Skip to content

Atomic Chat v2.0.5

Choose a tag to compare

View all tags
@github-actions github-actions released this 31 Jul 19:53
· 69 commits to main since this release
v2.0.5
e57b410

🚀 New Features

More GPUs, real acceleration — ROCm on Windows, CUDA on Linux

  • AMD GPUs now run on ROCm on Windows, with VRAM correctly detected (#195)

  • Linux now gets CUDA builds, not just Vulkan (#198)

  • Atomic Chat now picks the right GPU backend on its own and remembers it, so you get GPU speed instead of silently generating on CPU (#186)

  • llama.cpp and TurboQuant updated to the latest builds

  • Added turbo4 support to llama.cpp

  • Remote providers now work on the first request

  • The model catalog loads faster and no longer runs out of browser storage

🔧 Improvements & Fixes

  • Long generations no longer stop after 3 minutes — the limit is now 10 (#212)
  • Web search shows all results, not just the first (#171)
  • Models the Hub wrongly called "Heavy for your device" now show their Download button (#196)
  • Sorting on the Models tab now applies to Recommended too (#192)
  • You can drag the window from the Models tab again on Windows (#193)
  • Text-only MLX models no longer fail to load as vision models (#170)
  • MLX downloads no longer freeze at 0% (#184)
  • The Linux AppImage updater no longer leaves the app unable to launch (#164)
  • Tool call previews no longer render whole files as one line (#188)
  • Fixed a startup crash on Windows
  • Crashes are now written to app.log even with telemetry off, so you can attach it to a bug report
  • Clearer provider status in the sidebar and model dropdown
  • 10+ smaller stability and layout fixes

🙏 Contributors

Thanks to @Vect0rM, @dtorey-d, @cashcon57, @konytskyy, @aryapratham1601-sketch, @Eiion, @leleobhz, @linuxgamer, @dsstutts, @StephenLReed and @DjagbleyEmmanuel for their contributions to this release!

Assets 11
Loading