Skip to content

Conversation

@banach-space
Copy link

@banach-space banach-space commented Nov 21, 2025

Bumps torch-mlir release version so that both X86 and AArch64 wheels are
available.

Bumps torch-mlir release version so that both X86 and AArch64 wheels are
available.
@banach-space banach-space force-pushed the andrzej/update_torch_mlir_release branch from 53f61c1 to 1808d04 Compare November 21, 2025 09:13
@banach-space banach-space changed the title andrzej/update torch mlir release Bump torch-mlir version (CPU) Nov 21, 2025
@rengolin
Copy link
Member

CI failure looks weird, nothing to do with the actual test.

@rolfmorel
Copy link
Contributor

This is the issue from #14. As demonstrated there, this is not lighthouse specific. Rather it looks like torch-mlir changed (presumably around how nanobind gets linked). I will have a look at the torch-mlir repo, to try to understand what they did since beginning of October. Have also asked @makslevental if he happens to have any ideas as to what might be going on.

@rolfmorel
Copy link
Contributor

Could you please bump to today's nightly? I can confirm that with the following diff the CI tests all pass for me locally on x86_64:

$ git diff
diff --git a/pyproject.toml b/pyproject.toml
index c9cf1bd..bb4ecac 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -15,20 +15,20 @@ test = [
 # Additional "targets" which pull in optional dependencies -- use `uv sync --extra TARGET`
 ingress_torch_cpu = [
     "torch==v2.8.0+cpu",
-    "torch-mlir==20251003.589"
+    "torch-mlir==20251122.639"
 ]
 ingress_torch_nvidia = [
     "torch==2.8.0",  # Nvidia-enabled version of torch
-    "torch-mlir==20251003.589"
+    "torch-mlir==20251122.639"
 ]
 ingress_torch_rocm = [
     "torch==2.8.0+rocm6.4",  # AMD-enabled version of torch
-    "torch-mlir==20251003.589",
+    "torch-mlir==20251122.639",
     "pytorch_triton_rocm"  # Transitive dependency listed explicitly so that we can state which package repository it is supposed to come from
 ]
 ingress_torch_xpu = [
     "torch==2.8.0+xpu",  # Intel-enabled version of torch
-    "torch-mlir==20251003.589",
+    "torch-mlir==20251122.639",
     "pytorch_triton_xpu"  # Transitive dependency listed explicitly so that we can state which package repository it is supposed to come from
 ]

(I am having a look if there's syntax for deduplicating the torch-mlir version in pyproject.toml.)

@rolfmorel
Copy link
Contributor

rolfmorel commented Nov 22, 2025

Closing as #26 does the same whilst cleaning up pyproject.toml a bit. With #26 merged, we show in #22 that CI is now enabled on x86 and ARM!

@rolfmorel rolfmorel closed this Nov 22, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants