[Qwen3.8-Flash-Next] Remove torch.compile for NVIDIA implementation (#55272)

Signed-off-by: Thien Tran <gau.nernst@yahoo.com.sg>
This commit is contained in:
Thien Tran
2026-09-06 21:57:56 -07:00
committed by GitHub
parent 9cc7793e32
commit d9105ea800
13 changed files with 165 additions and 185 deletions
+7 -1
View File
@@ -338,9 +338,15 @@ def test_dsa_models_default_to_mrv2_and_breakable_cudagraph(
("DeepseekV32MTPModel", True, False),
("GlmMoeDsaForCausalLM", False, True),
("GlmMoeDsaForCausalLM", True, False),
("Qwen4ExpForCausalLM", False, True),
("Qwen4ExpForCausalLM", True, False),
("Qwen4ExpForConditionalGeneration", False, True),
("Qwen4ExpForConditionalGeneration", True, False),
("Qwen4ExpMTP", False, True),
("Qwen4ExpMTP", True, False),
],
)
def test_dsa_breakable_cudagraph_platform_default(
def test_breakable_cudagraph_platform_default(
monkeypatch, architecture, is_rocm, expected
):
from vllm.config.vllm import default_breakable_cudagraph_architectures