[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
+12
View File
@@ -49,6 +49,7 @@ from .interfaces import (
is_attention_free,
is_hybrid,
requires_raw_input_tokens,
supports_hc_sp,
supports_mamba_prefix_caching,
supports_multimodal,
supports_multimodal_encoder_tp_data,
@@ -873,6 +874,7 @@ class _ModelInfo:
requires_raw_input_tokens: bool
supports_multimodal_encoder_tp_data: bool
supports_pp: bool
supports_hc_sp: bool
has_inner_state: bool
is_attention_free: bool
is_hybrid: bool
@@ -903,6 +905,7 @@ class _ModelInfo:
model
),
supports_pp=supports_pp(model),
supports_hc_sp=supports_hc_sp(model),
has_inner_state=has_inner_state(model),
is_attention_free=is_attention_free(model),
is_hybrid=is_hybrid(model),
@@ -1463,6 +1466,15 @@ class _ModelRegistry:
model_info, _ = self.inspect_model_cls(architectures, model_config)
return model_info.supports_pp
def is_hc_sp_supported_model(
self,
architectures: str | list[str],
model_config: ModelConfig,
) -> bool:
"""Check support for HC SP without changing MoE parallelism."""
model_info, _ = self.inspect_model_cls(architectures, model_config)
return model_info.supports_hc_sp
def model_has_inner_state(
self,
architectures: str | list[str],