Pinned issues
Issues
- 0
Docs Revamp
#181 opened by msaroufim - 1
Saving autoquant quantization plan
#320 opened by RobinKa - 0
ARM builds in CI
#335 opened by msaroufim - 0
Improvement ideas for `hf_eval.py`
#332 opened by gau-nernst - 3
FloatQuantization subclass
#228 opened by msaroufim - 16
FP6 dtype!
#208 opened by NicolasMejiaPetit - 82
[feature request] np.packbits / np.unpackbits, general BitTensors (maybe can be just tensors with dtype torch.bits8 or have a new dtype torch.bits introduced) and bit packed tensors utilities for saving memory / accesses, support for BitTensors wherever BoolTensors are used
#292 opened by vadimkantorov - 3
[Tracker] WIP features for torchao 0.3
#252 opened by supriyar - 2
- 2
Bitnet 1.58 prework, POC, and staging
#281 opened by CoffeeVampir3 - 7
`dequantize_affine` modified the `input` in-place
#289 opened by yiliu30 - 2
apply_dynamic_quant for vit_b_16
#63 opened by cpuhrsch - 0
torch.iinfo() support for sub byte dtypes
#308 opened by vayuda - 4
custom cuda extensions make installing ao hard
#288 opened by msaroufim - 0
Numerics checks between NF4 and bnb nf4
#295 opened by msaroufim - 4
Generic packing algorithms from size N to M
#284 opened by vayuda - 0
- 1
- 1
torchao init: ImportError: libcudart.so.12: cannot open shared object file: No such file or directory
#260 opened by mikekgfb - 1
HQQ Tracker
#255 opened by HDCharles - 1
[Tracker] WIP Features for torchao v0.2
#132 opened by supriyar - 2
[Question] MBU in automated CI?
#237 opened by cadedaniel - 1
Building torchao from source installs unnecessary torch and nvidia packages every time
#231 opened by gau-nernst - 0
Doc build failing on main
#221 opened by msaroufim - 0
- 2
[RFC] Plans for sparsity
#143 opened by jcaip - 7
2:4 sparsity + PTQ(int8) model's inference
#134 opened by RanchiZhao - 1
[RFC] More general affine quantization primitives
#160 opened by jerryzh168 - 12
- 0
[NF4][FSDP2]: enable multi-gpu CI
#202 opened by weifengpy - 7
NF4Tensor uses 8 bits of memory
#209 opened by cuichenx - 0
Sparsity OSS colab tracker
#113 opened by jcaip - 21
[RFC] Plans for torchao
#47 opened by supriyar - 4
- 0
[RFC] Plans for LLM QAT
#86 opened by andrewor14 - 1
[Tracker] General feature requests for torchao
#65 opened by supriyar - 0
1 bit LLM implementation
#67 opened by msaroufim - 0
[NF4][FSDP2] DTensor + fused adam on cpu
#205 opened by weifengpy - 1
Semi-Structured Sparsity unsupported for Windows
#191 opened by philipbutler - 0
Custom CUDA extensions
#137 opened by msaroufim - 2
- 1
how do you implement int8 mm on cuda?
#34 opened by ThisisBillhe - 4
Project implicitly depends on torch nightly
#29 opened by ELanning - 2
- 1
Reconsider using class method for Int8DynamicallyQuantizedLinearWeight.from_float
#21 opened by HDCharles - 3
SDPA: scaled_dot_product_attention to use this attetino donot need to retain ?
#14 opened by ganliqiang