Shucai Xiao 79bebc4ffe fp8 type support (#357) ...

* add two fp8 data types `tl.float8e4b8` and `tl.float8e5b16` to triton.
* add SW type conversion between `tl.float8e4b8/tl.float8e5b16` and `fp16`
* change flashattention to support fp8 in q/k.

2023-11-02 15:51:23 -05:00

..

backends

[FRONTEND] drop the GIL around more CUDA ops (#2173)

2023-08-24 20:31:38 -07:00

__init__.py

[RUNTIME] refactor driver (#1515)

2023-04-12 23:50:44 -07:00

autotuner.py

[FA fwd D=128] Reduce LDS usage in epilogue (#340)

2023-10-25 12:10:34 -05:00

cache.py

[RUNTIME] Fix cache dir (#2196)

2023-08-29 21:07:16 -04:00

driver.py

Initial code merge of Hopper support (#2036)

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

errors.py

[FRONTEND] cleaned up language; added frontend function for globaltimer special register (#1525)

2023-04-14 15:29:27 -07:00

jit.py

fp8 type support (#357)

2023-11-02 15:51:23 -05:00