fix : 使用 bool 注意力掩码并支持打包 SFT 文档边界阻断

- 简化 process_attention_mask,通过广播返回 bool 掩码
- 新增 make_doc_boundary_mask 生成块对角因果掩码
- SFT strategy 传入文档边界掩码
This commit is contained in:
2026-06-05 17:02:28 +08:00
parent dc7d2cfbca
commit acd1103bd0
2 changed files with 26 additions and 10 deletions
+6 -9
View File
@@ -26,24 +26,21 @@ def process_attention_mask(
return input_mask
device = input_tensor.device
dtype = input_tensor.dtype
B, S = input_tensor.size()[:2]
B = input_tensor.size(0)
T = position_ids.max().item() + 1
if input_mask is None:
if position_ids.min().item() == 0 and is_causal:
return None
pad = torch.ones(B, T, dtype=torch.bool, device=device)
attend = torch.ones(B, 1, T, dtype=torch.bool, device=device)
else:
pad = input_mask[:, :T].to(device=device, dtype=torch.bool)
attend = input_mask[:, :T].to(device=device, dtype=torch.bool).unsqueeze(1)
attend = pad.view(B, 1, T).expand(B, S, T).clone()
if is_causal:
attend &= position_ids.unsqueeze(-1) >= torch.arange(T, device=device)
causal = position_ids.unsqueeze(-1) >= torch.arange(T, device=device)
attend = attend & causal
return torch.full(
(B, 1, S, T), -torch.finfo(dtype).max / 2, dtype=dtype, device=device
).masked_fill_(attend.unsqueeze(1), 0.0)
return attend.unsqueeze(1)
@AutoModel.register("autoregressive_lm")