biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[V1][Metrics] Remove metrics that were deprecated in 0.8 (#18837)

Browse Source 
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
This commit is contained in:
Mark McLoughlin
2025-05-28 19:54:12 +01:00
committed by GitHub
parent c68b5c63eb
commit 0e98964e94
6 changed files with 1 additions and 156 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
docs/usage/metrics.md
View File

@@ -35,19 +35,6 @@ The following metrics are exposed:
--8<-- "vllm/engine/metrics.py:metrics-definitions"
```
The following metrics are deprecated and due to be removed in a future version:
- `vllm:num_requests_swapped`, `vllm:cpu_cache_usage_perc`, and
`vllm:cpu_prefix_cache_hit_rate` because KV cache offloading is not
used in V1.
- `vllm:gpu_prefix_cache_hit_rate` is replaced by queries+hits
counters in V1.
- `vllm:time_in_queue_requests` because it duplicates
`vllm:request_queue_time_seconds`.
- `vllm:model_forward_time_milliseconds` and
`vllm:model_execute_time_milliseconds` because
prefill/decode/inference time metrics should be used instead.
Note: when metrics are deprecated in version `X.Y`, they are hidden in version `X.Y+1`
but can be re-enabled using the `--show-hidden-metrics-for-version=X.Y` escape hatch,
and are then removed in version `X.Y+2`.