minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Diffusion] Enable breakable CUDA graph for JoyEcho (#38110)

Browse Source
This commit is contained in:
Xiaoyu Zhang
2026-09-05 21:45:53 +08:00
committed by GitHub
parent 5df60a21cd
commit eda10c3678
2 changed files with 10 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/multimodal_gen/runtime/pipelines_core/stages/model_specific_stages/joy_echo/denoising.py
+3 -1
View File
@@ -515,7 +515,9 @@ class JoyEchoDMDDenoisingStage(LTX2AVDenoisingStage):
]
with self._ltx2_model_forward_context(ctx, step):
model_video, model_audio = step.current_model(**model_kwargs)
model_video, model_audio = self._ltx2_call_current_model(
ctx, step, model_kwargs
)
if memory_meta:
memory_video_len = memory_meta["memory_video_len"]
python/sglang/multimodal_gen/runtime/server_args/server_args.py
+7 -4
View File
@@ -180,6 +180,8 @@ BREAKABLE_CUDA_GRAPH_SUPPORTED_MODEL_IDS = frozenset(
"ideogram-v4-instant",
"ideogram-ai/ideogram-4-fp8",
"ideogram-ai/ideogram-4-nf4",
"jdopensource/joyai-echo",
"joyai-echo",
"lightricks/ltx-2",
"lightricks/ltx-2.3",
"meituan-longcat/longcat-image",
@@ -203,6 +205,7 @@ BREAKABLE_CUDA_GRAPH_SUPPORTED_PIPELINE_CONFIGS = frozenset(
{
"GlmImagePipelineConfig",
"Ideogram4PipelineConfig",
"JoyEchoPipelineConfig",
"LTX2PipelineConfig",
"LTX23PipelineConfig",
"LongCatImagePipelineConfig",
@@ -735,10 +738,10 @@ class ServerArgs(DisaggServerArgsMixin):
logger.warning(
"[Diffusion BCG] disabled for %s: only Ideogram-4, "
"Lightricks/LTX-2, LongCat-Image, MiniMax-H3, "
"Qwen/Qwen-Image, Qwen/Qwen-Image-2512, SANA1.5, SANA-Video, "
"Tongyi-MAI/Z-Image/Z-Image-Turbo, and zai-org/GLM-Image are "
"currently supported.",
"jdopensource/JoyAI-Echo, Lightricks/LTX-2, LongCat-Image, "
"MiniMax-H3, Qwen/Qwen-Image, Qwen/Qwen-Image-2512, SANA1.5, "
"SANA-Video, Tongyi-MAI/Z-Image/Z-Image-Turbo, and "
"zai-org/GLM-Image are currently supported.",
pipeline_config_name,
)
self.enable_breakable_cuda_graph = False