> For the complete documentation index, see [llms.txt](https://docs.akamas.io/akamas-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.akamas.io/akamas-docs/reference/optimization-packs/vllm-pack/vllm.md).

# vLLM

This page describes the Optimization Pack for the component type vLLM.

## Metrics

| Metric                                    | Unit                   | Description                                                |
| ----------------------------------------- | ---------------------- | ---------------------------------------------------------- |
| e2e\_request\_latency\_p95                | milliseconds           | 95th percentile of the end-to-end request latency          |
| e2e\_request\_latency\_avg                | milliseconds           | Average end-to-end request latency                         |
| request\_success\_rate                    | requests per second    | Rate of successfully completed requests                    |
| time\_to\_first\_token\_p95               | milliseconds           | 95th percentile of the time to first token (TTFT)          |
| time\_to\_first\_token\_avg               | milliseconds           | Average time to first token (TTFT)                         |
| prefill\_token\_throughput                | tokens                 | Prefill-phase token throughput                             |
| inter\_token\_latency\_p95                | milliseconds           | 95th percentile of the inter-token latency (ITL)           |
| inter\_token\_latency\_avg                | milliseconds           | Average inter-token latency (ITL)                          |
| decode\_token\_throughput                 | tokens                 | Decode-phase token throughput                              |
| num\_requests\_running                    | requests               | Number of requests currently being processed               |
| num\_requests\_waiting                    | requests               | Number of requests waiting in the queue                    |
| kv\_cache\_usage\_avg                     | percentage             | Average KV cache utilization                               |
| kv\_cache\_usage\_max                     | percentage             | Peak KV cache utilization                                  |
| preemption\_rate                          | preemptions per second | Rate of request preemptions                                |
| prefix\_cache\_hit\_rate                  | percentage             | Prefix cache hit rate                                      |
| input\_sequence\_length\_p95              | tokens                 | 95th percentile of the input (prompt) sequence length      |
| input\_sequence\_length\_avg              | tokens                 | Average input (prompt) sequence length                     |
| output\_sequence\_length\_p95             | tokens                 | 95th percentile of the output (generated) sequence length  |
| output\_sequence\_length\_avg             | tokens                 | Average output (generated) sequence length                 |
| fleet\_e2e\_latency\_percentile           | milliseconds           | Fleet-wide end-to-end request latency percentile           |
| fleet\_inter\_token\_latency\_percentile  | milliseconds           | Fleet-wide inter-token latency percentile                  |
| fleet\_request\_success\_rate             | requests per second    | Fleet-wide request success rate                            |
| fleet\_kv\_cache\_usage\_avg              | percentage             | Fleet-wide average KV cache utilization                    |
| inter\_token\_latency\_p95\_5m            | milliseconds           | 95th percentile inter-token latency over a 5-minute window |
| inter\_token\_latency\_p95\_max\_per\_pod | milliseconds           | Maximum per-pod 95th percentile inter-token latency        |
| inter\_token\_latency\_p95\_min\_per\_pod | milliseconds           | Minimum per-pod 95th percentile inter-token latency        |

## Parameters

| Parameter                 | Type    | Unit               | Default | Domain                | Restart | Description                                                                     |
| ------------------------- | ------- | ------------------ | ------- | --------------------- | ------- | ------------------------------------------------------------------------------- |
| gpu\_memory\_utilization  | real    | percentage         | `0.9`   | `0.0` → `1.0`         | yes     | Target fraction of GPU memory reserved for the model executor and KV cache      |
| max\_num\_seqs            | integer | sequences          | `128`   | `1` → `1000000000000` | yes     | Maximum number of sequences processed together in a single iteration (batch)    |
| max\_num\_batched\_tokens | integer | tokens             | `1024`  | `1` → `1000000000000` | yes     | Maximum number of tokens processed together in a single iteration (batch)       |
| max\_model\_len           | integer | tokens             | `8192`  | `1` → `1000000000000` | yes     | Maximum model context length, in tokens                                         |
| tensor\_parallel\_size    | integer | parallelism degree | `1`     | `1` → `1000000000000` | yes     | Number of GPUs across which each model instance is sharded (tensor parallelism) |


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.akamas.io/akamas-docs/reference/optimization-packs/vllm-pack/vllm.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.