wzxii's picture
Upload 2 files
e5d4298 verified
raw
history blame
9.26 kB
const models = [
{
name: 'GPT-3.5-turbo-1103',
MGI: 0.2326,
temp0_HumanEval: 0.7439,
temp0_HumanEval_ET: 0.5915,
temp0_8_HumanEval: 0.7248,
temp0_8_HumanEval_ET: 0.5778,
link: 'https://platform.openai.com/docs/models/gpt-3-5-turbo',
},
{
name: 'GPT-3.5-turbo-0613',
MGI: 0.1314,
temp0_HumanEval: 0.6585,
temp0_HumanEval_ET: 0.5122,
temp0_8_HumanEval: 0.612,
temp0_8_HumanEval_ET: 0.4776,
link: 'https://platform.openai.com/docs/models/gpt-3-5-turbo',
},
{
name: 'CodeGeeX4-All-9B',
MGI: 0.5182,
temp0_HumanEval: 0.7805,
temp0_HumanEval_ET: 0.622,
temp0_8_HumanEval: 0.7606,
temp0_8_HumanEval_ET: 0.6078,
link: 'https://huggingface.co/THUDM/codegeex4-all-9b',
},
{
name: 'Deepseek-Coder-6.7b',
MGI: 0.0602,
temp0_HumanEval: 0.4817,
temp0_HumanEval_ET: 0.3963,
temp0_8_HumanEval: 0.3388,
temp0_8_HumanEval_ET: 0.2798,
link: 'https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-base',
},
{
name: 'Deepseek-Coder-6.7b-instruct',
MGI: 0.0893,
temp0_HumanEval: 0.7134,
temp0_HumanEval_ET: 0.628,
temp0_8_HumanEval: 0.6494,
temp0_8_HumanEval_ET: 0.5673,
link: 'https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-instruct'
},
{
name: 'CodeLlama-7b',
MGI: 0.0417,
temp0_HumanEval: 0.311,
temp0_HumanEval_ET: 0.2378,
temp0_8_HumanEval: 0.2211,
temp0_8_HumanEval_ET: 0.1832,
link: 'https://huggingface.co/codellama/CodeLlama-7b-hf'
},
{
name: 'CodeLlama-7b-python',
MGI: 0.0525,
temp0_HumanEval: 0.3841,
temp0_HumanEval_ET: 0.3354,
temp0_8_HumanEval: 0.3062,
temp0_8_HumanEval_ET: 0.2568,
link: 'https://huggingface.co/codellama/CodeLlama-7b-Python-hf'
},
{
name: 'CodeLlama-7b-instruct',
MGI: 0.0908,
temp0_HumanEval: 0.3537,
temp0_HumanEval_ET: 0.2988,
temp0_8_HumanEval: 0.2876,
temp0_8_HumanEval_ET: 0.2455,
link: 'https://huggingface.co/codellama/CodeLlama-7b-Instruct-hf'
},
{
name: 'CodeGemma-2b',
MGI: 0.0112,
temp0_HumanEval: 0.3537,
temp0_HumanEval_ET: 0.2805,
temp0_8_HumanEval: 0.2161,
temp0_8_HumanEval_ET: 0.1771,
link: 'https://huggingface.co/google/codegemma-2b',
},
{
name: 'CodeGemma-7b',
MGI: 0.0548,
temp0_HumanEval: 0.4329,
temp0_HumanEval_ET: 0.3537,
temp0_8_HumanEval: 0.3316,
temp0_8_HumanEval_ET: 0.2779,
link: 'https://huggingface.co/google/codegemma-7b',
},
// {
// name: 'CodeGemma-7b-it',
// MGI: 0.0732,
// temp0_HumanEval: 0.5671,
// temp0_HumanEval_ET: 0.4756,
// temp0_8_HumanEval: 0.5076,
// temp0_8_HumanEval_ET: 0.4329,
// link: 'https://huggingface.co/google/codegemma-7b-it',
// },
{
name: 'AixCoder3-7b-base',
MGI: 0.0578,
temp0_HumanEval: 0.5427,
temp0_HumanEval_ET: 0.4451,
temp0_8_HumanEval: 0.4545,
temp0_8_HumanEval_ET: 0.3826,
link: 'https://github.com/aixcoder-plugin/aiXcoder-7B?ref=top.aibase.com',
},
{
name: 'CodeQwen1.5-7b',
MGI: 0.0361,
temp0_HumanEval: 0.439,
temp0_HumanEval_ET: 0.3841,
temp0_8_HumanEval: 0.3884,
temp0_8_HumanEval_ET: 0.3279,
link: 'https://huggingface.co/Qwen/CodeQwen1.5-7B',
},
{
name: 'CodeQwen1.5-7b-Chat',
MGI: 0.1098,
temp0_HumanEval: 0.6951,
temp0_HumanEval_ET: 0.6098,
temp0_8_HumanEval: 0.6368,
temp0_8_HumanEval_ET: 0.5566,
link: 'https://huggingface.co/Qwen/CodeQwen1.5-7B-Chat',
},
{
name: 'Starcoder2-7b',
MGI: 0.0579,
temp0_HumanEval: 0.3659,
temp0_HumanEval_ET: 0.3232,
temp0_8_HumanEval: 0.2905,
temp0_8_HumanEval_ET: 0.2438,
link: 'https://huggingface.co/bigcode/starcoder2-7b',
},
{
name: 'Starcoder2-15b',
MGI: 0.0552,
temp0_HumanEval: 0.4695,
temp0_HumanEval_ET: 0.4024,
temp0_8_HumanEval: 0.3406,
temp0_8_HumanEval_ET: 0.2795,
link: 'https://huggingface.co/bigcode/starcoder2-15b',
},
{
name: 'CodeLlama-13b',
MGI: 0.0412,
temp0_HumanEval: 0.3415,
temp0_HumanEval_ET: 0.2927,
temp0_8_HumanEval: 0.2405,
temp0_8_HumanEval_ET: 0.2015,
link: 'https://huggingface.co/codellama/CodeLlama-13b-hf',
},
{
name: 'CodeLlama-34b',
MGI: 0.0651,
temp0_HumanEval: 0.4817,
temp0_HumanEval_ET: 0.4146,
temp0_8_HumanEval: 0.3478,
temp0_8_HumanEval_ET: 0.2906,
link: 'https://huggingface.co/codellama/CodeLlama-34b-hf',
},
{
name: 'CodeLlama-70b',
MGI: 0.0523,
temp0_HumanEval: 0.5244,
temp0_HumanEval_ET: 0.4512,
temp0_8_HumanEval: 0.4317,
temp0_8_HumanEval_ET: 0.3615,
link: 'https://huggingface.co/codellama/CodeLlama-70b-hf',
}
]
function renderRows(model) {
return `
<tr>
<td class="td_a"><a href=${model.link} target="_blank">${model.name}</a></td>
<td class="td_value">${model.MGI}</td>
<td class="td_value">${model.temp0_HumanEval}</td>
<td class="td_value">${model.temp0_HumanEval_ET}</td>
<td class="td_value">${model.temp0_8_HumanEval}</td>
<td class="td_value">${model.temp0_8_HumanEval_ET}</td>
</tr>
`
}
function renderTables() {
let html = ''
// let rows = models.length + 1
// html += `
// <tr>
// <td class="td_HumanEval" rowspan="${rows}">HumanEval</td>
// </tr>
// `
models.forEach(model => {
html += renderRows(model);
});
document.querySelector('tbody').innerHTML = html;
}
function adjustBackgroundColor() {
var evalTable = document.getElementById('evalTable');
var rows = evalTable.getElementsByTagName('tr');
for(var i = 0; i < rows.length; i++){
rows[i].style.backgroundColor = (i < 2 ? '#ffffff' : (i % 2 === 0 ? '#fbfbfb' : '#ffffff'));
}
}
renderTables()
adjustBackgroundColor()
document.querySelector('table').addEventListener('click', (e) => {
if (e.target.matches('button')) {
let direction = e.target.dataset.direction
let type = e.target.dataset.type
models.sort(orderBy(direction, type))
renderTables()
adjustBackgroundColor()
e.target.dataset.direction = (direction === 'asc' ? 'desc' : 'asc')
e.target.style.transform = (direction === 'asc' ? 'rotate(0deg)' : 'rotate(180deg)')
}
})
function orderBy(direction, type) {
return function (a, b) {
let comparision = 0
let value = (direction === 'desc' ? 1 : -1)
switch (type) {
case 'name':
if (a.name.toLowerCase() > b.name.toLowerCase()) {
comparision = value
} else {
comparision = value * -1
}
break;
case 'MGI':
if (a.MGI > b.MGI) {
comparision = value
} else {
comparision = value * -1
}
break;
case 'temp0_HumanEval':
if (a.temp0_HumanEval > b.temp0_HumanEval) {
comparision = value
} else {
comparision = value * -1
}
break;
case 'temp0_HumanEval_ET':
if (a.temp0_HumanEval_ET > b.temp0_HumanEval_ET) {
comparision = value
} else {
comparision = value * -1
}
break;
case 'temp0_8_HumanEval':
if (a.temp0_8_HumanEval > b.temp0_8_HumanEval) {
comparision = value
} else {
comparision = value * -1
}
break;
case 'temp0_8_HumanEval_ET':
if (a.temp0_8_HumanEval_ET > b.temp0_8_HumanEval_ET) {
comparision = value
} else {
comparision = value * -1
}
break;
}
return comparision
}
}
var btns = document.getElementsByClassName("button_sort");
function adjustBtnForm(event) {
var btn = event.target;
if (btn.classList.contains("button_sort")) {
btn.style.borderTop = '8px solid orange';
for (var i = 0; i < btns.length; i++) {
if (btns[i] !== btn) {
btns[i].style.borderTop = '8px solid #9a9a9a';
btns[i].style.transform = 'rotate(180deg)';
}
}
}
}
for (var i = 0; i < btns.length; i++) {
btns[i].addEventListener("click", adjustBtnForm);
}