apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 08:37:06 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

Revert "[Bug fix] fix attention rank init (#2743)" (#2761)

Browse Source 
This reverts commit e8bbe7244b.
This commit is contained in:
RichardWooSJTU
2025-07-09 10:38:12 +08:00
committed by GitHub
parent f72c4de539
commit 6610aa29d0
4 changed files with 13 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
fastdeploy/model_executor/layers/attention/xpu_attn_backend.py
View File

@@ -91,7 +91,7 @@ class XPUAttentionBackend(AttentionBackend):
# self.use_speculate = self.speculate_method is not None
# self.speculate_max_draft_token_num = fd_config.parallel_config.speculate_max_draft_tokens
self.keep_pd_step_flag: bool = fd_config.speculative_config.model_type == "mtp"
self.rank: int = fd_config.parallel_config.expert_parallel_rank * fd_config.parallel_config.tensor_parallel_degree + fd_config.parallel_config.tensor_parallel_rank
self.rank: int = fd_config.parallel_config.tensor_parallel_rank
self.kv_num_heads: int = kv_num_heads
self.num_heads: int = num_heads