46 Commits

Author	SHA1	Message	Date
justheuristic	577275bd8c	cast properly	2022-09-18 00:30:57 +03:00
justheuristic	45dc1983e9	cast properly	2022-09-18 00:28:03 +03:00
justheuristic	702cc72018	debug asset	2022-09-18 00:26:46 +03:00
justheuristic	a214824f93	matmul -1- addmm	2022-09-18 00:24:59 +03:00
justheuristic	14048a3c16	safer cast	2022-09-18 00:24:20 +03:00
justheuristic	5b169f18e4	change typecast behavior	2022-09-18 00:21:15 +03:00
justheuristic	1da4880262	change typecast behavior	2022-09-18 00:19:22 +03:00
justheuristic	1145589f84	change typecast behavior	2022-09-18 00:15:57 +03:00
justheuristic	d6e25b5f5e	change typecast behavior	2022-09-18 00:15:18 +03:00
justheuristic	e2b523d071	change typecast behavior	2022-09-18 00:07:05 +03:00
justheuristic	85bf5294a6	debug assert	2022-09-18 00:01:25 +03:00
justheuristic	210b9ed9ce	debug assert	2022-09-18 00:00:45 +03:00
justheuristic	647c976a74	change order	2022-09-17 23:59:36 +03:00
justheuristic	0de1a4494b	change order	2022-09-17 23:53:49 +03:00
justheuristic	e9b87112ee	un-fuse bias	2022-09-17 23:51:28 +03:00
justheuristic	56a074f6dc	un-fuse bias	2022-09-17 23:46:37 +03:00
justheuristic	d9ca0ed905	un-fuse bias	2022-09-17 23:44:28 +03:00
justheuristic	eac9aca460	cast bias too	2022-09-17 23:38:09 +03:00
justheuristic	a9fe0ff98c	recast to fp16	2022-09-17 23:34:22 +03:00
justheuristic	fc4a135ed1	clearer assertions	2022-09-17 23:24:26 +03:00
justheuristic	cc4858c2fd	some kind of warning or something when this is first executed to make people aware that a cast happens and the operation quantization is performed in fp16.	2022-09-17 20:46:04 +03:00
dbaranchuk	e2a75769f2	bug fix	2022-09-11 21:41:46 +03:00
dbaranchuk	4dd475ced4	refactoring	2022-09-11 06:28:17 +03:00
dbaranchuk	d358999e9e	refactoring	2022-09-11 06:26:15 +03:00
dbaranchuk	ee325f0215	clarified an exception message	2022-09-11 06:18:44 +03:00
dbaranchuk	42b5fc9acc	add memory effcient backward option	2022-09-11 05:51:29 +03:00
Dmitry Baranchuk	843ad0631c	Merge pull request #1 from TimDettmers/main ... Update main branch	2022-09-10 19:33:21 -07:00
dbaranchuk	8d34d36f15	req_gradA for casted & more efficient and accurate fp16 backward	2022-08-29 00:56:08 +03:00
dbaranchuk	b3fee1ed6a	add dtype <-> fp16 cast	2022-08-26 04:11:40 +03:00
dbaranchuk	4d6174bc63	memory efficient fp16 backward	2022-08-25 19:09:23 +03:00
Max Ryabinin	9fc0ab415c	Remove unused code	2022-08-24 18:43:18 +03:00
dbaranchuk	876387dc0c	minor fixes	2022-08-24 01:12:48 +03:00
dbaranchuk	656de8ed11	minor fixes	2022-08-23 23:53:43 +03:00
dbaranchuk	1753aa0418	refactoring	2022-08-23 23:51:00 +03:00
dbaranchuk	8ae9bb23ad	add memory efficient backward	2022-08-23 23:39:54 +03:00
Tim Dettmers	de354f7ded	Added fused bias to matmullt.	2022-08-16 12:00:54 -07:00
Tim Dettmers	f9cbe2fe99	Fixed prod Python < 3.7 compatibility in function.py.	2022-08-08 09:13:22 -07:00
Tim Dettmers	62441815bc	Removed prod for Python <= 3.7 compatibility.	2022-08-08 05:20:36 -07:00
Tim Dettmers	758c7175a2	Merge branch 'debug' into cuda-bin-switch-and-cli	2022-08-04 08:03:00 -07:00
Tim Dettmers	cc5b323876	Merge branch 'extract_outliers' into debug	2022-08-04 07:40:48 -07:00
Tim Dettmers	451fd9506e	Added fixes for the case that matmullt dim A is zero, e.g. [0, 768].	2022-08-03 11:54:01 -07:00
Titus von Koeller	ea7c14f8ef	reran black with linelength 80 for greater readability	2022-08-01 09:32:47 -07:00
Titus von Koeller	bfa0e33294	ran black and isort for coherent code formatting	2022-08-01 03:31:48 -07:00
Tim Dettmers	389f66ca5a	Fixed direct extraction masking.	2022-07-27 01:46:35 -07:00
Tim Dettmers	47a73d94c3	Matmullt with direct outlier extraction for 8-bit inference.	2022-07-26 19:15:35 -07:00
Tim Dettmers	c771b3a75a	Most tests passing.	2022-07-22 14:41:05 -07:00