GitLab

[Core][Metrics][BugFix] Replace num_cached_tokens/num_external_computed_tokens with PrefillStats (#37460)

Related to `Counters can only be incremented by non-negative amounts`
error with the `vllm:prompt_tokens_by_source_total` metric.
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Co-authored-by: Or Ozeri <or@ozery.com>

Signed-off-by: Martin Hickey <martin.hickey@ie.ibm.com>
Co-authored-by: Or Ozeri <or@ozery.com>

Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>

[BugFix] --max-model-len=-1 causes over-limit requests to hang and starve the entire service (#39102)
Signed-off-by: triangle14 <y1019026570@gmail.com>
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: mgoin <mgoin64@gmail.com>

Signed-off-by: Yifan <yzong@redhat.com>

Signed-off-by: Itay Alroy <ialroy@nvidia.com>
Co-authored-by: Yongji Wu <wuyongji317@gmail.com>
Co-authored-by: Ron Tourgeman <rtourgeman@nvidia.com>

Signed-off-by: Sage Ahrac <sagiahrak@gmail.com>

[Perf] Compute maxsim in worker side, reducing redundant copies, 2.7% E2E throughput improvement (#36159)
Signed-off-by: yewentao256 <zhyanwentao@126.com>

Signed-off-by: majiayu000 <1835304752@qq.com>
Co-authored-by: mcelrath

Signed-off-by: Andy Xie <andy.xning@gmail.com>

Signed-off-by: Nick Hill <nickhill123@gmail.com>

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

Signed-off-by: ahao-anyscale <ahao@anyscale.com>
Signed-off-by: Aaron Hao <ahao@anyscale.com>
Signed-off-by: hao-aaron <ahao@anyscale.com>
Signed-off-by: Nick Hill <nickhill123@gmail.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Nick Hill <nickhill123@gmail.com>

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

Signed-off-by: ahao-anyscale <ahao@anyscale.com>
Signed-off-by: Aaron Hao <ahao@anyscale.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>

Signed-off-by: Nick Hill <nickhill123@gmail.com>

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

Signed-off-by: Tianchen Ding <dtcccc@linux.alibaba.com>

Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>

Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>

Signed-off-by: Andreas Karatzas <akaratza@amd.com>

Signed-off-by: Ryan Rock <ryan.rock@amd.com>

Signed-off-by: John Calderon <jcalderon@nvidia.com>
Co-authored-by: Benjamin Chislett <bchislett@nvidia.com>

Signed-off-by: Nick Hill <nhill@redhat.com>
Signed-off-by: njhill <nickhill123@gmail.com>

[Prefix Cache] Include lora_name in BlockStored event for deterministic KV-cache reconstruction (#27577)
Signed-off-by: Sage Ahrac <sagiahrak@gmail.com>
Co-authored-by: Sage <80211083+sagiahrac@users.noreply.github.com>

Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>

Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Nick Hill <nhill@redhat.com>
Co-authored-by: Nick Hill <nhill@redhat.com>

Signed-off-by: Divakar Verma <divakar.verma@amd.com>

Signed-off-by: Seiji Eicher <seiji@anyscale.com>

Signed-off-by: Nick Hill <nhill@redhat.com>

Signed-off-by: inkcherry <mingzhi.liu@amd.com>

Signed-off-by: Shivam <shivamprasad91@gmail.com>
Signed-off-by: shivampr <shivampr.dev@gmail.com>
Co-authored-by: Chen Zhang <zhangch99@outlook.com>

Signed-off-by: Or Ozeri <oro@il.ibm.com>

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

Currently, when requests are cancelled while executing their final
step, "completion" is handled based on normal stop processing (e.g.
length or stop token), so the abort has no effect. This is typically
not a problem, but when a kv connector is involved it thinks the
request completed successfully rather than being aborted.

This is problematic for disaggregated prefill which will free kv
cache blocks if the request was aborted but not if it completed
successfully—since the cancelled request will never be sent to
the decode side, kv cache blocks remain pinned until the fall-back
timeout expires. The problem is exacerbated when many requests
are cancelled and/or there are large prefills whose forward pass
takes a long time (since the window is bigger).

This PR fixes the problem by processing pending aborts
immediately prior to processing model output each step; we process
only aborts, not new requests, since it's preferable for latency to
process model outputs before new incoming requests.

Fixes #26400.
Signed-off-by: Nick Hill <nhill@redhat.com>

Signed-off-by: LuminolT <lumischen01@gmail.com>
Signed-off-by: Lumis Chen <lumischen01@gmail.com>
Co-authored-by: Russell Bryant <rbryant@redhat.com>