Vram semaphore infra (#44465)

Signed-off-by: Brandon Pelfrey <bpelfrey@nvidia.com>
Co-authored-by: Roger Wang <hey@rogerw.io>
This commit is contained in:
Brandon Pelfrey
2026-06-26 17:32:51 -07:00
committed by GitHub
co-authored by Roger Wang
parent 3f67477497
commit af16446bf3
11 changed files with 1087 additions and 15 deletions
+6
View File
@@ -582,6 +582,7 @@ class EngineArgs:
skip_mm_profiling: bool = MultiModalConfig.skip_mm_profiling
video_pruning_rate: float | None = MultiModalConfig.video_pruning_rate
mm_tensor_ipc: MMTensorIPC = MultiModalConfig.mm_tensor_ipc
mm_ipc_gpu_memory_gb: float = MultiModalConfig.mm_ipc_gpu_memory_gb
# LoRA fields
enable_lora: bool = False
max_loras: int = LoRAConfig.max_loras
@@ -1294,6 +1295,10 @@ class EngineArgs:
multimodal_group.add_argument(
"--mm-tensor-ipc", **multimodal_kwargs["mm_tensor_ipc"]
)
multimodal_group.add_argument(
"--mm-ipc-gpu-memory-gb",
**multimodal_kwargs["mm_ipc_gpu_memory_gb"],
)
# LoRA related configs
lora_kwargs = get_kwargs(LoRAConfig)
@@ -1660,6 +1665,7 @@ class EngineArgs:
logits_processors=self.logits_processors,
video_pruning_rate=self.video_pruning_rate,
mm_tensor_ipc=self.mm_tensor_ipc,
mm_ipc_gpu_memory_gb=self.mm_ipc_gpu_memory_gb,
io_processor_plugin=self.io_processor_plugin,
renderer_num_workers=self.renderer_num_workers,
)