505 Commits

This Branch

This Branch

All Branches

Author	SHA1	Message	Date
RandNMR73	efcb3c4058	8 gpus	2025-12-25 02:00:45 +00:00
RandNMR73	36cdddfa9a	bigger batch size + no grad around wrapped_flash_attn	2025-12-25 01:36:17 +00:00
RandNMR73	56faaa37bd	re-enable delta_s since inference is terrible without it	2025-12-25 00:09:36 +00:00
RandNMR73	d0e44170b2	disable delta_s	2025-12-24 23:55:11 +00:00
RandNMR73	f2eb855c2a	change block size back to 64x64	2025-12-24 23:36:37 +00:00
RandNMR73	f6080f7cee	add sageattn3 inference example	2025-12-24 22:59:02 +00:00
RandNMR73	8678798e58	add delta_s back	2025-12-24 22:58:10 +00:00
RandNMR73	346bbfaf6d	revert block size to 128x128	2025-12-24 22:42:52 +00:00
RandNMR73	52480b2514	adjus quant kernel block size	2025-12-24 21:59:04 +00:00
RandNMR73	460df09e20	adjust sage3 block size to 64x64	2025-12-24 21:49:43 +00:00
RandNMR73	3641487374	print sageattn file	2025-12-24 20:16:07 +00:00
RandNMR73	8d68a4fd46	add SageAttn3 with QAT	2025-12-24 09:28:22 +00:00
RandNMR73	7afe836f28	5090 testing	2025-12-24 06:04:36 +00:00
RandNMR73	3b70702af9	fix masking + causal and non-causal logic in qat attn	2025-12-23 01:36:49 +00:00
RandNMR73	9bcc41164a	checkpoint (qat attn in progress)	2025-12-23 01:36:46 +00:00
Matthew Noto	d41912cd7f	fix import	2025-12-23 01:34:18 +00:00
Matthew Noto	b06c03845b	qat attn in progress + refactor nvfp4 utils	2025-12-23 01:34:18 +00:00
Matthew Noto	6a7f08e73b	fix DeepGEMM path	2025-12-23 01:34:18 +00:00
Matthew Noto	fe9e94ed0c	fix DeepGEMM path	2025-12-23 01:34:18 +00:00
Matthew Noto	2fa6fc102b	add inference repo	2025-12-23 01:34:15 +00:00
Matthew Noto	9b125520f0	nvfp4 utils in progress	2025-12-23 01:31:21 +00:00
Matthew Noto	ebdb160af9	checkpoint	2025-12-23 01:31:21 +00:00
Peiyuan Zhang	b628b5fc7e	save	2025-12-23 01:31:21 +00:00
Peiyuan Zhang	3a9e4d3b89	fake quant done	2025-12-23 01:31:21 +00:00
Matthew Noto	c137c9f9c5	add real and fake quant precision tests	2025-12-23 01:31:21 +00:00
Peiyuan Zhang	18aebfadc7	update	2025-12-23 01:31:17 +00:00
Peiyuan Zhang	84ed3ceead	1005 morning	2025-12-23 01:30:18 +00:00
Peiyuan Zhang	130b46634a	save	2025-12-23 01:30:16 +00:00
Peiyuan Zhang	fa91001127	update	2025-12-23 01:28:09 +00:00
Peiyuan Zhang	c47be5c795	update	2025-12-23 01:28:08 +00:00
Peiyuan Zhang	032016f7dd	update	2025-12-23 01:28:08 +00:00
Peiyuan Zhang	2a28c081be	update	2025-12-23 01:28:08 +00:00
Peiyuan Zhang	a1ab4a7eeb	stash	2025-12-23 01:28:05 +00:00
Peiyuan Zhang	8abfe234af	update	2025-12-23 01:27:22 +00:00
Peiyuan Zhang	9161ef60da	update	2025-12-23 01:25:05 +00:00
Peiyuan Zhang	6a58c3aa64	+ generator sage 3	2025-12-23 01:21:59 +00:00
Peiyuan Zhang	85553c2717	+ fp4 linear + fp4 attn, all with 16-bit bwd	2025-12-20 11:09:33 +00:00
Peiyuan Zhang	da3f43c7fa	+baseline	2025-12-20 11:09:32 +00:00
Loay Rashid	e53730f324	[docs] Minor Fixes (#942)	2025-12-19 16:48:16 -06:00
Loay Rashid	7a4fe9086a	[feat] Support sequence packing and shard after pachification for USP (#894)	2025-12-19 16:19:46 -06:00
Ohm-Rishabh	d277361aae	[misc] add schedule configurations to pytorch profiler (#934)	2025-12-18 01:45:23 -06:00
alexzms	734a54e7a9	[ci]: Use pre-built docker image & skip VSA compilation (#939)	2025-12-16 23:14:11 -08:00
alexzms	91364982df	[Feature] Support for Variable Q/KV Sequence Lengths in VSA ThunderKittens kernel (#911)	2025-12-16 20:08:15 -08:00
William Lin	50145e4fcb	[CI] Fix CI tests (#935)	2025-12-16 04:59:43 -08:00
William Lin	4112507e99	[misc] upgrade pytorch version to 2.9.0 (#928)	2025-12-15 04:12:43 -08:00
William Lin	424fc2b4ae	[bugfix] [lora] [distillation] Fix lora distillation bug (#933)	2025-12-15 04:12:02 -08:00
William Lin	e6066223e6	[bugfix] [VSA] [distillation] Various bugfixes for VSA and distillation and nightly tests (#932)	2025-12-12 16:51:54 -08:00
William Lin	b6fa3d24d8	[misc] update wechat image (#931)	2025-12-11 21:22:21 -08:00
Ketaki Tank	55c2e7cd76	[feat] Add fvd implementation (#923)	2025-12-11 19:06:19 -08:00
Tuyabei	5a549af823	[bugfix] [VSA] Fix block_size computation in backward kernel (#925)	2025-12-10 14:36:40 -08:00