Releases: IST-DASLab/FP-Quant
Releases · IST-DASLab/FP-Quant
v0.3.2: Removed hadamard Randomness
Full Changelog: v0.3.1...v0.3.2
v0.3.1: Fixed compat with QuTLASS v0.2.0
Full Changelog: v0.3.0...v0.3.1
v0.3.0: MXFP8 and MXFP4 backward passes
What's Changed
- enable FP_Qaunt on xpu, validated by @yao-matrix in #11
- MXFP4 and MXFP8 backward passes by @BlackSamorez in #12
New Contributors
- @yao-matrix made their first contribution in #11
Full Changelog: v0.2.0...v0.3.0
v0.2.0: FP-Quant Pre-Print Release
Main additions
- NVFP4 support
- Various transform sizes (16-128)
- Various transform initializations
What's Changed
- Saving fix by @Godofnothing in #1
- NVFP4 Support by @BlackSamorez in #3
New Contributors
- @Godofnothing made their first contribution in #1
- @BlackSamorez made their first contribution in #3
Full Changelog: v0.1.6...v0.2.0
v0.1.6: better triton dispatch
Full Changelog: v0.1.5...v0.1.6
v0.1.5: pseudoquantization
Full Changelog: v0.1.4...v0.1.5
v0.1.4: no fast_hadamard_transform
Full Changelog: v0.1.3...v0.1.4
v0.1.3: Initial Release
Full Changelog: v0.1.2...v0.1.3
v0.1.2: cutlass>=0.0.1
Full Changelog: v0.1.1...v0.1.2
v0.1.1: Release kernel interfaces
Full Changelog: v0.1.0...v0.1.1