vrr/kvcache-ai-ktransformers
2
0
Fork 0
mirror of https://github.com/kvcache-ai/ktransformers.git synced 2026-04-28 11:49:51 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

(kt-kernel): add numa_nodes parameter for explicit NUMA node mapping (#1891)

Browse source 
Add numa_nodes parameter to BaseMoEWrapper and all subclasses, allowing
users to explicitly specify which NUMA node IDs to use for subpool
mapping instead of always defaulting to sequential [0, 1, ..., N-1].

This enables running multiple KTransformers instances on different NUMA
nodes of the same machine, e.g. --kt-threadpool-count 1 --kt-numa-nodes 1
to bind to NUMA node 1. Previously this required external numactl
workarounds since subpool_numa_map was hardcoded to start from 0.
This commit is contained in:
ErvinXie 2026-03-31 10:27:50 +08:00 • committed by GitHub
parent bdf4bb76c5
commit 3903c9afcc
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
5 changed files with 34 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

3
kt-kernel/python/experts.py
View file

@ -65,6 +65,7 @@ class KTMoEWrapper:
cpu_save: bool = False,
max_deferred_experts_per_token: Optional[int] = None,
method: str = "AMXINT4",
numa_nodes: Optional[List[int]] = None,
):
"""
Factory method to create the appropriate backend implementation.
@ -85,6 +86,7 @@ class KTMoEWrapper:
chunked_prefill_size: Maximum prefill chunk size
cpu_save: Whether to save weights to CPU memory
max_deferred_experts_per_token: Number of experts per token to defer. Defaults to 0.
numa_nodes: Explicit list of NUMA node IDs for subpool mapping. If None, defaults to sequential.
method: Backend method ("AMXINT4", "AMXINT8", "RAWINT4", "FP8", "BF16", "LLAMAFILE", "MOE_INT4", "MOE_INT8")
Returns:
@ -117,6 +119,7 @@ class KTMoEWrapper:
cpu_save=cpu_save,
max_deferred_experts_per_token=max_deferred_experts_per_token,
method=method,
numa_nodes=numa_nodes,
)
# Forward static methods to the base class