-
Notifications
You must be signed in to change notification settings - Fork 66
Closed
Description
Path to v1.3.0
- cache-dit-generate command line tool @DefTruth feat: add cache-dit-generate cli tool #752
- Optimize VAE Parallel comm, use batched isend/irecv @DefTruth chore: use batched isend/irecv for vae-p #757 feat: tile batched p2p comm for vae-p #758 [2/N] reduce comm overhead for vae-p #763 reduce comm overhead for vae-p #762
- Hybrid CP(USP) + TP, e.g, SP2 + TP2 @DefTruth feat: support hybrid CP/SP + TP #765
- Support USP (hybrid ulysses and ring attention) @DefTruth feat: support ring attn p2p comm #754 feat: support USP -> Ulysses + Ring #755
- New models support: GLM-Image, FLUX.2-Klein, Helios, FireRed-Image-Edit, and more. @DefTruth feat: support tensor parallel for glm-image #794 feat: support cache for glm-image #787 feat: support 🚀flux2-klein series #717 feat: support 🔥FireRed-Image-Edit-1.0 #797 feat: support cache for Helios-14B #834 feat: support FireRed-Image-Edit-1.1 #854
- Support pass a quantize_config to
enable_cacheAPI @DefTruth feat: support load quantize config yaml #847 - FP8 Blockwise dynamic quantization support @DefTruth feat: support blockwise fp8 #822
- AMD GPU support feat: Add AMD GPU support #841
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels