pytorch/benchmark

Pull Requests Commits

fix bug for import error when nvml is not installed

FindHao committed 3 years ago

a320f758

add nvml monitor

FindHao committed 3 years ago

0352d6f6

Handle run.py model_analyzer is none (#1297)

aaronenyeshi committed 3 years ago

7ad47ee0

Fix a bug in run.py related to model_analyzer

xuzhao9 committed 3 years ago

7b932305

Align run.py profile_one_step and run_one_step (#1296)

aaronenyeshi committed 3 years ago

022dfe3d

Only check the internal userbenchmarks when the directory exists (#1290)

xuzhao9 committed 3 years ago

f28c72d8

Fix e2e hf_bert model (#1291)

xuzhao9 committed 3 years ago

b35e3d86

Suppress `pretrained` warnings from torchvision (#1280)

davidberard98 committed 3 years ago

2bb181ce

Add resnet152 (#1279)

davidberard98 committed 3 years ago

1168045b

add metrics support for CPU GPU peak memory usage (#1274)

FindHao committed 3 years ago

e9994db3

Correctness Check - Return False instead of assert (#1283)

brad-mengchi committed 3 years ago

eb736f86

Fix the broken main branch with the latest kornia package (#1286)

xuzhao9 committed 3 years ago

48e783a8

add dcgm guidance for sync with upstream (#1284)

FindHao committed 3 years ago

b7ee9565

Add benchmark script for inductor+dynamo vs. eager mode

xuzhao9 committed 3 years ago

ad578741

use cpus_per_task=12 instead of 10 (#1281)

davidberard98 committed 3 years ago

66d60d3c

Allow passing in batch_size as an argument to distributed trainer (#1276)

davidberard98 committed 3 years ago

63d4037c

Update conda installation command to handle the installation difference with cudatoolkit and pytorch-cuda (#1260)

weiwangmeta committed 3 years ago

2fcedc4b

ddp_experiments: disable cudagraphs for inductor & disable static_graph (#1265)

davidberard98 committed 3 years ago

39fb58af

ddp_experiments: improve error handling (#1264)

davidberard98 committed 3 years ago

8fb03251

update dcgm APIs to dcgm 3.0 (#1270)

FindHao committed 3 years ago

1e8a917e

replace env.seed by end.reset(seed=seed) #1271 (#1272)

Delaunay committed 3 years ago

b751e650

Fix the broken bisection workflows (#1267)

xuzhao9 committed 3 years ago

aa594b4a

check and then call opt

brad-mengchi committed 3 years ago

a8bdfa47

Migrate to enable Python 3.10 (#1261)

xuzhao9 committed 3 years ago

94078d92

Reduce profiling runs on distributed models (#1243)

davidberard98 committed 3 years ago

7debab95

add option to control inductor+cudagraph

jspark1105 committed 3 years ago

39b84d2b

Run PR test on GCP A100 runner (#1255)

xuzhao9 committed 3 years ago

90ba2a76

Add gcp a100 dind docker build (#1247)

xuzhao9 committed 3 years ago

d586cf6c

Add regex as a new dependency (#1253)

xuzhao9 committed 3 years ago

5226453a

torchinductor -> torch._inductor (#1254)

davidberard98 committed 3 years ago

5e33c732

Older