COMMITS
/ torchao/float8/float8_linear.py March 12, 2025
D
[float8] Bug fix: do not override requires_grad=False when enable_float8_all_gather=True (#1873)
Daniel Vega-Myhre committed
February 22, 2025
V
delete delayed scaling from torchao.float8 (#1753)
Vasiliy Kuznetsov committed
February 18, 2025
V
fix tensor parallelism for float8 training with rowwise scaling (#1718)
Vasiliy Kuznetsov committed
February 10, 2025
D
Support power of 2 scaling factors in float8 training and use e4m3 everywhere (#1670)
Daniel Vega-Myhre committed
January 31, 2025
V
float8 rowwise training: add FSDP workaround (#1629)
Vasiliy Kuznetsov committed
January 13, 2025
V
[cleanup][3/x] unify dynamic input and grad_output casting (#1480)
Vasiliy Kuznetsov committed
V
[cleanup][2/x] split float8 mm by delayed vs dynamic (#1461)
Vasiliy Kuznetsov committed
V
[cleanup][1/x] make hp_tensor_to_float8_dynamic only work with hp inputs (#1458)
Vasiliy Kuznetsov committed
December 20, 2024
V
for now, delete the float8-all-gather-only functionality from float8 … (#1451)
Vasiliy Kuznetsov committed
December 19, 2024
V
refactor: move delayed and static scaling to StatefulFloat8Linear (#1439)
Vasiliy Kuznetsov committed
December 4, 2024
L
[float8] Allow specifying arbitrary dtype for each tensor (#1378)
Luca Wehrstedt committed
November 25, 2024
V
float8 training: move module attribute setting to sync function (#1341)
Vasiliy Kuznetsov committed
November 22, 2024
V
float8 training: fix bug with AC + compile (#1329)
Vasiliy Kuznetsov committed
V
fix bug with float8 training + FSDP2 + TP (#1327)
Vasiliy Kuznetsov committed
November 19, 2024
V
float8 with delayed scaling: fix autocast handling (#1306)
Vasiliy Kuznetsov committed
November 15, 2024
V
float8 delayed scaling: remove need to use workaround for AC (#1291)
Vasiliy Kuznetsov committed
November 11, 2024
A
Ruff Lint fixes (torchao/float8) (#1239)
Apurva Jain committed
October 31, 2024
Y
October 7, 2024
V
float8 training axiswise scaling support with per-gemm-argument configuration (#940)
Vasiliy Kuznetsov committed
V
add axiswise scaling to Float8Linear (#920)
Vasiliy Kuznetsov committed
V
add axiswise granularity to Float8Tensor (#919)
Vasiliy Kuznetsov committed
October 2, 2024
Y
Use checkpoint to enforece the recomputation of fp8 weight
y-sq committed
August 28, 2024
V
float8 training: add static scaling (#760)
Vasiliy Kuznetsov committed
July 30, 2024
V
move float8_experimental to torchao/float8
Vasiliy Kuznetsov committed