docs: add API benchmark example

[zcxGGmu]

Summary

• Add an online API benchmarking example to doc/en/benchmark.md.
• Show a small EvalScope evalscope perf command against /v1/chat/completions.
• Recommend validating the server with direct curl requests first, then increasing benchmark parallelism gradually.

Refs #853
Refs #1269

Test Plan

• python documentation guard check for the new EvalScope benchmark guidance
• git diff --check HEAD~1..HEAD

[greptile-apps]

Greptile Summary

This PR adds an "Online API Benchmarking" section to doc/en/benchmark.md, giving users a practical walk-through for load-testing the KTransformers /v1/chat/completions endpoint with EvalScope.

• Introduces a two-step workflow: first validate the server with a direct curl request, then run evalscope perf starting at low concurrency and scale up.
• Includes an inline pip install evalscope hint and moves the optional --tokenizer-path flag outside the code block with a clear explanation, addressing feedback from prior review rounds.
• Adds guidance on using a single curl request to reproduce and isolate issues when a benchmark run causes a crash.

Confidence Score: 5/5

Pure documentation addition with no code changes; safe to merge.

The change is a single documentation section with no logic, configuration, or dependency changes. The curl and evalscope commands are well-formed, the optional tokenizer-path is clearly documented outside the runnable block, and the inline install hint is present. Prior reviewer concerns have been addressed in this revision.

Files Needing Attention: No files require special attention.

Important Files Changed

Filename	Overview
doc/en/benchmark.md	Adds a new "Online API Benchmarking" section with a curl verification step, an EvalScope perf command, and inline install guidance; previously flagged issues (optional tokenizer-path, missing install note) are addressed in this revision.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[KTransformers server running] --> B[Verify with direct curl request]
    B --> C{Server responds OK?}
    C -- No --> D[Debug server-side issue]
    C -- Yes --> E[Run evalscope perf --parallel 1 --number 4]
    E --> F{Benchmark stable?}
    F -- No --> G[Reproduce failing prompt with plain curl]
    G --> D
    F -- Yes --> H[Increase --parallel and --number gradually]
    H --> I[Collect throughput / latency metrics]

Loading

_{Reviews (2): Last reviewed commit: "[docs]: refine EvalScope benchmark examp..." | Re-trigger Greptile}

[zcxGGmu]

Thanks for the review — I pushed b556223e6c58 addressing the EvalScope benchmark feedback.

Changes made:

• Added an inline EvalScope install note before the benchmark command.
• Removed the required --tokenizer-path placeholder from the openqa example and documented when to add it optionally.

Validation:

• python content guard — passed, confirmed the install note, valid default command, and optional tokenizer guidance.
• git diff --check — passed.

[greptile-apps]

Want your agent to iterate on Greptile's feedback? Try greploops.