holsee/chroxy 177
Headless Chrome as a Service
Expert System to Categorise Beer (in Elixir)
belfastfp/Belfast-Kats-Katas 5
Functional Kats Belfast meetup Kata solutions
Client application for interacting with the Chroxy Headless Chrome Proxy Service
Programming Exercise that covers: storing state, functions over that state and Console IO
belfastfp/belfastkats.github.io 1
katsconf.com website
Code Kata: Countdown Numbers Game Solver
Solutions to advent of code 2018
Ruby algorithms and data structures. C extensions
push eventelixir-nx/nx
commit sha c72b036483b5d63226aaf502f27077adfaf40693
Ensure outputs are correct
push time in 5 hours
push eventelixir-nx/nx
commit sha e6495fe066f6e870440524e7b2820f1cad20e3af
Fix bugs
push time in 6 hours
startedmbuhot/chinook
started time in 7 hours
starteddmjio/stripe
started time in 7 hours
push eventelixir-nx/nx
commit sha a8a75db624c877ddc384aeb3bca339d98521f4d1
Remove debug log
push time in 7 hours
PR opened elixir-nx/nx
Implements host callbacks for intermediate values on EXLA CPU. I can mess around with a GPU version later. I'm open to change the API in any way.
pr created time in 7 hours
The Windows App SDK empowers all Windows Desktop apps with modern Windows UI, APIs, and platform features, including back-compat support, shipped via NuGet.
fork in 10 hours
startedslundberg/shap
started time in 11 hours
startedneedle-tools/shadergraph-markdown
started time in 19 hours
push eventbastion-rs/bastion
commit sha 627646b084634074d828f3f804014f9e5c0ed4a6
wip, lotable backed global state
push time in 19 hours
push eventelixir-nx/nx
commit sha f820a3564028157c341a35a7caeb2377a42fa160
Fix same padding window aggregate bug (#428)
push time in 19 hours
issue closedelixir-nx/nx
Window max/min fail for f64 input types on binary backend
The issue is somewhere in pad. Two tests that fail (they raise, no result is returned):
test "window max f64" do
t = Nx.iota({3, 3}, type: {:f, 64})
assert Nx.window_max(t, {2, 1}) == Nx.tensor([[3, 4, 5], [6, 7, 8]], type: {:f, 64})
end
test "window min f64" do
t = Nx.iota({3, 3}, type: {:f, 64})
assert Nx.window_min(t, {2, 1}) == Nx.tensor([[0.0, 1.0, 2.0], [3.0, 4.0, 5.0]], type: {:f, 64})
end
closed time in 19 hours
seanmor5push eventelixir-nx/nx
commit sha 0c228dc4a0f5093863358ea0ddeb2c3406071473
Fix lack of ordering on maps in defn
commit sha 3f733c6b9135d9a2b6b17f585b8f3bf76b675156
Merge branch 'main' into sm-window-aggregate-bug
push time in 19 hours
startedgoogle/leveldb
started time in a day
push eventbastion-rs/bastion
commit sha 871ed16d06d5e9a712ef54c1ba4a86aa33dc5ddb
Implemented reading and writing value in global state
push time in a day
push eventrusterlium/rustler
commit sha 25f01d3e10626f39e820df16c08642b0420b75fa
Rename release.sh -> prepare_release.sh
commit sha f02fa08c0e1d83b7d3e2f9c604979eff63178d2b
Do not attempt to publish in script
commit sha d2391a76b32f120d9c2a8d05464f57d474aa68d7
Update RELEASE_CHECKLIST.md
commit sha c882c95aed3eed5cdbf193a8e44c46ef60ad6be8
Merge pull request #369 from evnu/prepare_release Replace release.sh with prepare_release.sh
push time in a day
PR merged rusterlium/rustler
After preparing a release (bumping version numbers, etc.), manually pushing the crates and packages simplifies error handling if one of the release operations fails.
pr closed time in a day
Pull request review commentrusterlium/rustler
Replace release.sh with prepare_release.sh
mix deps.get mix compile popd -echo-echo "This script will run:"-echo "rustler_mix $ mix hex.publish"-echo "rustler_codegen $ cargo publish"-echo "rustler $ cargo publish"-echo " $ git push"-echo- read -p "Everything OK? [yN] " -n 1 -r-echo-if [[ $REPLY =~ ^[Yy]$ ]] && [[ -z $DRYRUN ]]; then- # At this point, we cannot reliably revert on errors anymore, as we might- # have published below already.-- cannot_revert() {- echo "Errors detected, but cannot revert."- }-- trap cannot_revert INT EXIT-- # Update and publish- pushd rustler_codegen- cargo publish- popd- pushd rustler- cargo publish- popd- pushd rustler_mix- mix hex.publish- popd-- git push- git push origin "$TAG"-- trap "echo done" EXIT-fi++trap "" EXIT++cat <<_++To release, run the following manually:++cd rustler_codegen; cargo publish+cd rustler; cargo publish+cd rustler_mix; hex.publish
Typo, missing mix
comment created time in 2 days
issue commentelixir-nx/nx
Add mechanism for optional backend functions
I propose adding a custom_call/2 metadata node or similar which looks something like:
custom_call(:target, fallback_impl_fn)
Then compilers and backends can register custom calls for specific targets, during evaluation and compilation those special targets are dispatched to the registered custom call implementation. In the backward pass it is just metadata so the gradient is still valid.
This would be useful in defn as well. For example in Axon some functions or layers might have specialized implementations. BatchNorm for example has an XLA specialized implementation which is faster than the one that is generated from the code. Torchx would benefit as well because things like softmax and other activations would not need to fall back on the evaluator
comment created time in 2 days
PR opened rusterlium/rustler
After preparing a release (bumping version numbers, etc.), manually pushing the crates and packages simplifies error handling if one of the release operations fails.
pr created time in 2 days
issue closedrusterlium/rustler
Integrate cargo tests into Elixir test build.
It would be great if we could integrate cargo tests into the mix test task. If this is already there, pardon me but I have looked for docs.
closed time in 2 days
ghostissue commentrusterlium/rustler
Integrate cargo tests into Elixir test build.
Closing, there hasn't been activity here in a long time and it is not yet clear how to proceed. Feel free to reopen if this changes!
comment created time in 2 days