Commit Graph

485 Commits

Author SHA1 Message Date
xiongliangcheng
289ad342f8 !1903 【mcore】添加MiniCPM3-4b适配
Merge pull request !1903 from xiongliangcheng/master
2024-11-29 08:02:16 +00:00
wucong
3f52981fb0 !1962 修复llama2-7b训练16步左右oom
Merge pull request !1962 from wucong/fix_oom
2024-11-28 14:16:08 +00:00
商元义
40d16b17ba !1933 添加Qwen2.5-LoRA微调,修改全参微调bug
Merge pull request !1933 from 商元义/qwen25_0point5b
2024-11-27 08:09:36 +00:00
glhyy
7092f3c537 !1939 文档更新
Merge pull request !1939 from glhyy/master
2024-11-27 06:39:09 +00:00
glhyy
db300e7a9f !1918 更新dpo、simpo相关文档
Merge pull request !1918 from glhyy/master
2024-11-26 01:36:39 +00:00
wucong
b7dce4d1e0 !1917 添加分支与标签说明
Merge pull request !1917 from wucong/addBranchReadme
2024-11-25 10:37:48 +00:00
qyz
718d8018e2 !1911 增加llama2、mixtral的lora权重转换脚本
Merge pull request !1911 from qyz/master
2024-11-25 01:23:41 +00:00
Yulin Cao
e1eb630895 !1915 InternLM预训练脚本问题修改、用户指南问题修改
Merge pull request !1915 from Yulin Cao/Internlm
2024-11-22 06:29:09 +00:00
Yulin Cao
3e9f90fbf7 !1913 Qwen2.5-Math系列模型适配
Merge pull request !1913 from Yulin Cao/master
2024-11-22 06:13:09 +00:00
Yulin Cao
65ff3e2e35 !1916 yi1.5-6b最佳性能更新
Merge pull request !1916 from Yulin Cao/yi
2024-11-22 02:39:13 +00:00
glhyy
00c6b6b262 !1858 dpo、simpo方案特性支持:支持vpp、dpp、ep、cp、断点续训等
Merge pull request !1858 from glhyy/master
2024-11-21 03:31:39 +00:00
wurongrong008
295497a44a !1830 Qwen2.5代码大模型适配
Merge pull request !1830 from wurongrong008/qwen25_coder
2024-11-19 03:42:05 +00:00
商元义
94768fc246 !1904 添加Qwen2.5-72B模型
Merge pull request !1904 from 商元义/qwen25_0point5b
2024-11-18 08:59:00 +00:00
RuanZhiXiang
d77ddc5e17 !1806 Optim: llama3 qwen系列模型 预训练性能提升
Merge pull request !1806 from RuanZhiXiang/optimization-llama3-qwen
2024-11-18 08:29:28 +00:00
商元义
75ea1ebb86 !1901 添加Qwen2.5-0.5B适配
Merge pull request !1901 from 商元义/qwen25_0point5b
2024-11-16 06:25:33 +00:00
徐源徽
6bc81ad1f3 !1839 新增qwen2.5-7B全参微调脚本
Merge pull request !1839 from 徐源徽/master
2024-11-16 06:24:48 +00:00
wurongrong008
000fd4b687 !1838 强化学习框架搭建,支持llama3.1-8b SimPO训练
Merge pull request !1838 from wurongrong008/simpo
2024-11-15 08:59:07 +00:00
徐源徽
4a3577a50f !1860 Qwen1.5-14b mcore MFU优化
Merge pull request !1860 from 徐源徽/master
2024-11-15 07:21:37 +00:00
RuanZhiXiang
cf51a26e9e !1893 optim: sft微调性能提升
Merge pull request !1893 from RuanZhiXiang/optim-sft
2024-11-15 04:34:08 +00:00
徐源徽
31f2861303 !1900 llama2-70B README性能数据更新
Merge pull request !1900 from 徐源徽/master
2024-11-15 00:52:42 +00:00
Yulin Cao
fdf6998543 !1812 新增Yi1.5系列模型适配
Merge pull request !1812 from Yulin Cao/master
2024-11-15 00:51:38 +00:00
RuanZhiXiang
876c9638b5 !1888 fix: 限制batch_p2p_comm参数关闭条件
Merge pull request !1888 from RuanZhiXiang/pp2vpp-bugfix
2024-11-14 03:15:44 +00:00
RuanZhiXiang
c2e9c1961e !1894 fix: problem of SFT failure caused by wrong trainer usage
Merge pull request !1894 from RuanZhiXiang/sft-bugfix
2024-11-14 02:36:59 +00:00
商元义
b06ad7ee79 !1886 Qwen2-LoRA微调
Merge pull request !1886 from 商元义/LoRA
2024-11-13 01:30:54 +00:00
qyz
93f78fe5d1 !1837 delete old code of ckpt
Merge pull request !1837 from qyz/master
2024-11-12 12:25:13 +00:00
shenjiarun
9d85c0df2c !1875 延长多集群等待时间
Merge pull request !1875 from shenjiarun/master
2024-11-12 01:47:17 +00:00
guoxinjie
c653aaed08 !1855 订正资料中的重计算参数
Merge pull request !1855 from guoxinjie/amend_recompute
2024-11-07 02:16:12 +00:00
liuxinyang
4b28e4f9ea !1853 奖励模型增加加速特性
Merge pull request !1853 from liuxinyang/master
2024-11-07 01:05:52 +00:00
LeiZhenzhen
c0be616e7e !1814 refactor trainer
Merge pull request !1814 from LeiZhenzhen/master
2024-11-06 10:53:02 +00:00
丁子叉
d119d7b927 !1844 [mcore-llm]MLA结构及group_limited_greedy适配CP标准流程
Merge pull request !1844 from 丁子叉/1102_cp
2024-11-05 07:18:18 +00:00
shenjiarun
1813958b96 !1745 新增baichuan2全参微调脚本和相应模版
Merge pull request !1745 from shenjiarun/master
2024-11-04 13:03:12 +00:00
徐源徽
5b913a1cf6 !1840 llama2-13B mcore MFU优化,更新README性能数据
Merge pull request !1840 from 徐源徽/master
2024-11-02 03:14:14 +00:00
徐源徽
9b70bd5cff !1829 llama2-7B mcore MFU优化,更新README性能数据
Merge pull request !1829 from 徐源徽/master
2024-10-31 10:47:55 +00:00
shenjiarun
7dc8bd4d28 !1831 README新增全参微调长序列性能和对应脚本
Merge pull request !1831 from shenjiarun/master
2024-10-31 09:20:06 +00:00
qyz
44715d4fec !1798 Aquila2权重转换mg-hf切换新框架
Merge pull request !1798 from qyz/master
2024-10-31 06:36:48 +00:00
AresLzk
3e9fd1d561 !1827 新增InternLM2.5系列模型适配ModelLink-mcore
Merge pull request !1827 from AresLzk/master
2024-10-31 01:32:22 +00:00
njupt_sjj
ee59b5208d !1828 llama3.1模型mcore适配
Merge pull request !1828 from njupt_sjj/master
2024-10-30 11:02:56 +00:00
ningbenzhe1
9711128526 !1742 预训练在general mask场景支持CP,微调场景在pack场景支持CP
Merge pull request !1742 from ningbenzhe1/master
2024-10-29 12:03:04 +00:00
njupt_sjj
85345147e1 !1821 【mcore】llama3模型微调适配
Merge pull request !1821 from njupt_sjj/master
2024-10-29 11:10:42 +00:00
shenjiarun
f79c5dcabe !1817 整改两处公网地址
Merge pull request !1817 from shenjiarun/master
2024-10-29 11:05:59 +00:00
徐源徽
de4d0664b8 !1759 新增Qwen2-7B/Qwen1.5-4B mcore全参微调脚本
Merge pull request !1759 from 徐源徽/master
2024-10-26 06:55:21 +00:00
caoruichao
45ca70b2aa !1793 添加Qwen2.5-1.5B模型
Merge pull request !1793 from caoruichao/master
2024-10-26 06:38:36 +00:00
闻江
bf92daf093 !1813 新增MiniCPM-2B微调
Merge pull request !1813 from 闻江/master
2024-10-26 01:29:16 +00:00
DONGHAORAN
fd0903e8be !1774 lora适配moe模型
Merge pull request !1774 from DONGHAORAN/master
2024-10-25 06:28:18 +00:00
guoxinjie
670f729060 !1760 整理主线分支 README
Merge pull request !1760 from guoxinjie/arrange_readme
2024-10-25 01:33:31 +00:00
linqihong
1ea3922682 !1811 新增奖励模型llama2-7b训练脚本
Merge pull request !1811 from linqihong/llama2-rm
2024-10-24 06:14:05 +00:00
shengjy
2757579530 !1809 新增codellama lora推理
Merge pull request !1809 from shengjy/master
2024-10-23 07:42:57 +00:00
RuanZhiXiang
e93769afd2 !1511 refactor: support Deepseek Specification
Merge pull request !1511 from RuanZhiXiang/refactor-deepseek
2024-10-21 07:57:37 +00:00
shengjy
6e515bf03f !1795 新增codellama微调
Merge pull request !1795 from shengjy/master
2024-10-21 02:47:31 +00:00
曲玥泽
3afe8cde43 !1768 codellama-34b、llama-7b、yi-34b权重转换适配新框架
Merge pull request !1768 from 曲玥泽/master
2024-10-17 03:28:34 +00:00