pytorch/benchmark

Pull Requests Commits

Another refactor, to move things under E2Emodel API

wconstab committed 3 years ago

46414ecd

wconstab committed 3 years ago

7b168d80

Separate distributed component requirements

wconstab committed 3 years ago

a528b1bb

Add deepspeed support -- WIP

wconstab committed 3 years ago

6e8ca4c7

Fix DDP? runs on 2gpu but need to confirm actually uses DDP

wconstab committed 3 years ago

d9273baa

Refactor ddp.py to make measure() reusable

wconstab committed 3 years ago

02199466

wconstab committed 3 years ago

f7fc86f6

Add arg for running dist benchmarks locally

wconstab committed 3 years ago

72a34904

Add unittest-xml-reporting deps to fix the broken CI (#958)

xuzhao9 committed 3 years ago

52c067fc

Add userbenchmark metrics to the distributed benchmark (#954)

xuzhao9 committed 3 years ago

f30d537c

update regnet and detectron models (#951)

FindHao committed 3 years ago

96938549

Pin numpy version and mkl libraries (#944)

xuzhao9 committed 3 years ago

9bb0bd70

Swap fx2trt_oss to torch_tensorrt (#92)

Wei Wei committed 3 years ago

476ebd74

Support ctr_mbl_feed_30x_over model in TorchBench Internal

xuzhao9 committed 3 years ago

5aad9057

Benchmark aotautograd+nvfuser (#932)

xuzhao9 committed 3 years ago

45a845ea

Fix a few bugs in the userbenchmark abtest script (#940)

xuzhao9 committed 3 years ago

51a70428

Add the MPS device for the runs. (#917)

kulinseth committed 3 years ago

4cc1de9e

Add abtest script for userbenchmark PR testing (#936)

xuzhao9 committed 3 years ago

6200aa1b

Update the config to use a100 runner (#926)

xuzhao9 committed 3 years ago

d7ec7b2b

Pre-download multi30k dataset for attention model (#937)

xuzhao9 committed 3 years ago

8300ff42

Add support for NVIDIA DCGM based FLOPs/sec calculation (#929)

FindHao committed 3 years ago

b5e0b83c

Fix a minor issue in the domain score calculation. (#908)

xuzhao9 committed 3 years ago

8b2dc796

Pin the version of dill for attention_is_all_you_need_pytorch (#922)

xuzhao9 committed 3 years ago

eb672435

Add amp+train support and channels_last support to timm models (#900)

xuzhao9 committed 3 years ago

5a152887

Add required example_inputs argument to prepare_fx and prepare_qat_fx (#77608)

jerryzh168 committed 3 years ago

6f0d2fbd

NVFuser userbenchmarks - use JSON output format (#920)

davidberard98 committed 3 years ago

ae802fdd

Increase the default batch size of alexnet (#919)

xuzhao9 committed 3 years ago

af847a96

Pin the omegaconf package version for xlmr (#918)

xuzhao9 committed 3 years ago

113887cb

Patch torchdynamo to workaround dynamo#159 (#874)

xuzhao9 committed 3 years ago

75ca4de4

Fix fp16 precision and cos sim correctness (#913)

xuzhao9 committed 3 years ago

b81d24d2

Older