biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Feature] Support sequence parallelism for static fp8 quantization (#19181)

Browse Source 
Signed-off-by: cascade812 <cascade812@outlook.com>
This commit is contained in:
cascade
2025-06-23 13:09:02 -07:00
committed by GitHub
parent d0132f025d
commit e6327c9b3e
7 changed files with 531 additions and 195 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/compilation/pass_manager.py
View File

@@ -51,15 +51,15 @@ class PostGradPassManager(CustomGraphPass):
if self.pass_config.enable_noop:
self.passes += [NoOpEliminationPass(config)]
if self.pass_config.enable_fusion:
self.passes += [FusionPass.instance(config)]
self.passes += [ActivationQuantFusionPass(config)]
if self.pass_config.enable_sequence_parallelism:
self.passes += [SequenceParallelismPass(config)]
if self.pass_config.enable_async_tp:
self.passes += [AsyncTPPass(config)]
if self.pass_config.enable_fusion:
self.passes += [FusionPass.instance(config)]
self.passes += [ActivationQuantFusionPass(config)]
if self.pass_config.enable_attn_fusion:
self.passes += [AttnFusionPass(config)]