ViperEkura 1d54491809 refactor: 改用递归子模块 init 替代统一 normal_(0.006) ...

- Embedding.reset_parameters: normal_(std=0.02)
- Linear.reset_parameters: kaiming_uniform_ + uniform_ bias
- Transformer._init_weights 通过 apply 递归调用子模块 reset_parameters
- 移除全局 normal_(0.006) 覆盖，各模块使用更合适的分布

2026-05-17 10:44:18 +08:00

..

config

refactor: Config序列化统一BaseConfig基类

2026-05-16 22:06:39 +08:00

dataset

refactor: Storage 改用工厂模式，server reload 接入 uvicorn

2026-05-16 17:00:26 +08:00

inference

refactor: Storage 改用工厂模式，server reload 接入 uvicorn

2026-05-16 17:00:26 +08:00

model

refactor: 改用递归子模块 init 替代统一 normal_(0.006)

2026-05-17 10:44:18 +08:00

parallel

refactor: 移除 device_ids 参数设计，统一通过 CUDA_VISIBLE_DEVICES 控制 GPU 分配；更新 README 训练示例

2026-05-09 14:55:43 +08:00

tokenize

fix: 修复 CLI 参数缺失/重复、device_ids 越界、generate 参数名不一致、scheduler 时序、非流式截断等 bug

2026-05-09 14:36:42 +08:00

trainer

refactor: Config序列化统一BaseConfig基类

2026-05-16 22:06:39 +08:00

__init__.py

chore: 版本号升至 1.3.5

2026-05-15 18:23:27 +08:00

factory.py

refactor: 工厂 kwargs 过滤及组件参数清理

2026-05-16 16:47:41 +08:00

serialization.py

refactor: Config序列化统一BaseConfig基类

2026-05-16 22:06:39 +08:00