songsenand
  • Joined on 2024-01-17
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-09 12:37:52 +08:00
6ee28e0aa5 feat: 实现并行化 MoE 层以兼容 torch.compile 和 AMP
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-09 08:02:10 +08:00
e1efcc75a8 feat(base.html): 优化页面样式与结构,提升移动端兼容性
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-08 18:26:02 +08:00
d5daba182a docs: 将Streamlit替换为Flask以支持移动端监控界面
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-08 06:38:21 +08:00
7ac44a2731 refactor(trainer): 优化两阶段训练器代码结构和注释格式
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-08 00:21:28 +08:00
5dda0e6f85 feat(BigExpert): 添加 torch.compile 支持并优化编译参数
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-07 23:09:40 +08:00
813dce2224 删除 uv.lock 文件以清理依赖锁定信息
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-07 14:47:41 +08:00
c9a96651cd feat: 添加模型扩容两阶段训练功能,支持冻结层训练与全量微调切换
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-07 08:01:26 +08:00
d14fd09f41 feat(reverse_proxy): 添加 Apache 和 Nginx 反向代理配置支持 WebSocket 和 CORS
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-06 22:53:38 +08:00
3da8ae8876 feat(docs): 添加HTTP静态文件服务与远程监控说明
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-06 12:29:33 +08:00
2f0166c8ce feat: 添加模型权重检查与推理调试工具脚本
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-06 06:13:09 +08:00
a203e67aff fix(trainer): 修正总训练步数计算逻辑以支持多轮训练
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-06 06:06:12 +08:00
6ad003133c fix(dataset): 添加异常捕获防止标签生成失败
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-05 23:19:21 +08:00
493bfdec1a refactor(MoELayer): 并行化前向传播以兼容 torch.compile 和 AMP
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-05 22:17:00 +08:00
7143896f4d fix: 启用 TensorFloat32 加速矩阵乘法并解决 UserWarning
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-05 22:06:49 +08:00
59bb29e4fd feat(benchmark): 添加性能基准测试脚本用于诊断模型训练瓶颈
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-05 19:38:41 +08:00
c0489e538c feat(docs): 添加基于JSON旁路记录法的移动端监控方案文档
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-05 10:38:25 +08:00
c31ec3990f fix(trainer): 添加键盘中断处理以保存训练进度
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-05 10:30:03 +08:00
f838ec9b22 docs: 更新 README 中代码示例和训练说明
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-05 08:08:31 +08:00
1e9f1e8bd6 fix(dependency): 降级 torch 版本要求至 2.10.0
songsenand pushed to main at songsenand/SUimeModelTraner 2026-04-05 07:54:36 +08:00
310a926c98 refactor(trainer): 优化检查点保存逻辑以支持定期覆盖保存