Spaces:
Sleeping
Sleeping
File size: 4,443 Bytes
df490c1 dbfe5ff df490c1 437e5ce df490c1 0e7c595 df490c1 0e7c595 df490c1 0e7c595 df490c1 f216670 df490c1 f216670 df490c1 f216670 df490c1 63a2b2d df490c1 32dc043 ada2d66 df490c1 f216670 df490c1 8a53052 600915c df490c1 41ae4ba df490c1 |
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 |
import copy as cp
import json
from collections import defaultdict
from urllib.request import urlopen
import gradio as gr
import numpy as np
import pandas as pd
from meta_data import DEFAULT_BENCH, META_FIELDS, RESULTS
def load_results_local():
with open(RESULTS, 'r') as infile:
data = json.load(infile)
return data
def nth_large(val, vals):
return sum([1 for v in vals if v > val]) + 1
def model_size_flag(sz, FIELDS):
if pd.isna(sz) and 'Unknown' in FIELDS:
return True
if pd.isna(sz):
return False
if '7B' in FIELDS and sz == 7:
return True
if '13B' in FIELDS and sz == 13:
return True
if '70B' in FIELDS and sz == 70:
return True
return False
def model_type_flag(line, FIELDS):
if 'OpenSource' in FIELDS and line['OpenSource'] == 'Yes':
return True
if 'API' in FIELDS and line['OpenSource'] == 'No' and line['Verified'] == 'Yes':
return True
# if 'Proprietary' in FIELDS and line['OpenSource'] == 'No' and line['Verified'] == 'No':
# return True
if 'Commercial LLMs' in FIELDS and line['Commercial LLMs'] == 'Yes':
return True
if 'General LLMs' in FIELDS and line['General LLMs'] == 'Yes':
return True
if 'Medical LLMs' in FIELDS and line['Medical LLMs'] == 'Yes':
return True
if 'SOTA' in FIELDS and line['SOTA'] == 'Yes':
return True
return False
def BUILD_L1_DF(results, fields):
check_box = {}
check_box['essential'] = ['Method', 'Param (B)']
# revise there to set default dataset
check_box['required'] = ['Avg Score', 'Avg Rank'] + DEFAULT_BENCH
check_box['avg'] = ['Avg Score', 'Avg Rank']
check_box['all'] = check_box['avg'] + fields
type_map = defaultdict(lambda: 'number')
type_map['Method'] = 'html'
type_map['Language Model'] = type_map['Vision Model'] = type_map['OpenSource'] = type_map['Verified'] = 'str'
check_box['type_map'] = type_map
df = generate_table(results, fields)
return df, check_box
def generate_table(results, fields):
def get_mmbench_v11(item):
assert 'MMBench_TEST_CN_V11' in item and 'MMBench_TEST_EN_V11' in item
val = (item['MMBench_TEST_CN_V11']['Overall'] + item['MMBench_TEST_EN_V11']['Overall']) / 2
val = float(f'{val:.1f}')
return val
res = defaultdict(list)
for i, m in enumerate(results):
item = results[m]
meta = item['META']
for k in META_FIELDS:
if k == 'Param (B)':
param = meta['Parameters']
res[k].append(float(param.replace('B', '')) if param != '' else None)
elif k == 'Method':
name = meta['Method'][0]
res[k].append(f'{name}')
res['name'].append(name)
else:
res[k].append(meta[k])
scores, ranks = [], []
for d in fields:
key_name = 'Overall' if d != 'OCRBench' else 'Final Score'
# Every Model should have MMBench_V11 results
if d == 'MMBench_V11':
val = get_mmbench_v11(item)
res[d].append(val)
scores.append(val)
ranks.append(nth_large(val, [get_mmbench_v11(x) for x in results.values()]))
elif d in item:
res[d].append(item[d][key_name])
if d == 'MME':
scores.append(item[d][key_name] / 28)
elif d == 'OCRBench':
scores.append(item[d][key_name] / 10)
else:
scores.append(item[d][key_name])
ranks.append(nth_large(item[d][key_name], [x[d][key_name] for x in results.values() if d in x]))
else:
res[d].append(None)
scores.append(None)
ranks.append(None)
res['Avg Score'].append(round(np.mean(scores), 1) if None not in scores else None)
res['Avg Rank'].append(round(np.mean(ranks), 2) if None not in ranks else None)
df = pd.DataFrame(res)
valid, missing = df[~pd.isna(df['Avg Score'])], df[pd.isna(df['Avg Score'])]
valid = valid.sort_values('Avg Score')
valid = valid.iloc[::-1]
if len(fields):
missing = missing.sort_values('MMBench_V11' if 'MMBench_V11' in fields else fields[0])
missing = missing.iloc[::-1]
df = pd.concat([valid, missing])
return df
|