DeepSpeed
Sparse attn triton v1.0 support + torch1.8 test runner
#1374
Merged

Sparse attn triton v1.0 support + torch1.8 test runner #1374

jeffra merged 23 commits into master from reyazda/test-sparse-v2
jeffra
let the sparse tests run
18a8dffe
fixing the sparse-APIs to use the latest triton version
80dbe2f7
jeffra update with assert and some fixes
be55c85c
jeffra add torch18 tests and fix sparse-attn checks
9e2dee4b
jeffra turn back on tests
e7028669
jeffra use relative paths for megatron jsons
54752d0e
jeffra factor out relative path for unit test files
741f8b55
jeffra set test path
626a51e4
jeffra jeffra requested a review from awan-10 awan-10 4 years ago
jeffra jeffra requested a review from cli99 cli99 4 years ago
jeffra jeffra requested a review from conglongli conglongli 4 years ago
jeffra jeffra requested a review from eltonzheng eltonzheng 4 years ago
jeffra jeffra requested a review from minjiaz minjiaz 4 years ago
jeffra jeffra requested a review from niumanar niumanar 4 years ago
jeffra jeffra requested a review from RezaYazdaniAminabadi RezaYazdaniAminabadi 4 years ago
jeffra jeffra requested a review from samyam samyam 4 years ago
jeffra jeffra requested a review from ShadenSmith ShadenSmith 4 years ago
jeffra jeffra requested a review from tjruwase tjruwase 4 years ago
jeffra jeffra marked this pull request as draft 4 years ago
jeffra refactor sparse attn imports
318ae956
jeffra Merge branch 'master' into reyazda/test-sparse-v2
99634357
jeffra fix relative import
49514bd3
jeffra rename test_path so pytest doesn't think its a test
61bf986e
jeffra skip test_configurable_parallel for now until fixed
35245fef
jeffra moe fix
0d28040f
jeffra fixes random connection reset test failures for some unit tests
d1615ff0
resolve the TK with correct setting based on dtype and block size
940abc6c
Merge branch 'reyazda/test-sparse-v2' of github.com:microsoft/DeepSpe…
bd13efde
jeffra fix megatron regression, add moe unit test, fix moe ckpt comparison
7d9c5fb1
jeffra add sparse-attn skip if not compatible
7733f00d
jeffra Merge branch 'master' into reyazda/test-sparse-v2
80011245
jeffra skip moe ckpt test if old torch
08695e29
jeffra turn back on test_configurable_parallel
1324589e
jeffra tear down torch dist pg when test completes
3beb9fd9
jeffra jeffra marked this pull request as ready for review 4 years ago
tjruwase
tjruwase approved these changes on 2021-09-21
jeffra jeffra merged 6996bb01 into master 4 years ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone