ViperEkura ef8783b7e3 fix: separate attn_mask and loss_mask in get_logprobs, compose causal masking in strategy ...

- add loss_mask parameter to get_logprobs to decouple attention from loss masking
- DPO/GRPO strategies compose key-padding + causal mask before model forward
- prevents prompt tokens from being masked out of attention and missing causal masking

2026-07-21 23:47:28 +08:00

..

__init__.py

chore : 升级 torch 2.11.0+cu128，移除自定义 Muon，修复 gloo device_id

2026-06-27 16:10:37 +08:00

metric_util.py

refactor : grad_norm 指标简化，clip_grad_norm 移至 executor

2026-06-30 14:59:43 +08:00

rollout.py

refactor: map instruction/input/output to chat roles

2026-07-20 13:55:25 +08:00

schedule.py

fix: scheduler division-by-zero, loss_mask bool

2026-07-03 22:04:55 +08:00

strategy.py

fix: separate attn_mask and loss_mask in get_logprobs, compose causal masking in strategy

2026-07-21 23:47:28 +08:00

train_callback.py

fix: init last_ckpt_step and last_log_flush_step from context.optimizer_step

2026-07-15 22:15:55 +08:00

train_context.py

refactor : align config field names with Hugging Face

2026-07-20 22:05:31 +08:00

trainer.py

feat: add --resume flag to decouple weight loading from training resumption

2026-07-16 14:23:23 +08:00