Skip to content

Update libtorch to 2.9 #1636

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 5 commits into from
Oct 18, 2025
Merged

Update libtorch to 2.9 #1636

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
03bb18c
Update libtorch to 2.9
stewartj3 Oct 18, 2025
6279bb0
Replace README supported CUDA versions with table
stewartj3 Oct 18, 2025
e4a1bf6
Update parsing for libtorch versions to select correct cuda versions
stewartj3 Oct 18, 2025
192f14a
Add CUDA to the version numbers on the tables
stewartj3 Oct 18, 2025
18fa6de
Apply suggestion from @polvalente
polvalente Oct 18, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion torchx/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,7 @@ CUDA versions. The current supported targets are:
- `cu118` CUDA 11.8 and CPU version (no macOS support, libtorch `< 2.8.0` only)
- `cu126` CUDA 12.6 and CPU version (no macOS support)
- `cu128` CUDA 12.8 and CPU version (no macOS support)
- `cu129` CUDA 12.9 and CPU version (no macOS support, libtorch `>= 2.8.0` only)
- `cu130` CUDA 13.0 and CPU version (no macOS support, libtorch `>= 2.9.0` only)
Copy link
Contributor

@polvalente polvalente Oct 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
- `cu128` CUDA 12.8 and CPU version (no macOS support)
- `cu129` CUDA 12.9 and CPU version (no macOS support, libtorch `>= 2.8.0` only)
- `cu130` CUDA 13.0 and CPU version (no macOS support, libtorch `>= 2.9.0` only)
- `cu128` CUDA 12.8 and CPU version (no macOS support)
- `cu129` CUDA 12.9 and CPU version (no macOS support, libtorch `>= 2.8.0 && < 2.9.0` only)
- `cu130` CUDA 13.0 and CPU version (no macOS support, libtorch `>= 2.9.0` only)

Copy link
Contributor Author

@stewartj3 stewartj3 Oct 18, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

With the constant updates in supported CUDA versions it may be easier to display this a a table with torch version and support CUDA versions as the columns.

torch supported CUDA versions
2.7.0 cu118, cu126, cu128
2.8.0 cu126, cu128, cu129
2.9.0 cu126, cu128, cu130

Copy link
Contributor

@polvalente polvalente Oct 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I like that idea

Copy link
Contributor Author

@stewartj3 stewartj3 Oct 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Updated the readme with the new table.


Once downloaded, we will compile `Torchx` bindings. You will need `make`/`nmake`,
`cmake` (3.12+) and a `C++` compiler. If building on Windows, you will need:
Expand Down
10 changes: 5 additions & 5 deletions torchx/mix.exs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -75,12 +75,12 @@ defmodule Torchx.MixProject do

defp libtorch_config() do
target = System.get_env("LIBTORCH_TARGET", "cpu")
version = System.get_env("LIBTORCH_VERSION", "2.8.0")
version = System.get_env("LIBTORCH_VERSION", "2.9.0")
env_dir = System.get_env("LIBTORCH_DIR")

# 2.8.0 is the first version that supports cu129 and drops cu118
# cu118 might still be needed for older hardware, so we're keeping it
# for now.
# cu118 support was dropped in 2.8, which might still be needed for older hardware,
# so we're keeping it for now.
# cu129 was supported in 2.8, but replaced with cu130 in 2.9.
valid_targets = ["cpu", "cu118", "cu126", "cu128"]

valid_targets =
Expand All @@ -89,7 +89,7 @@ defmodule Torchx.MixProject do
if Version.match?(parsed, "<= 2.7.0") do
valid_targets
else
(valid_targets -- ["cu118"]) ++ ["cu129"]
(valid_targets -- ["cu118"]) ++ ["cu130"]
end
Copy link
Contributor

@polvalente polvalente Oct 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
if Version.match?(parsed, "<= 2.7.0") do
valid_targets
else
(valid_targets -- ["cu118"]) ++ ["cu129"]
(valid_targets -- ["cu118"]) ++ ["cu130"]
end
cond do
Version.match?(parsed, "< 2.8.0") ->
valid_targets
Version.match?(parsed, "< 2.9.0") ->
(valid_targets -- ["cu118"]) ++ ["cu129"]
true ->
(valid_targets -- ["cu118"]) ++ ["cu130"]
end

Copy link
Contributor

@polvalente polvalente Oct 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

maybe even just have explicit lists in each clause

Copy link
Contributor Author

@stewartj3 stewartj3 Oct 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Having a supported CUDA list for each version (or when libtorch changes the supported CUDA versions) would be easy to maintain. These only change every few months and are easy to update when adding a new version.


_ ->
Expand Down