github Fangyuan025/Chaty v0.2.0
Chaty v0.2.0 — GPU acceleration

latest releases: v2.2.1, v2.2.0, v2.1.9...
3 months ago

Chaty v0.2.0 — GPU acceleration + hardware panel.

The headline of this release is GPU offload for the LLM, so larger models run much faster.

New

  • GPU acceleration (Vulkan, auto‑tuned). Cross‑vendor — works on NVIDIA / AMD / Intel. Chaty detects your VRAM and automatically offloads as many transformer layers as fit, backing off gracefully if a GPU allocation fails and falling back to CPU when there's no usable GPU. No configuration needed.
  • Hardware panel (top‑right). Shows your CPU, RAM, GPU + VRAM, the active GPU backend, and the current model's offload (e.g. “27/33 layers on GPU”).
  • GPU setting — Settings → GPU acceleration: Auto / Off (CPU) / a manual layer count.

Notes

  • GPU uses Vulkan; the binary only needs the system vulkan-1.dll that ships with your graphics driver — nothing extra to install.
  • With no usable GPU, it runs on CPU exactly as before.
  • Voice remains English‑only and CPU‑based (independent of the LLM's VRAM).

Install (Windows x64)

Download Chaty_0.2.0_x64-setup.exe below and run it (per‑user, no admin). Voice models (~0.5 GB) still download on first voice use.

Don't miss a new Chaty release

NewReleases is sending notifications on new releases.