Commit Graph

Select branches

Hide Pull Requests

main

v1.2.2

v1.3.0

v1.3.1

v1.3.10

v1.3.11

v1.3.12

v1.3.2

v1.3.3

v1.3.4

v1.3.5

v1.3.6

v1.3.7

v1.3.8

v1.3.9

Mono Color

• 30ac07418c feat(train): 添加多轮对话训练支持 ViperEkura 2025-09-28 15:38:53 +08:00
• 1169cfad82 fix(trainer): 修复多轮对话中的因果注意力掩码计算逻辑等 ViperEkura 2025-09-28 15:15:19 +08:00
• 0b96b11a6e test(trainer): 增加训练中断与检查点恢复测试 ViperEkura 2025-09-28 14:38:23 +08:00
• 25ec56a1f5 fix(trainer): 修复训练器恢复检查点时的学习率初始化问题 ViperEkura 2025-09-28 14:38:02 +08:00
• c8a38743a4 fix(tests): 更新测试代码以验证优化器状态的保存与加载 ViperEkura 2025-09-28 14:00:38 +08:00
• f25a249291 feat(khaosz): 优化模型参数保存与加载逻辑 ViperEkura 2025-09-28 14:00:21 +08:00
• 4fcdc87c95 feat(trainer): 重构数据集与策略模块以支持字典形式的数据返回 ViperEkura 2025-09-27 14:11:27 +08:00
• 9fbc9481b5 refactor(core): 修改注意力掩码处理函数并重命名参数 ViperEkura 2025-09-27 13:37:10 +08:00
• 053f4a4dad feat( StrategyFactory): 添加 SFT 策略初始化参数并完善工厂方法调用 ViperEkura 2025-09-27 13:24:16 +08:00
• 676fdd59d7 feat(strategy): 重构mask构建逻辑并优化策略工厂参数传递 ViperEkura 2025-09-27 13:12:57 +08:00
• a4443765ee Initial commit ViperEkura 2025-09-27 12:02:22 +08:00