Brian Coutinho
Brian Coutinho
Add a few test cases to verify newly added NCCL metadata in profiler events The test looks at the following blocks record_param_comms ``` { "ph": "X", "cat": "cpu_op", "name": "record_param_comms",...
## Summary Currently the memory profiler feature in PyTorch is available via the [profiler API](https://pytorch.org/tutorials/recipes/recipes/profiler_recipe.html#using-profiler-to-analyze-memory-consumption) by passing `profile_memory=True` in the interface. It is desirable to also enable memory profiling using...
Summary: There is an ODR one definition rule violation that was causing a crash on sigrid https://fb.workplace.com/groups/560979627394613/posts/2909061125919773/?comment_id=2909752389183980&reply_comment_id=2910102119149007 Sigrid includes both kineto and ipcfabric via dynolog, and on kineto the class...
Related to an issue we saw on FAIR research cluster, some of the compille time flags were not set as expected. This change prints them so we can easily debug...
## TLDR Dynolog provides system telemetry at Meta as well as in open source environments. Metric logging using Prometheus - an industry standard framework for logging/exporting metrics. This can also...
Fixes #125272 ## About (This is a re-spin of PR #106617) Kineto introduced a new profiler to read performance counters from NVIDIA GPUs (CUPTI Range Profiler API) added in PR[75616](https://github.com/pytorch/pytorch/pull/75616)....
# Overview Originally, this was part 1 of splitting PR #1148. It supports a new kind of GPU Counter events that will be published to the timeline as a time...
# Enabling dynamic profiler plugins (Feature Proposal) Authors: @yisitu, @zli669, @briancoutinho (NVIDIA) ## TLDR * We would like to contribute a feature to Kineto to enable the plugging in of...