apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 16:48:03 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

Revert "【FIX】Change the name of sparse attn from moba to plas (#3845)" (#4001)
Some checks failed
CE Compile Job / ce_job_pre_check (push) Has been cancelled
Details
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
Details
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
Details
CE Compile Job / Show Code Archive Output (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
Details
CE Compile Job / CE_UPLOAD (push) Has been cancelled
Details
Deploy GitHub Pages / deploy (push) Has been cancelled
Details

Browse Source 
This reverts commit e31c8f7336.
This commit is contained in:
Jiang-Jia-Jun
2025-09-09 11:08:23 +08:00
committed by GitHub
parent bbd548ceb6
commit c60adf4281
13 changed files with 150 additions and 150 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
fastdeploy/model_executor/layers/attention/attention.py
View File

@@ -119,19 +119,19 @@ class Attention(nn.Layer):
self.init_weight()
if (
fd_config.plas_attention_config is not None
and fd_config.plas_attention_config.plas_encoder_top_k_left is not None
and fd_config.plas_attention_config.plas_encoder_top_k_right is not None
and fd_config.plas_attention_config.plas_decoder_top_k_left is not None
and fd_config.plas_attention_config.plas_decoder_top_k_right is not None
fd_config.moba_attention_config is not None
and fd_config.moba_attention_config.moba_encoder_top_k_left is not None
and fd_config.moba_attention_config.moba_encoder_top_k_right is not None
and fd_config.moba_attention_config.moba_decoder_top_k_left is not None
and fd_config.moba_attention_config.moba_decoder_top_k_right is not None
):
mlp_weight_path = os.path.join(
fd_config.model_config.model, fd_config.plas_attention_config.mlp_weight_name
fd_config.model_config.model, fd_config.moba_attention_config.mlp_weight_name
)
self.plas_use_mlp = mlp_weight_path is not None and os.path.exists(mlp_weight_path)
plas_block_size = fd_config.plas_attention_config.plas_block_size
plas_max_seq_length = fd_config.plas_attention_config.plas_max_seq_length
if self.plas_use_mlp:
self.moba_use_mlp = mlp_weight_path is not None and os.path.exists(mlp_weight_path)
moba_block_size = fd_config.moba_attention_config.moba_block_size
moba_max_seq_length = fd_config.moba_attention_config.moba_max_seq_length
if self.moba_use_mlp:
mlp_weight = {}
with safe_open(mlp_weight_path, framework="np", device="cpu") as f:
for key_name in f.keys():
@@ -148,12 +148,12 @@ class Attention(nn.Layer):
* self.kv_num_heads : (fd_config.parallel_config.tensor_parallel_rank + 1)
* self.kv_num_heads
]
assert self.attn_gate_weight.shape[1] % plas_block_size == 0
assert self.attn_gate_weight.shape[1] % moba_block_size == 0
self.cache_k_block_means = paddle.zeros(
[
fd_config.parallel_config.max_num_seqs,
plas_max_seq_length // plas_block_size,
moba_max_seq_length // moba_block_size,
self.kv_num_heads,
self.head_dim,
],