[ROCm][Qwen4Exp] Support HC and MoE sequence parallelism on ROCm (#61471)
Signed-off-by: Andreas Karatzas <akaratza@amd.com> Co-authored-by: Claude <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude
parent
a956a8126e
commit
29345f42a6
@@ -992,7 +992,7 @@ class Qwen4ExpForConditionalGenerationConfig(Qwen3_5ForConditionalGenerationConf
|
||||
parallel_config = vllm_config.parallel_config
|
||||
# MoE SP also enables HC SP without the explicit HC flag.
|
||||
if (
|
||||
current_platform.is_cuda()
|
||||
current_platform.is_cuda_alike()
|
||||
and parallel_config.use_sequence_parallel_moe
|
||||
and parallel_config.pipeline_parallel_size != 1
|
||||
):
|
||||
|
||||
Reference in New Issue
Block a user