[Refactor] Cuda Graph Runner/Backend Refactor (#23906)

Co-authored-by: BBuf <1182563586@qq.com>
Co-authored-by: Cheng Wan <54331508+ch-wan@users.noreply.github.com>
Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com>
This commit is contained in:
Yuwei An
2026-06-09 21:36:57 -07:00
committed by GitHub
co-authored by BBuf Cheng Wan Lianmin Zheng
parent 56f06278c6
commit 2495c02c2c
160 changed files with 5212 additions and 3083 deletions
@@ -23,7 +23,7 @@ class TestExternalModels(CustomTestCase):
engine = sgl.Engine(
model_path=model_path,
cuda_graph_max_bs=1,
cuda_graph_max_bs_decode=1,
max_total_tokens=64,
enable_multimodal=True,
)
@@ -188,7 +188,7 @@ def init_process_dst(
model_path=model_name,
base_gpu_id=base_gpu_id,
tp_size=tp_size,
cuda_graph_max_bs=2,
cuda_graph_max_bs_decode=2,
tokenizer_path=model_name,
remote_instance_weight_loader_seed_instance_ip=seed_instance_ip,
remote_instance_weight_loader_seed_instance_service_port=seed_instance_service_port,
@@ -197,7 +197,7 @@ def init_process_dst(
model_path=model_name,
base_gpu_id=base_gpu_id,
tp_size=tp_size,
cuda_graph_max_bs=2,
cuda_graph_max_bs_decode=2,
tokenizer_path=model_name,
remote_instance_weight_loader_seed_instance_ip=seed_instance_ip,
remote_instance_weight_loader_seed_instance_service_port=seed_instance_service_port,
@@ -26,7 +26,7 @@ class TestPrefetchCheckpointsMultiGPU(CustomTestCase):
enable_dp_attention=True,
disable_radix_cache=True,
weight_loader_prefetch_checkpoints=True,
cuda_graph_max_bs=1,
cuda_graph_max_bs_decode=1,
max_total_tokens=256,
)
@@ -28,7 +28,7 @@ class TestRunaiModelLoader(CustomTestCase):
cls.engine = sgl.Engine(
model_path=TEST_GCS_MODEL,
load_format="runai_streamer",
cuda_graph_max_bs=1,
cuda_graph_max_bs_decode=1,
max_total_tokens=64,
)