release: 0.2.6

What's Changed

• Add hip support by @Disty0 in #330
• Switched linters, black -> ruff by @ishandeva in #334
• Add marlin int4 kernel by @dacorvo and @shcho1118 in #333
• fix: use reshape instead of view by @dacorvo in #338
• Support QLayerNorm without weights by @dacorvo in #341

New Contributors

• @ishandeva made their first contribution in #334
• @Disty0 made their first contribution in #330
• @shcho1118 made their first contribution in #333

Full Changelog: v0.2.5...v0.2.6

release: 0.2.5

New features

• Load and save models from the Hugging Face hub #263 by @sayakpaul
• Add support for float8 e4f3mnuz #310 (from #281) by @maktukmak
• Faster and less memory-intensive requantization #290 by @latentCall145
• Support torch.equal for QTensor #294 by @dacorvo
• Add Marlin Float8 kernel #296 (from #241) by @fxmarty
• Add Whisper for speech recognition example #298 (from #242) by @mattiadg
• Add ViT classification example #308 by @shovan777

Bug fixes

• Fix include patterns in quantize #271 by @kaibioinfo
• Enable non-strict loading of state dicts #295 by @BenjaminBossan
• Fix transformers forward error #303 by @dacorvo
• Fix missing call in transformers models #325 by @dacorvo
• Fix 8-bit mm calls for 4D inputs #326 by @dacorvo

Full Changelog: v0.2.4...v0.2.5

release: 0.2.4

Bug Fixes:

• fix import error in optimum-cli when diffusers is not installed by @dacorvo

Full Changelog: v0.2.3...v0.2.4

release: 0.2.3

What's Changed

• Use new int8 torch kernels by @dacorvo in #222
• Rebuild extension when pytorch is updated by @dacorvo in #223
• Use tinygemm bfloat16 / int4 kernel whenever possible by @dacorvo in #234
• Add HQQ optimizer by @dacorvo in #235
• Add QuantizedModelForCausalLM by @dacorvo in #243
• Integrate quanto commands to optimum-cli by @dacorvo in #244
• Add pixart-sigma test to image example by @dacorvo in #247
• Support diffusion models. by @sayakpaul in #255

Bug fixes

• Fix: align extension on max arch by @dacorvo in #227
• Fix TinyGemmQBitsTensor move by @dacorvo in #246
• Fix stream-lining bug by @dacorvo in #249
• Fix float/int8 matrix multiplication latency regression by @dacorvo in #250
• Fix serialization issues by @dacorvo in #258

New Contributors

• @sayakpaul made their first contribution in #255

Full Changelog: v0.2.2...v0.2.3

release: 0.2.2

New features

• add OWLv2 detection example by @dacorvo,
• use new torch quantization kernels by @dacorvo.

Bug fixes

• avoid CUDA compilation errors on older Nvidia cards (pre Ampere) by @dacorvo,
• recompile extensions when pytorch is updated and prevent segfault by @dacorvo.

release: 0.2.1

This release does not contain any new feature, but it is the first one with the new package name.

release: 0.2.0

New features

• requantize helper by @calmitchell617,
• StableDiffusion example by @thliang01,
• improved linear backward path by @dacorvo ,
• AWQ int4 kernels by @dacorvo .

release: v0.1.0

New features

• group-wise quantization,
• safe serialization.

release(0.0.13):

New features

• new QConv2d quantized module,
• official support for float8 weights.

Bug fixes

• fix QbitsTensor.to() that was not moving the inner tensors,
• prevent shallow QTensor copies when loading weights that do not move inner tensors.

release: 0.0.12

New features

• quanto kernels library (not used for now in quantize).

Breaking changes

• quantization types are now all quanto.dtype