 songsenand
|
7ac44a2731
|
refactor(trainer): 优化两阶段训练器代码结构和注释格式
|
2026-04-08 06:37:47 +08:00 |
|
songsenand
|
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
|
2026-04-07 14:46:50 +08:00 |
|
songsenand
|
3da8ae8876
|
feat(docs): 添加HTTP静态文件服务与远程监控说明
|
2026-04-06 22:53:15 +08:00 |
|
songsenand
|
2f0166c8ce
|
feat: 添加模型权重检查与推理调试工具脚本
|
2026-04-06 12:29:22 +08:00 |
|
songsenand
|
a203e67aff
|
fix(trainer): 修正总训练步数计算逻辑以支持多轮训练
|
2026-04-06 06:12:58 +08:00 |
|
songsenand
|
493bfdec1a
|
refactor(MoELayer): 并行化前向传播以兼容 torch.compile 和 AMP
|
2026-04-05 23:19:11 +08:00 |
|
songsenand
|
7143896f4d
|
fix: 启用 TensorFloat32 加速矩阵乘法并解决 UserWarning
|
2026-04-05 22:16:48 +08:00 |
|
songsenand
|
59bb29e4fd
|
feat(benchmark): 添加性能基准测试脚本用于诊断模型训练瓶颈
|
2026-04-05 22:06:36 +08:00 |
|
songsenand
|
c0489e538c
|
feat(docs): 添加基于JSON旁路记录法的移动端监控方案文档
|
2026-04-05 19:38:30 +08:00 |
|
songsenand
|
c31ec3990f
|
fix(trainer): 添加键盘中断处理以保存训练进度
|
2026-04-05 10:38:14 +08:00 |
|
songsenand
|
f838ec9b22
|
docs: 更新 README 中代码示例和训练说明
|
2026-04-05 10:29:23 +08:00 |
|
songsenand
|
310a926c98
|
refactor(trainer): 优化检查点保存逻辑以支持定期覆盖保存
|
2026-04-05 07:54:16 +08:00 |
|
songsenand
|
3e529d805f
|
fix(trainer): 调整模型保存频率以避免频繁写盘
|
2026-04-05 01:35:53 +08:00 |
|
songsenand
|
369424be28
|
feat(trainer): 添加GPU可用性日志并调整pinyin_ids处理逻辑
|
2026-04-05 00:58:51 +08:00 |
|
songsenand
|
b3055656d1
|
fix(trainer): 检测GPU可用性并提示用户退回CPU训练
|
2026-04-05 00:32:12 +08:00 |
|
songsenand
|
69349a88a6
|
feat(train): 添加训练脚本并重构模型输入处理逻辑
|
2026-04-05 00:08:29 +08:00 |