[Qwen4Exp] support Sequence Parallelism for Qwen3.8-Flash-Next (#56322)

Signed-off-by: huanghaoyan.hhy <huanghaoyan.hhy@alibaba-inc.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Harry Huang
2026-10-12 20:56:44 +08:00
committed by GitHub
co-authored by mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 258b97ff6b
commit c75cb63cd2
23 changed files with 870 additions and 178 deletions
+16
View File
@@ -127,6 +127,22 @@ def test_registry_is_pp(model_arch, is_pp):
assert model_info.supports_pp is is_pp
@pytest.mark.parametrize(
"model_arch,supported",
[
("Qwen4ExpForCausalLM", not current_platform.is_rocm()),
("Qwen4ExpForConditionalGeneration", not current_platform.is_rocm()),
("Qwen4ExpMTP", not current_platform.is_rocm()),
("LlamaForCausalLM", False),
],
)
def test_registry_supports_hc_sp(model_arch, supported):
"""HC SP is opt-in for each model implementation, including MTP."""
model_info = ModelRegistry._try_inspect_model_cls(model_arch)
assert model_info is not None
assert model_info.supports_hc_sp is supported
@create_new_process_for_each_test()
@pytest.mark.parametrize(
"model_arch",