Thomas Raoux 31b0c52142 [FRONTEND][BACKEND] Add flag to control accumulation for fp8 (#2300) ...

Change the dot to allow taking an initial accumulator and add a flag
that will allow the compiler to accumulate in a lower precision than the
output type.
On Hopper this flag is on by default which allows accumualting with
lower precision.
This only affect Hopper fp8 dot.

2023-09-15 18:42:54 -07:00

..

blocksparse

Initial code merge of Hopper support (#2036)

2023-08-07 09:53:04 +08:00

__init__.py

[TESTS] Added test for flash-attention (#1138)

2023-02-01 11:26:29 -08:00

cross_entropy.py

[FRONTEND] switch absolute imports to relative v2 (#1833)

2023-06-26 04:13:12 +00:00

flash_attention.py

[OPS] Remove unnecessary perf bug workaround (#2240)

2023-09-04 21:30:54 -07:00

matmul_perf_model.py

[FRONTEND] switch absolute imports to relative v2 (#1833)

2023-06-26 04:13:12 +00:00

matmul.py

[FRONTEND][BACKEND] Add flag to control accumulation for fp8 (#2300)

2023-09-15 18:42:54 -07:00