Create README for distributed benchmark (pytorch#1183)

Summary:
Add readme so distributed benchmark is easy to run and understand.

Pull Request resolved: pytorch#1183

Reviewed By: xuzhao9

Differential Revision: D39559898

Pulled By: erichan1

fbshipit-source-id: 2a77a72e3f03dd5acb2ad388412a0a1ef65e0a64

userbenchmark/distributed/README.md

@@ -0,0 +1,11 @@
+This is a benchmark for measuring PyTorch Distributed performance. 
+
+An example run command. Results are outputted as a json file in --job_dir FOLDER.
+```
+python run_benchmark.py distributed --ngpus 8 --nodes 1  --model torchbenchmark.e2e_models.hf_bert.Model --trainer torchbenchmark.util.distributed.trainer.Trainer --distributed ddp --job_dir $PWD/.userbenchmark/distributed/e2e_hf_bert --profiler False
+```
+Supported options (not-exhaustive):
+* --distributed {torchbenchmark.e2e_models.hf_bert.Model, torchbenchmark.e2e_models.hf_t5.Model}  
+* --distributed {ddp, fsdp, deepspeed, none}
+* --profiler {True, False}
+  * If set to True, returns one trace for every GPU, saved into --job_dir FOLDER.