[Mini-VLLM] Remove orphan subsystems: snapshot, CPU workers, dead decoders
- Delete the snapshot package, CLI command, and its launch tests - Delete CPU workers, cpu_communicator, and the hw_agnostic layer package - Delete the EC CPU connector, keeping the EC base glue intact - Delete dead composite/triton_flashinfer backends and third_party/flashmla - Delete dead video decoders, video-bound tests, and orphan utils - Remove the deep_gemm_mega_moe option and dead allreduce fusion pair Co-authored-by: Hinata <hinata@idihore.id> Signed-off-by: MOVZX <movzx@yahoo.com>
This commit is contained in:
@@ -124,7 +124,6 @@ MoEBackend = Literal[
|
||||
"triton",
|
||||
"batched_triton",
|
||||
"deep_gemm",
|
||||
"deep_gemm_mega_moe",
|
||||
"cutlass",
|
||||
"flashinfer_trtllm",
|
||||
"flashinfer_cutlass",
|
||||
@@ -163,7 +162,7 @@ PASSTHROUGH_ALL2ALL_BACKEND: Final = "passthrough"
|
||||
# "Native" here meaning that the backend does not use the fused MoE path.
|
||||
# Currently only relevant for the vLLM provided DeepGEMM MegaMoE.,
|
||||
# TODO: migrate to the fused MoE path ultimately.
|
||||
NATIVE_MEGA_MOE_BACKENDS = frozenset({"deep_gemm_mega_moe"})
|
||||
NATIVE_MEGA_MOE_BACKENDS: frozenset[str] = frozenset()
|
||||
|
||||
# Every megakernel backend. The FlashInfer variants run through the standard
|
||||
# fused-MoE model path; the quantization method binds the kernel.
|
||||
|
||||
Reference in New Issue
Block a user