[V0 Deprecation] Remove unused swap_space parameter (#36216)

Signed-off-by: majiayu000 <1835304752@qq.com>
Co-authored-by: mcelrath
This commit is contained in:
lif
2026-03-07 22:09:55 +08:00
committed by GitHub
co-authored by mcelrath
parent ee8a29511f
commit 00b814ba5a
22 changed files with 19 additions and 79 deletions
+4 -4
View File
@@ -507,10 +507,10 @@ longer relevant in v1:
- `vllm:num_requests_swapped`
- `vllm:cpu_cache_usage_perc`
In this mode, when a request is preempted (e.g. to make room in KV
cache to complete other requests), we swap kv cache blocks out to CPU
memory. This is also known as "KV cache offloading" and is configured
with `--swap-space` and `--preemption-mode`.
In this mode, when a request was preempted (e.g. to make room in KV
cache to complete other requests), kv cache blocks were swapped out to
CPU memory. The `--swap-space` flag has been removed as this feature
is no longer used in V1.
Historically, [vLLM has long supported beam search](https://github.com/vllm-project/vllm/issues/6226). The
SequenceGroup encapsulated the idea of N Sequences which
+1 -1
View File
@@ -17,7 +17,7 @@ llm = Vllm(
model="microsoft/Orca-2-7b",
tensor_parallel_size=4,
max_new_tokens=100,
vllm_kwargs={"swap_space": 1, "gpu_memory_utilization": 0.5},
vllm_kwargs={"gpu_memory_utilization": 0.5},
)
```