[Frontend] Consolidate scale out entrypoints (#44512)

Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
This commit is contained in:
wang.yuqi
2026-06-29 03:18:53 -07:00
committed by GitHub
parent eddfd4cf21
commit 3483240b7e
45 changed files with 422 additions and 289 deletions
+2 -1
View File
@@ -13,5 +13,6 @@ vLLM's examples are organized into the following categories:
- **[`rl/`](../../examples/rl)** – Reinforcement learning examples.
- **[`deployment/`](../../examples/deployment)** – Examples for deploying vLLM in production.
- **[`ray_serving/`](../../examples/ray_serving)** – Scalable serving using Ray.
- **[`disaggregated/`](../../examples/disaggregated)** – Examples for disaggregated serving (separate prefill and decode), including various kv cache connectors (LMCache, Mooncake, FlexKV, P2P NCCL) and failure recovery.
- **[`disaggregated/`](../../examples/disaggregated)** – Examples for Disaggregated P/D (Prefill/Decoding) inference, including various kv cache connectors (LMCache, Mooncake, FlexKV, P2P NCCL) and failure recovery.
- **[`scale_out/`](../../examples/scale_out)** – Examples for Token In <> Token Out API Server.
- **[`observability/`](../../examples/observability)** – Metrics, logging, tracing (OpenTelemetry), and dashboards (Grafana, Perses).
+2 -2
View File
@@ -119,9 +119,9 @@ For further details on profiling vLLM, please refer to [this page](../../contrib
- `/ping` - SageMaker health check
- `/invocations` - SageMaker-compatible endpoint (routes to the same inference functions as `/v1` endpoints)
## Disaggregated Everything
## Scale-Out APIs
### Tokens IN <> Tokens OUT
### Tokens IN <> Tokens OUT APIs
- `/inference/v1/generate` - Generate completions
- `/abort_requests` - Abort in-flight requests (only when `--tokens-only` is also set)