flagos-ai/FlagGems

Python 904 stars

FlagGems is an operator library for large language models implemented in the Triton Language.

✓ Synced 3h ago

1.1k Merged PRs

6 days Avg Merge Time

0m Fastest PR

6 months Slowest PR

      #247
      Global Speed Rank
    

Top Reviewers

#	Title	Author	Time	Reviews	Blocks
#1667	Fix PR label caching problem	@tengqm	2.8h	0	✓
#1666	Sync labels for labeler	@tengqm	3m	0	✓
#1665	Bump wait-for-workflow version and revise size labels	@tengqm	16m	0	✓
#1654	Auto label a PR regarding its size	@tengqm	1m	0	✓
#1645	[TSINGMICRO] fix hard code cuda in flag_gems.fused.FLA.utils.input_guard	@tsingmicro-public-e	5.1h	1	✓
#1633	[TSINGMICRO] add custom op for tsingmicro backend	@tsingmicro-public-e	1.9h	1	✓
#1626	[Iluvatar] use torch_moe_align_block_size as reference.	@awayzjj	1 day	6	✓
#1631	Remove redundant settings for coverage tool	@tengqm	3.6h	0	✓
#1630	Merge unit tests for operators, examples, and utils	@tengqm	1.8h	0	✓
#1628	Extract PR ID determination logic from individual workflows	@tengqm	2.1h	0	✓
#1604	[KUNLUNXIN] enable vdot benchmark test	@dongjibin1996	2 days	1	✓
#1627	Merge cpp op test	@tengqm	1.6h	0	✓
#1624	Refactor CI tests for experimental operators	@tengqm	55m	0	✓
#1623	Merge Nvidia backend test to the unit test workflow	@tengqm	2.3h	0	✓
#1616	[MTHREADS]: Remove the incorrect sqmma prompt	@Kylin1207	3.8h	2	✓
#1618	Extract Nvidia backend tests into shell script	@tengqm	1.1h	0	✓
#1617	Converge backend tests	@tengqm	44m	0	✓
#1613	Add Iluvatar/Ascend/Moore CI and refactor backend workflows into reusable sub-workflow	@douxetpur	10.4h	3	✓
#1568	[ASCEND]8.5.0 fix part1	@qianjinqiu	4 days	1	✓
#1575	fix concat_and_cache_mla UT failed when using cpu as reference.	@awayzjj	3 days	2	✓