pytorch/ao

Issues

• [RFC] Long Term QAT Flow

  #987 opened in 14 hours by andrewor14
  4

• is this only for linux?

  #957 opened 4 days ago by FurkanGozukara
  15

• Tensor Parallelism Support for AffineQuantizedTensor

  #988 opened in 14 hours by jerryzh168
  0

• Does torch.export preserve the quantize_per_tensor/dequantize_per_tensor ops?

  #986 opened in 14 hours by justinchuby
  0

• [feat] int8 flash attention

  #952 opened 5 days ago by felipemello1
  2

• installation fails on ARM64 Linux (aka Raspberry Pi 5)

  #913 opened 10 days ago by sunshinesfbay
  0

• Add `aarch64` support for torchao binaries

  #922 opened 7 days ago by jerryzh168
  0

• optimizer CPU offload doesn't work outside of CUDA

  #958 opened 4 days ago by bghira
  3

• [ROCm] Unable to Run FPX Weights

  #967 opened 3 days ago by Beinsezii
  4

• CUDA environment errors after installing ao

  #974 opened a day ago by philipbutler
  3

• [Feature request] Support bias in linear_nf4

  #979 opened 10 hours ago by ebsmothers
  6

• How do I perform Int8 activation and int8 weight QAT and export to onnx?

  #975 opened a day ago by ben-da6
  1

• `test_choose_qparams_token_asym` failing in nightly

  #970 opened 2 days ago by gau-nernst
  2

• [ao/float8] Discussion of the current Float8 recipe

  #972 opened 2 days ago by yanbing-j
  1

• RuntimeError: CUDA error: named symbol not found CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1 Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

  #968 opened 3 days ago by kolyan288
  1

• Building from source caused uint issue + suggested fix

  #951 opened 5 days ago by EnragedAntelope
  2

• [MPS] torchao low-bit-precision optim does not expose 'backend' argument to torch.compile

  #955 opened 4 days ago by bghira
  10

• int8 mixed precision training flaky tests?

  #945 opened 3 days ago by jerryzh168
  3

• torchao.ops import error on aarch64 linux

  #929 opened 4 days ago by ng-05
  10

• CPUOffloadOptimizer incompatible with learning rate schedulers

  #959 opened 4 days ago by bghira
  5

• float8 training with rowwise scaling

  #889 opened 15 days ago by vkuzo
  10

• [Question] Difference in MXLinear vs MXInferenceLinear grouping direction

  #932 opened 6 days ago by Abhijit-2592
  4

• run_vit_b_quant.py runs slower than run_bit_b.py

  #898 opened 14 days ago by jerryzh168
  9

• [torchao]NotImplementedError: AffineQuantizedTensor dispatch: attempting to run unimplemented operator/function: aten.permute.default

  #890 opened 8 days ago by yanbing-j
  9

• More fine-grained documentation needed for `torchao.autoquant()`

  #907 opened 11 days ago by suvadityamuk
  3

• Unify scaled INT8 matmul

  #862 opened 21 days ago by gau-nernst
  1

• we should ensure activation checkpointing with Float8Linear behaves optimally

  #893 opened 14 days ago by vkuzo
  1

• ValueError: ('Unsupported kind: ', 'FRAGMENT')

  #900 opened 14 days ago by loretoparisi
  1

• Repr of LinearActivationQuantizedTensor is broken

  #871 opened 14 days ago by BenjaminBossan
  3

• [wip] unification of torchao.float8 with the rest of torchao

  #894 opened 14 days ago by vkuzo
  1

• [CI] Add CI test for PyTorch 2.5.0rc

  #888 opened 16 days ago by gau-nernst
  0

• AffineQuantizedTensor dispatch: attempting to run unimplemented operator/function: `aten.linear.default`

  #875 opened 17 days ago by agunapal
  1

• Plans for 8da4w quantization

  #883 opened 17 days ago by sanchitintel
  6

• Out Of Memory Error

  #879 opened 19 days ago by bianxuxuxu
  2

• fpx test failing on main

  #792 opened 18 days ago by msaroufim
  5

• Update Eval scripts

  #853 opened 21 days ago by drisspg
  1

• Expected Tensor argument scales to have dtype torch.bfloat16, but got torch.float32 instead

  #876 opened 19 days ago by agunapal
  1

• AO dtype composability tracker

  #844 opened 22 days ago by msaroufim
  1

• Device error on 8/31 nightlies

  #795 opened 20 days ago by ebsmothers
  3

• Create report for Autoquant

  #855 opened 21 days ago by drisspg
  0

• quantization subclass typo

  #846 opened 21 days ago by YihengBrianWu
  2

• 'FakeTensor' object has no attribute 'layout_type'

  #789 opened a month ago by WAS-PlaiLabs
  4

• Add MyPy support

  #856 opened 21 days ago by drisspg
  0

• Add Autoquant Cache

  #854 opened 21 days ago by drisspg
  1

• int4_weight_only dequant does not restore original shape

  #848 opened 21 days ago by jerryzh168
  0

• [CI] Mac M1 build wheels should not run for PR commits

  #841 opened 22 days ago by gau-nernst
  0

• installing torchao from source does not work with cuda on latest torchao + latest PT nightly

  #810 opened a month ago by vkuzo
  2

• NotImplementedError: aten.linear.default not implemented when using MXTensor

  #796 opened a month ago by Ali-Flt
  7

• [Quantization + FSDP] Support `quantize_()` for DTensor

  #803 opened a month ago by gau-nernst
  1

• placeholder for float8 README.md images

  #793 opened a month ago by vkuzo
  1