Stas Bekman c2bb53f20f TiledMLP + SequenceTiledCompute: improve the bs>1 use-case (#7422) ...

Improved TiledMLP and SequenceTiledCompute for bs>1

This PR:
- extends the testing utils to add `CaptureStd*`, `CaptureLogger`
context managers
- extends the test to run both bs=1 and bs=2
- use an uneven seqlen to test varlen shards
- flattens bs+seqlen dim, to avoid problems with grad tensor strides
when bs>1 - mlp doesn't care for the bs dimension so using a pretend
`bs*seqlen` seqlen instead and restoring the shape at the end for the
grad.

---------

Signed-off-by: Stas Bekman <stas@stason.org>
Co-authored-by: Logan Adams <114770087+loadams@users.noreply.github.com>

2025-07-16 09:30:08 -07:00

..

accelerator

Update DeepSpeed copyright license to Apache 2.0 (#3111)

2023-03-30 17:14:38 -07:00

benchmarks

Fix the openfold training. (#4657)

2023-11-09 02:04:12 +00:00

hybrid_engine

DeepSpeed Chat (#3186)

2023-04-11 11:53:38 -07:00

lightning

Update DeepSpeed copyright license to Apache 2.0 (#3111)

2023-03-30 17:14:38 -07:00

model

Sanitize inputs to eval() (#6745)

2024-11-13 09:04:56 -08:00

onebit

Add Compressedbackend for Onebit optimizers (#5473)

2024-06-05 20:28:46 +00:00

perf

CPUAdam fp16 and bf16 support (#5409)

2024-05-20 12:50:20 +00:00

small_model_debugging

Guanhua/partial offload rebase v2 (#590) (#4636)

2023-11-06 14:15:16 -08:00

torch_compile

[compile] Show breakdown of graph break (#6601)

2024-10-14 17:31:34 +00:00

unit

TiledMLP + SequenceTiledCompute: improve the bs>1 use-case (#7422)

2025-07-16 09:30:08 -07:00

.coveragerc

Reduce Unit Test Times (Part 3) (#3850)

2023-07-12 00:35:49 +00:00

conftest.py

tests/conftest.py: automatically add local deepspeed repo when running tests (#7317)

2025-05-28 23:32:49 +00:00

pytest.ini

Inference V2 Human Eval (#4804)

2024-02-22 22:55:40 +00:00