""
with gr.Blocks() as demo:
with gr.Column():
# gr.Markdown(description)
gr_plot_agg = gr.Plot(label="Aggregated by Category and Language")
# gr_plot_bylang = gr.Plot(label='Breakdown by language')
demo.load(plot_agg_fn, [], gr_plot_agg)
# demo.load(plot_by_lang_fn, [], gr_plot_bylang)
demo.launch()
|