v0.8.4

What's Changed

• chore: Update torch to 2.11 by @gkumbhat in #206

Full Changelog: v0.8.3...v0.8.4

v0.8.3

What's Changed

• fix: deprecated import by @maxdebayser in #202
• feat: update transformers to 5.x by @andrea-fasoli in #204
• feat: update readme with fp8 dependency groups by @andrea-fasoli in #205

New Contributors

• @maxdebayser made their first contribution in #202

Full Changelog: v0.8.2...v0.8.3

v0.8.2

What's Changed

• fix: FP8 fallback for AIU addons running on CPU by @andrea-fasoli in #200
• fix: update FP8 syntax for custom torch._scaled_mm on CPU by @andrea-fasoli in #201

Full Changelog: v0.8.1...v0.8.2

v0.8.1

What's Changed

• chore: Update torch to 2.10 by @gkumbhat in #198

New Contributors

• @gkumbhat made their first contribution in #198

Full Changelog: v0.8.0...v0.8.1

v0.8.0

What's Changed

• fix: Fix build and check packages flow by @tharapalanivel in #192
• chore: upgrade torch to allow 2.9 by @Ssukriti in #190
• fix: Fixes for paged fp8 attention with chunked prefill by @ani300 in #191

New Contributors

• @Ssukriti made their first contribution in #190

Full Changelog: v0.7.0...v0.8.0

v0.7.0

What's Changed

• feat: Quantization Refactor by @BrandonGroth in #169
• fix: remove custom scaled bmm op on cpu and fix fp8 test by @andrea-fasoli in #187
• chore(deps): Update torch requirement from <2.8,>=2.2.0 to >=2.2.0,<2.9 by @dependabot[bot] in #177
• chore(deps): Update accelerate requirement from !=0.34,<1.10,>=0.20.3 to >=0.20.3,!=0.34,<1.11 by @dependabot[bot] in #179
• chore(deps): Update transformers requirement from <4.56,>=4.45 to >=4.45,<4.58 by @dependabot[bot] in #186

Full Changelog: v0.6.0...v0.7.0

v0.6.0

What's Changed

• fix: enabling block-by-block evaluation for granite-3.x-models by @bayo-ibm in #165
• fix: pylint false alarm on libdevice functions by @chichun-charlie-liu in #166
• fix: Add version limits for torchao, ensure compat with 0.12 + AIU by @ani300 in #168
• feat: Change paged FP8 prefill back to regular attention by @ani300 in #171
• feat: FP8 requested changes by @ani300 in #173
• chore(deps): Update triton requirement from <3.4,>=3.0 to >=3.0,<3.5 by @dependabot[bot] in #170
• chore(deps): Update transformers requirement from <4.54,>=4.45 to >=4.45,<4.56 by @dependabot[bot] in #172
• fix: FP8 TP fixes by @ani300 in #176

Full Changelog: v0.5.0...v0.6.0

v0.5.0

What's Changed

• chore(deps): Update transformers requirement from <4.53,>=4.45 to >=4.45,<4.54 by @dependabot[bot] in #151
• fix: Mark FP8 scale to have the same batch size as input by @ani300 in #163
• chore: Update torch requirement from <2.6,>=2.2.0 to >=2.2.0,<2.8 by @dependabot[bot] in #100
• feat: Add QmaxDynamic to allow unify Qmax , Qminmax, pertokenmax by @iqbal-saraf in #139
• feat: GPTQv2 enablement for fms_mo by @bayo-ibm in #138
• chore(deps): Update accelerate requirement from !=0.34,<1.9,>=0.20.3 to >=0.20.3,!=0.34,<1.10 by @dependabot[bot] in #164

New Contributors

• @bayo-ibm made their first contribution in #138

Full Changelog: v0.4.1...v0.5.0

v0.4.1

What's Changed

• feat: Per-sequence scaling in FP8 attention, FP8 fixes by @ani300 in #162

Full Changelog: v0.4.0...v0.4.1

v0.4.0

What's Changed

• feat: add guards to sawb recomputation by @andrea-fasoli in #131
• build: Move torchvision to an optional dependency by @BrandonGroth in #144
• fix: feat: fix for new transformers (>4.48) and new QLinear for INT8 training with HW emulation by @chichun-charlie-liu in #141
• chore(deps): Update transformers requirement from <4.52,>=4.45 to >=4.45,<4.53 by @dependabot[bot] in #127
• build: Move triton to an optional dependency by @BrandonGroth in #146
• chore(deps): Update accelerate requirement from !=0.34,<1.7,>=0.20.3 to >=0.20.3,!=0.34,<1.9 by @dependabot[bot] in #143
• build: Make non-essential dependencies optional by @BrandonGroth in #147
• fix: fix available_packages by @chichun-charlie-liu in #153
• fix: Saved qconfig recipe being overwritten with defaults by @BrandonGroth in #152
• fix: Remove gptqmodel Warning on startup by @BrandonGroth in #156
• fix: Remove llmcompressor oneshot import deprecation warning by @BrandonGroth in #157
• feat: addons for FP8 attention bmm, paged attention, and linear in FMS by @ani300 in #154
• feat: addons for FP8 attention bmm and linear in FMS by @andrea-fasoli in #149
• feat: add QA and MaskedLM task for FP8 encoder instantiation by @andrea-fasoli in #148
• feat: AIU sim for FP8 (DL8/DL16) added to triton kernel by @chichun-charlie-liu in #159
• fix: qkvsync bug fix by @chichun-charlie-liu in #161
• chore(deps): Update datasets requirement from <4.0,>=3.0.0 to >=3.0.0,<5.0 by @dependabot[bot] in #160

New Contributors

• @ani300 made their first contribution in #154

Full Changelog: v0.3.0...v0.4.0