apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-04 16:22:57 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

remove cum_offsets from ForwardMeta (#2925)
Some checks failed
Deploy GitHub Pages / deploy (push) Has been cancelled
Details

Browse Source
This commit is contained in:
周周周
2025-07-19 23:57:27 +08:00
committed by GitHub
parent 25698d56d1
commit 8c5407d9e4
3 changed files with 2 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
fastdeploy/model_executor/forward_meta.py
View File

@@ -85,8 +85,6 @@ class ForwardMeta:
# The sequence length processed in the current step # The sequence length processed in the current step
seq_lens_this_time: Optional[paddle.Tensor] = None seq_lens_this_time: Optional[paddle.Tensor] = None
# Accumulated offset
cum_offsets: Optional[paddle.Tensor] = None
# batch_id_per_token tensor, used to indicate which token belongs which batch after padding removal to the original input_ids # batch_id_per_token tensor, used to indicate which token belongs which batch after padding removal to the original input_ids
batch_id_per_token: Optional[paddle.Tensor] = None batch_id_per_token: Optional[paddle.Tensor] = None
# Accumulated sequence length of query # Accumulated sequence length of query
@@ -112,7 +110,8 @@ class XPUForwardMeta(ForwardMeta):
""" """
XPUForwardMeta is used to store the global meta information of the forward, and some XPU specific meta info. XPUForwardMeta is used to store the global meta information of the forward, and some XPU specific meta info.
""" """
# Accumulated offset
cum_offsets: Optional[paddle.Tensor] = None
# TODO(wanghaitao): Supplementary notes # TODO(wanghaitao): Supplementary notes
# #
encoder_batch_map: Optional[paddle.Tensor] = None encoder_batch_map: Optional[paddle.Tensor] = None

1
fastdeploy/spec_decode/mtp.py
View File

@@ -397,7 +397,6 @@ class MTPProposer(Proposer):
seq_lens_encoder=self.model_inputs["seq_lens_encoder"], seq_lens_encoder=self.model_inputs["seq_lens_encoder"],
seq_lens_decoder=self.model_inputs["seq_lens_decoder"], seq_lens_decoder=self.model_inputs["seq_lens_decoder"],
seq_lens_this_time=self.model_inputs["seq_lens_this_time"], seq_lens_this_time=self.model_inputs["seq_lens_this_time"],
cum_offsets=self.model_inputs["cum_offsets"],
batch_id_per_token=self.model_inputs["batch_id_per_token"], batch_id_per_token=self.model_inputs["batch_id_per_token"],
cu_seqlens_q=self.model_inputs["cu_seqlens_q"], cu_seqlens_q=self.model_inputs["cu_seqlens_q"],
cu_seqlens_k=self.model_inputs["cu_seqlens_k"], cu_seqlens_k=self.model_inputs["cu_seqlens_k"],

1
fastdeploy/worker/gpu_model_runner.py
View File

@@ -680,7 +680,6 @@ class GPUModelRunner(ModelRunnerBase):
seq_lens_encoder=self.share_inputs["seq_lens_encoder"], seq_lens_encoder=self.share_inputs["seq_lens_encoder"],
seq_lens_decoder=self.share_inputs["seq_lens_decoder"], seq_lens_decoder=self.share_inputs["seq_lens_decoder"],
seq_lens_this_time=self.share_inputs["seq_lens_this_time"], seq_lens_this_time=self.share_inputs["seq_lens_this_time"],
cum_offsets=self.share_inputs["cum_offsets"],
batch_id_per_token=self.share_inputs["batch_id_per_token"], batch_id_per_token=self.share_inputs["batch_id_per_token"],
cu_seqlens_q=self.share_inputs["cu_seqlens_q"], cu_seqlens_q=self.share_inputs["cu_seqlens_q"],
cu_seqlens_k=self.share_inputs["cu_seqlens_k"], cu_seqlens_k=self.share_inputs["cu_seqlens_k"],