Skip to content

v2.9.6 - Low-VRAM 4-bit PersonaPlex for 8GB GPUs

Choose a tag to compare

View all tags
@lukejmorrison lukejmorrison released this 25 Feb 02:07
· 56 commits to main since this release
v2.9.6
2e9b85d

VoiceDNA v2.9.6

Low-VRAM PersonaPlex support for consumer GPUs (GTX 1070 Ti class).

Highlights

  • Added optional extra: personaplex-lowvram (bitsandbytes + 4-bit runtime stack)
  • PersonaPlex now auto-switches to 4-bit quantized model on sub-12GB VRAM
  • Added CLI --lowvram for voicedna speak and voicedna test-natural
  • Added explicit status: Detected 8 GB VRAM → loading 4-bit PersonaPlex (low-VRAM mode)
  • Omarchy daemon/installer now supports low-VRAM mode
  • clear-vram.sh now performs more aggressive CUDA cache cleanup
  • Added low-VRAM backend decision tests and docs updates

Validation

  • pytest: 11 passed
  • Simulated 8GB backend decision confirms 4-bit PersonaPlex path
Assets 2
Loading