A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[KV offload][1b/N] rename offloading to kv_offload (#25191)
Signed-off-by: Or Ozeri <oro@il.ibm.com>
Or Ozeri committed
Sep 18, 2025 at 23:53 UTC
a53ad626d629e79264f0a6ab6820a4b547f3b1c4
5 files changed
+5
-5
.buildkite/test-pipeline.yaml
+1
-1
@@ -280,7 +280,7 @@ steps:
280
# split the test to avoid interference
281
- pytest -v -s v1/core
282
- pytest -v -s v1/executor
283
- - pytest -v -s v1/offloading
283
+ - pytest -v -s v1/kv_offload
284
- pytest -v -s v1/sample
285
- pytest -v -s v1/logits_processors
286
- pytest -v -s v1/worker
tests/v1/kv_offload/test_worker.py
renamed
+2
-2
@@ -1,7 +1,7 @@
1
# SPDX-License-Identifier: Apache-2.0
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
-from vllm.v1.offloading.abstract import LoadStoreSpec
4
-from vllm.v1.offloading.worker.worker import (OffloadingHandler,
3
+from vllm.v1.kv_offload.abstract import LoadStoreSpec
4
+from vllm.v1.kv_offload.worker.worker import (OffloadingHandler,
5
OffloadingWorker, TransferResult,
6
TransferSpec)
7
vllm/v1/kv_offload/abstract.py
renamed
vllm/v1/kv_offload/mediums.py
renamed
+1
-1
@@ -4,7 +4,7 @@ from abc import ABC
4
5
import numpy as np
6
7
-from vllm.v1.offloading.abstract import LoadStoreSpec
7
+from vllm.v1.kv_offload.abstract import LoadStoreSpec
8
9
10
class BlockIDsLoadStoreSpec(LoadStoreSpec, ABC):
vllm/v1/kv_offload/worker/worker.py
renamed
+1
-1
@@ -3,7 +3,7 @@
3
from abc import ABC, abstractmethod
4
5
from vllm.logger import init_logger
6
-from vllm.v1.offloading.abstract import LoadStoreSpec
6
+from vllm.v1.kv_offload.abstract import LoadStoreSpec
7
8
# a single transfer spec (src_blocks_spec, dst_blocks_spec)
9
TransferSpec = tuple[LoadStoreSpec, LoadStoreSpec]