feat(fp8): run scaled and raw fp8 checkpoints on the fp8 tensor cores - #9478

Open

Pfannkuchensack wants to merge 55 commits into

invoke-ai:mainfrom

Pfannkuchensack:feat/fp8_compute_raw

Commits on Jul 31, 2026

feat(fp8): enable FP8 storage for Z-Image
Pfannkuchensack
committed
93ecda8
Chore openapi
Pfannkuchensack
committed
cd8f052
feat(fp8): enable FP8 storage for Anima
Pfannkuchensack
committed
b47a92c
fix(fp8): never apply FP8 storage to already-quantized weights
Pfannkuchensack
committed
3feb027
feat(quantization): run ComfyUI scaled-fp8 checkpoints on the fp8 tensor cores
Pfannkuchensack
committed
1bec901
feat(krea2): keep ComfyUI scaled-fp8 checkpoints quantized and run them on fp8 tensor cores
Pfannkuchensack
committed
580813c
test(fp8): verify LoRA sidecar patching over fp8 weights, harden metadata read
Pfannkuchensack
committed
efdf552

Commits on Aug 1, 2026

Commits on Aug 2, 2026

Merge branch 'main' into feat/fp8_quantized_guard
Pfannkuchensack
authored
8826e8d
Merge branch 'main' into feat/fp8_anima
Pfannkuchensack
authored
ec0b1f3
Merge branch 'main' into feat/fp8_zimage
Pfannkuchensack
authored
8734f29

Commits on Aug 4, 2026

fix(fp8): keep the Qwen3-VL encoder's unquantized layers fp8-resident
Pfannkuchensack
committed
a3784eb

Commits on Aug 5, 2026

Commits on Aug 6, 2026

Merge branch 'main' into feat/fp8_zimage
Pfannkuchensack
authored
6b52aec
Merge branch 'main' into feat/fp8_anima
Pfannkuchensack
authored
6c38314
Merge branch 'main' into feat/fp8_quantized_guard
Pfannkuchensack
authored
5d81932

Commits on Aug 7, 2026

feat(fp8): run raw fp8 checkpoints on the tensor cores
Pfannkuchensack
committed
e93854e

Commits on Aug 8, 2026

Commits on Aug 14, 2026

Commits on Aug 15, 2026

test(fp8): capture a mixed fp8 FLUX.2 layout with input scales
Pfannkuchensack
committed
83dd95e

Commits on Aug 17, 2026

Commits on Aug 19, 2026

Merge remote-tracking branch 'upstream/main' into feat/fp8_zimage
Pfannkuchensack
committed
4ccb6b1
Merge remote-tracking branch 'upstream/main' into feat/fp8_compute_raw
Pfannkuchensack
committed
2c00114
chore(fp8): regenerate openapi schema for the fp8_compute settings

Pfannkuchensack
and
claude
committed
9ef7a2d
test(fp8): drop the Z-Image entry from the exclusion parametrize

Pfannkuchensack
and
claude
committed
5c3f8aa
Merge branch 'feat/fp8_zimage' into feat/fp8_anima
Pfannkuchensack
committed
6db534a
Merge branch 'feat/fp8_anima' into feat/fp8_quantized_guard
Pfannkuchensack
committed
93d3ce5
Merge branch 'feat/fp8_quantized_guard' into feat/fp8_compute_raw
Pfannkuchensack
committed
decabd5
docs(config): list fp8_compute_full_precision_hints in the config docstring

Pfannkuchensack
and
claude
committed
060f879
feat(fp8): run scaled fp8 FLUX.1 checkpoints on the tensor cores

Pfannkuchensack
and
claude
committed
b2854a1
feat(fp8): run scaled fp8 FLUX.2 checkpoints on the tensor cores

Pfannkuchensack
and
claude
committed
587c0fa
feat(fp8): run scaled fp8 Mistral encoders on the tensor cores

Pfannkuchensack
and
claude
committed
27a138f
feat(fp8): run scaled fp8 Anima checkpoints on the tensor cores

Pfannkuchensack
and
claude
committed
9f25f1e

Commits on Aug 24, 2026

Merge branch 'main' into feat/fp8_compute_raw
Pfannkuchensack
authored
d8f1618

Commits on Aug 25, 2026

Commits on Aug 26, 2026

Merge branch 'main' into feat/fp8_compute_raw
Pfannkuchensack
authored
922fb13

Commits on Aug 28, 2026

Merge branch 'main' into feat/fp8_compute_raw
Pfannkuchensack
authored
04b6f14

Commits on Aug 29, 2026

fix(fp8): address round-3 review — Qwen3-VL split, scale-axis and probe caching

Pfannkuchensack
and
claude
committed
6fa8ac8