COMMITS
/ torchao/utils.py March 17, 2025
V
enable tests on mx_formats + Blackwell (#1905)
Vasiliy Kuznetsov committed
A
Add copyright header and pre-commit validator (#1900)
Apurva Jain committed
March 5, 2025
A
Revert "Move torchao/_models to benchmarks/_models" (#1844)
Apurva Jain committed
V
add mxfp8_cublas recipe to mx_formats (#1831)
Vasiliy Kuznetsov committed
D
Add support for copy_ for plain layout and tensor core tiled layout (#1791) (#1804)
Driss Guessous committed
P
ROCm OCP FP8 Support (#1677)
Peter Yeh committed
March 4, 2025
A
Move torchao/_models to benchmarks/_models (#1784)
Apurva Jain committed
February 22, 2025
J
move decorators to testing/utils.py (#1761)
Jesse Cai committed
February 21, 2025
P
[Reland] ROCm CI (Infra + Skips) (#1581)
Peter Yeh committed
February 7, 2025
A
Add boiler plate code to Tensor subclass (#1663)
Apurva Jain committed
January 30, 2025
V
skip failing MX tests on cuda capability 10.0 (#1624)
Vasiliy Kuznetsov committed
January 17, 2025
A
Revert "Enable ROCM in CI" (#1583)
andrewor14 committed
M
Enable ROCM in CI (#999)
Mark Saroufim committed
January 9, 2025
J
Skip calling unwrap_tensor_subclass for torch 2.7+ (#1531)
Jerry Zhang committed
December 1, 2024
A
Update hardware check conditions (#1356)
Apurva Jain committed
November 26, 2024
Y
Enable CPU Offload for Intel GPU (#1324)
Yang Yang committed
A
Add hardware check to fp8 quant (#1314)
Apurva Jain committed
November 12, 2024
D
Fix Safe Load for NF4 (#1241)
Driss Guessous committed
November 5, 2024
J
Add a developer guide for exporting to executorch (#1219)
Jerry Zhang committed
October 29, 2024
J
Update utils.py (#1186)
Jerry Zhang committed
October 10, 2024
A
Rename AQT#2 LayoutType -> Layout (#1049)
Apurva Jain committed
A
Aqt rename#1 Layout -> TensorImpl (#1046)
Apurva Jain committed
October 8, 2024
M
Revert "Rename Layout -> TensorImpl" (#1040)
Mark Saroufim committed
A
Rename Layout -> TensorImpl (#1028)
Apurva Jain committed
October 4, 2024
S
training ir torchao migration
Shangdi Yu committed
October 1, 2024
J
Print args/kwargs types for utils and examples (#963)
Jerry Zhang committed
September 30, 2024
M
Skip test_choose_qparams_token_asym on pt 2.6 (#981)
Mark Saroufim committed
September 27, 2024
J
Supporting tensor parallelism for int8 weight only quant (#939)
Jerry Zhang committed
September 3, 2024
J
Move more utils to TorchAOBaseTensor (#784)
Jerry Zhang committed
August 30, 2024
D
Add Float8 Weight Only and FP8 weight + dynamic activation (#740)
Driss Guessous committed
August 23, 2024
D
Add sparse marlin 2:4 gemm op (#733)
Diogo Venâncio committed
August 22, 2024
J
Fix affine quantized tensor to device calls (#726)
Jerry Zhang committed
August 15, 2024
M
Fix source version check (#684)
Mark Saroufim committed
August 14, 2024
M
retry version guard fix (#679)
Mark Saroufim committed
August 1, 2024
J
Allow `benchmark_model` to accept args and kwargs (#586)
Jerry Zhang committed
July 16, 2024
M
fix infer_schema bc change (#509)
Mark Saroufim committed
July 10, 2024
A
Added support to benchmark_model for cpu and mps (#406)
Apurva Jain committed
July 5, 2024
B
fix torchtune in Genie (#480) (#480)
Botao Chen committed
July 2, 2024
J
Add decorator for custom op and inductor decomp registration (#434)
Jerry Zhang committed
June 17, 2024
M
Remove all dependencies except torch (#369)
Mark Saroufim committed
June 14, 2024
H
Generalize Model Size Code (#364)
HDCharles committed
June 13, 2024
J
Deprecate top level quantization APIs (#344)
Jerry Zhang committed
June 12, 2024
V
make torchao test discovery pass in fbcode
Vasiliy Kuznetsov committed
June 7, 2024
J
Move some util functions from quantization.utils to torchao.utils (#337)
Jerry Zhang committed
June 4, 2024
J
Refactor int4 and int8 weight only quantization to use `quantize` (#301)
Jerry Zhang committed
May 28, 2024
V
Add a prototype of MX format training and inference (#264)
Vasiliy Kuznetsov committed
May 24, 2024
M
Make fp8 test explicit (#266)
Mark Saroufim committed
May 20, 2024
L
In tutorials/quantize_vit, extract common methods to util.py (#238)
lancerts committed