minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

[AMD] Fix AMD Multimodal Test - skip nvfp4 tests (#23045)

Browse Source
This commit is contained in:
YC Yen-Ching Tseng
2026-04-17 09:02:39 -07:00
committed by GitHub
parent 2bac219d0c
commit 5d4e899477
1 changed files with 52 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/multimodal_gen/test/server/gpu_cases.py
+52 -47
View File
@@ -361,53 +361,58 @@ if not current_platform.is_hip():
T2V_sampling_params, T2V_sampling_params,
) )
) )
ONE_GPU_CASES_C = [ # Skip all ModelOpt tests on AMD: FP8 requires torch._scaled_mm (HIPBLAS_STATUS_NOT_SUPPORTED
_make_modelopt_ci_case( # on ROCm), NVFP4 requires flashinfer or sgl_kernel FP4 kernels (CUDA-only)
"flux1_modelopt_fp8_t2i", if current_platform.is_hip():
model_path=DEFAULT_FLUX_1_DEV_MODEL_NAME_FOR_TEST, ONE_GPU_CASES_C = []
modality="image", else:
sampling_params=MODELOPT_T2I_CI_sampling_params, ONE_GPU_CASES_C = [
extras=["--transformer-path", MODELOPT_FLUX1_FP8_TRANSFORMER], _make_modelopt_ci_case(
), "flux1_modelopt_fp8_t2i",
_make_modelopt_ci_case( model_path=DEFAULT_FLUX_1_DEV_MODEL_NAME_FOR_TEST,
"flux2_modelopt_fp8_t2i", modality="image",
model_path=DEFAULT_FLUX_2_DEV_MODEL_NAME_FOR_TEST, sampling_params=MODELOPT_T2I_CI_sampling_params,
modality="image", extras=["--transformer-path", MODELOPT_FLUX1_FP8_TRANSFORMER],
sampling_params=MODELOPT_T2I_CI_sampling_params, ),
extras=["--transformer-path", MODELOPT_FLUX2_FP8_TRANSFORMER], _make_modelopt_ci_case(
), "flux2_modelopt_fp8_t2i",
_make_modelopt_ci_case( model_path=DEFAULT_FLUX_2_DEV_MODEL_NAME_FOR_TEST,
"wan22_modelopt_fp8_t2v", modality="image",
model_path=DEFAULT_WAN_2_2_T2V_A14B_MODEL_NAME_FOR_TEST, sampling_params=MODELOPT_T2I_CI_sampling_params,
modality="video", extras=["--transformer-path", MODELOPT_FLUX2_FP8_TRANSFORMER],
sampling_params=MODELOPT_T2V_CI_sampling_params, ),
extras=["--transformer-path", MODELOPT_WAN22_FP8_TRANSFORMER], _make_modelopt_ci_case(
), "wan22_modelopt_fp8_t2v",
_make_modelopt_ci_case( model_path=DEFAULT_WAN_2_2_T2V_A14B_MODEL_NAME_FOR_TEST,
"flux1_modelopt_nvfp4_t2i", modality="video",
model_path=DEFAULT_FLUX_1_DEV_MODEL_NAME_FOR_TEST, sampling_params=MODELOPT_T2V_CI_sampling_params,
modality="image", extras=["--transformer-path", MODELOPT_WAN22_FP8_TRANSFORMER],
sampling_params=MODELOPT_T2I_CI_sampling_params, ),
extras=["--transformer-path", MODELOPT_FLUX1_NVFP4_TRANSFORMER], _make_modelopt_ci_case(
env_vars=MODELOPT_NVFP4_B200_ENV_VARS, "flux1_modelopt_nvfp4_t2i",
), model_path=DEFAULT_FLUX_1_DEV_MODEL_NAME_FOR_TEST,
_make_modelopt_ci_case( modality="image",
"flux2_modelopt_nvfp4_t2i", sampling_params=MODELOPT_T2I_CI_sampling_params,
model_path=DEFAULT_FLUX_2_DEV_MODEL_NAME_FOR_TEST, extras=["--transformer-path", MODELOPT_FLUX1_NVFP4_TRANSFORMER],
modality="image", env_vars=MODELOPT_NVFP4_B200_ENV_VARS,
sampling_params=MODELOPT_T2I_CI_sampling_params, ),
extras=["--transformer-weights-path", MODELOPT_FLUX2_NVFP4_WEIGHTS], _make_modelopt_ci_case(
env_vars=MODELOPT_NVFP4_B200_ENV_VARS, "flux2_modelopt_nvfp4_t2i",
), model_path=DEFAULT_FLUX_2_DEV_MODEL_NAME_FOR_TEST,
_make_modelopt_ci_case( modality="image",
"wan22_modelopt_nvfp4_t2v", sampling_params=MODELOPT_T2I_CI_sampling_params,
model_path=DEFAULT_WAN_2_2_T2V_A14B_MODEL_NAME_FOR_TEST, extras=["--transformer-weights-path", MODELOPT_FLUX2_NVFP4_WEIGHTS],
modality="video", env_vars=MODELOPT_NVFP4_B200_ENV_VARS,
sampling_params=MODELOPT_T2V_CI_sampling_params, ),
extras=["--transformer-path", MODELOPT_WAN22_NVFP4_TRANSFORMER], _make_modelopt_ci_case(
env_vars=MODELOPT_NVFP4_B200_ENV_VARS, "wan22_modelopt_nvfp4_t2v",
), model_path=DEFAULT_WAN_2_2_T2V_A14B_MODEL_NAME_FOR_TEST,
] modality="video",
sampling_params=MODELOPT_T2V_CI_sampling_params,
extras=["--transformer-path", MODELOPT_WAN22_NVFP4_TRANSFORMER],
env_vars=MODELOPT_NVFP4_B200_ENV_VARS,
),
]
TWO_GPU_CASES_A = [ TWO_GPU_CASES_A = [
DiffusionTestCase( DiffusionTestCase(