apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-05-06 15:40:33 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[XPU] support XPU VL model inference (#4030)

Browse Source 
* [XPU] support XPU VL model inference

* fix image op import and device check

* rebase develop

* fix perf
This commit is contained in:
Lucas
2025-09-25 14:34:15 +08:00
committed by GitHub
parent e36eccfdad
commit 87179cb744
18 changed files with 1300 additions and 146 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/model_executor/layers/attention/xpu_attn_backend.py
+2
View File
@@ -183,5 +183,7 @@ class XPUAttentionBackend(AttentionBackend):
forward_meta.encoder_batch_map_cpu,
forward_meta.decoder_context_len_cpu,
forward_meta.decoder_batch_map_cpu,
forward_meta.pos_emb_type,
self.rope_3d,
)
return res
fastdeploy/model_executor/layers/backends/xpu/moe/fused_moe.py
+2 -2
View File
@@ -72,7 +72,7 @@ class XPUMoEMethod(UnquantizedFusedMoEMethod):
layer.top_k,
False, # moe group, used in deepseek
)
if layer.tp_size > 1:
if layer.reduce_results and layer.tp_size > 1:
from fastdeploy.distributed.communication import (
tensor_model_parallel_all_reduce,
)
@@ -252,7 +252,7 @@ class XPUWeightOnlyMoEMethod(QuantMethodBase):
layer.top_k,
False, # moe group, used in deepseek
)
if layer.tp_size > 1:
if layer.reduce_results and layer.tp_size > 1:
from fastdeploy.distributed.communication import (
tensor_model_parallel_all_reduce,
)