Skip to content

GuideLLM v0.6.0

Latest

Choose a tag to compare

@dbutenhof dbutenhof released this 01 Apr 21:44
· 38 commits to main since this release
Immutable release. Only release title and notes can be modified.
a963ae9

Overview

GuideLLM v0.6.0 is a feature release adding multi-turn, Responses API, GeoSpatial model support, and in-process vLLM Python backend along with bug fixes.

To get started, install with:

pip install guidellm[recommended]==0.6.0

Or from source with:

pip install 'guidellm[recommended] @ git+https://github.com/vllm-project/guidellm.git'@v0.6.0

Compatibility Notes

  • Python: 3.10–3.13
  • OS: Linux, MacOS

What's New:

  • Added basic Responses API support: tool calling support will be added later
  • Added multi-turn support for both datasets and synthetic data
  • Added vLLM Python (in-process) backend
  • Added TerraTorch GeoSpacial model support

What's Fixed:

  • Allow disabling vLLM-specific body options in HTTP backend
  • Fix --sample-requests to limit sampling in output
  • Fix HTML references in html report
  • Fixed container image HOME permissions for OpenShift

Change Log

Features

Internal refactoring & cleanup

Fixes

  • fix(cli): validate --output-path against --output-dir by @aiwantaozi in #561
  • Fix the guidellm benchmark --sample-requests command line option by @natoscott in #591
  • Drop various depricated settings and remove the default OpenAI request timeout by @sjmonson in #589
  • Fix /v1/chat/completions formatting by @sjmonson in #595
  • Containerfile: ensure that HOME can be used by any user ID by @kpouget in #601
  • Fix JSON serialization for binary request payloads via base64 bytes config by @ushaket in #612
  • Move html template source location to raw github by @sjmonson in #629
  • Fix file extension not being sent to output handler by @jaredoconnell in #639
  • Check if deserialization path is vaild safely by @sjmonson in #659
  • Support removing keys from HTTP request bodies by @sjmonson in #661
  • Replace line iter with bytes to lines wrapper by @sjmonson in #663
  • Revert back to iterating over lines by @sjmonson in #680

CI environment

Documentation

  • Add multimodal benchmarking usage docs by @markurtz in #568
  • Add data parameter to benchmark command in README by @S1ro1 in #616
  • Add detail in benchmark profile documentation by @dbutenhof in #619
  • docs: add documentation for passing sampling parameters via --backend-kwargs by @cemigo114 in #626
  • docs: Fixing a broken link of docs/guides/outputs.md. by @theodor2311 in #642

Dependency updates

New Contributors

Full Changelog: v0.5.3...v0.6.0