Luca Wehrstedt

Allow services to handle many contests at the same time

14

At the moment many services require a contest ID as a command line argument. I think that their use of this ID is very limited (except in some cases like...

An int may not be enough to store the amount of memory used

2

A signed int on 32 bit allows to store the amount of bytes used by a program up to about 2.1GB. In some (very rare) cases this may not be...

I wanted to land a PR on a GitHub repo that I had opened with ghstack. ``` $ sl ghstack land https://github.com/myorg/myrepo/pull/299 ** Sapling SCM (version 0.1.20221201-095354-r360873f1) has crashed: Traceback...

Calling .to(...) on scalar arg breaks when specialization kicks in

2

One of our kernels has a scalar parameter. To avoid an overflow, at some point we cast it to int64 (i.e., we call `v.to(tl.int64)`). On occasion we pass a value...

Allow setting kwargs to special value during autotuning

5

Some of our kernels have side-effects, such as incrementing an extrernal counter. This counter is consumed in "real-time" by a _concurrent_ kernel. It breaks our code if this counter is...

Decide how we round the scores

23

At the moment the ScoreTypes are rounding the score and the max_score (both the public and the private ones) at two decimal places and, where applicable, the subtask score isn't...

Add a Herring-like benchmark

1

Summary: This diff adds two benchmarks that perform an allreduce operation using internally a technique inspired from Amazon's Herring paper. The two versions, one for TCP and one for InfiniBand...

Add test that stresses races during transport shutdown

1

Summary: This test wants to do two things: defer functions to the loop while the context is shutting down, and create new objects (connections and listeners) before and/or just after...

Make SHM support arbitrarily large libnop objects

1

In #58 and #97 we created a fastpath in SHM to serialize and deserialize protobufs directly to and from the ringbuffer. However, due to a limitation in protobuf's zero copy...

TSAN failures in MPT's ContextIsNotJoined test

1

https://app.circleci.com/pipelines/github/pytorch/tensorpipe/1456/workflows/d987d8ee-9b5e-422a-aa38-1c7d5f8fd3f7/jobs/11139