Namespace immutable constants per method in `to_native` (#23891) by JakeStevens · Pull Request #23891 · pytorch/executorch

JakeStevens · 2026-10-15T13:52:33Z

Summary:

Exported methods can reuse a constant name with different immutable tensors, causing to_native to reject the combined program. Namespace every immutable constant as <method>/<key>, including single-method programs, and update its graph references, including quantization metadata and nested graphs. Identical tensors still deduplicate when saving the .ptn.

Keys mutated by any method use shared state/<key> names throughout the program, preserving state shared by readers and writers. Update non-persistent buffer bindings and mutation outputs consistently, and reserve the state method namespace. Reject differing values for shared mutable keys and collisions between final names.

Authored with assistance from OpenAI Codex.

Reviewed By: digantdesai

Differential Revision: D125294428

pytorch-bot · 2026-10-15T13:52:37Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/23891

📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 2 New Failures, 1 Unrelated Failure

As of commit 53825b9 with merge base e683360 ():

NEW FAILURES - The following jobs have failed:

pull / unittest-buck / linux / linux-job (gh)
RuntimeError: Command bash /__w/_temp/exec_script failed with exit code 3
pull / unittest-buck / macos / macos-job (gh)
RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 3

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / test-qnn-buck-build-linux / linux-job (gh) (trunk failure)
RuntimeError: Command bash /__w/_temp/exec_script failed with exit code 3

This comment was automatically generated by Dr. CI and updates every 15 minutes.

meta-codesync · 2026-10-15T13:52:42Z

@JakeStevens has exported this pull request. If you are a Meta employee, you can view the originating Diff in D125294428.

github-actions · 2026-10-15T13:53:19Z

This PR needs a `release notes:` label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

Summary: Edge lowering mints `_lifted_tensor_constant{i}` names independently per method, so unrelated scalars (e.g. an fp32 attention scale and an int64 zero) collide under one data key when packaging multiple methods, and `to_native` rejected the package. Namespace only those auto-generated keys per method (`<method>/<key>`) in both the merged graph blob (new `rename_data_keys` in `graph_serialize.py`, walking the same references as `collect_data_keys`) and the constant dicts; identical tensors stay shared, mutated keys are never renamed, and genuine parameter/buffer fqn collisions still raise. Differential Revision: D125294428

Summary: Pull Request resolved: pytorch#23891 Edge lowering mints `_lifted_tensor_constant{i}` names independently per method, so unrelated scalars (e.g. an fp32 attention scale and an int64 zero) collide under one data key when packaging multiple methods, and `to_native` rejected the package. Namespace only those auto-generated keys per method (`<method>/<key>`) in both the merged graph blob (new `rename_data_keys` in `graph_serialize.py`, walking the same references as `collect_data_keys`) and the constant dicts; identical tensors stay shared, mutated keys are never renamed, and genuine parameter/buffer fqn collisions still raise. Differential Revision: D125294428

Summary: Edge lowering mints `_lifted_tensor_constant{i}` names independently per method, so unrelated scalars (e.g. an fp32 attention scale and an int64 zero) collide under one data key when packaging multiple methods, and `to_native` rejected the package. Namespace only those auto-generated keys per method (`<method>/<key>`) in both the merged graph blob (new `rename_data_keys` in `graph_serialize.py`, walking the same references as `collect_data_keys`) and the constant dicts; identical tensors stay shared, mutated keys are never renamed, and genuine parameter/buffer fqn collisions still raise. Differential Revision: D125294428

Summary: Pull Request resolved: pytorch#23891 Edge lowering mints `_lifted_tensor_constant{i}` names independently per method, so unrelated scalars (e.g. an fp32 attention scale and an int64 zero) collide under one data key when packaging multiple methods, and `to_native` rejected the package. Namespace only those auto-generated keys per method (`<method>/<key>`) in both the merged graph blob (new `rename_data_keys` in `graph_serialize.py`, walking the same references as `collect_data_keys`) and the constant dicts; identical tensors stay shared, mutated keys are never renamed, and genuine parameter/buffer fqn collisions still raise. Differential Revision: D125294428

…23891) Summary: Pull Request resolved: pytorch#23891 Edge lowering generates `_lifted_tensor_constant{i}` keys independently for each method. When methods use different scalar values under the same generated key, `to_native` rejects the combined program. Always namespace generated immutable lifted constants as `<method>/<key>`, including single-method programs. Update the constant dictionaries and graph references together through a shared `map_data_keys` traversal, including quantization metadata and nested graphs. Equal tensors under different keys still deduplicate when saving the `.ptn`. Mutable keys keep their existing names, parameter/buffer conflicts continue to raise, and explicit keys occupying a generated namespace are rejected. Authored with assistance from OpenAI Codex. Differential Revision: D125294428

…23891) Summary: Edge lowering generates `_lifted_tensor_constant{i}` keys independently for each method. When methods use different scalar values under the same generated key, `to_native` rejects the combined program. Always namespace generated immutable lifted constants as `<method>/<key>`, including single-method programs. Update the constant dictionaries and graph references together through a shared `map_data_keys` traversal, including quantization metadata and nested graphs. Equal tensors under different keys still deduplicate when saving the `.ptn`. Mutable keys keep their existing names, parameter/buffer conflicts continue to raise, and explicit keys occupying a generated namespace are rejected. Authored with assistance from OpenAI Codex. Differential Revision: D125294428

digantdesai

Review automatically exported from Phabricator review in Meta.

Summary: Pull Request resolved: pytorch#23891 Exported methods can reuse a constant name with different immutable tensors, causing `to_native` to reject the combined program. Namespace every immutable constant as `<method>/<key>`, including single-method programs, and update its graph references, including quantization metadata and nested graphs. Identical tensors still deduplicate when saving the `.ptn`. Keys mutated by any method retain their original names throughout the program, preserving shared state for readers and writers. Reject differing values for shared mutable keys and collisions between final names. Authored with assistance from OpenAI Codex. Differential Revision: D125294428

digantdesai

Review automatically exported from Phabricator review in Meta.

Summary: Exported methods can reuse a constant name with different immutable tensors, causing `to_native` to reject the combined program. Namespace every immutable constant as `<method>/<key>`, including single-method programs, and update its graph references, including quantization metadata and nested graphs. Identical tensors still deduplicate when saving the `.ptn`. Keys mutated by any method use shared `state/<key>` names throughout the program, preserving state shared by readers and writers. Update non-persistent buffer bindings and mutation outputs consistently, and reserve the `state` method namespace. Reject differing values for shared mutable keys and collisions between final names. Authored with assistance from OpenAI Codex. Reviewed By: digantdesai Differential Revision: D125294428

Summary: Add `EXECUTORCH_BUILD_NATIVE_LLM_RUNNER` to build and install `run_hf_llm_vulkan` and `run_hf_llm_ggml` for the enabled engines. Share gflags setup with the vision runner, resolve tokenizers with regex lookahead support, and add `native_llm_runner_utils_test`. Document building and running the GGML CPU LLM runner. Authored with assistance from Codex. Reviewed By: digantdesai Differential Revision: D125479665

Summary: Exported methods can reuse a constant name with different immutable tensors, causing `to_native` to reject the combined program. Namespace every immutable constant as `<method>/<key>`, including single-method programs, and update its graph references, including quantization metadata and nested graphs. Identical tensors still deduplicate when saving the `.ptn`. Keys mutated by any method use shared `state/<key>` names throughout the program, preserving state shared by readers and writers. Update non-persistent buffer bindings and mutation outputs consistently, and reserve the `state` method namespace. Reject differing values for shared mutable keys and collisions between final names. Authored with assistance from OpenAI Codex. Reviewed By: digantdesai Differential Revision: D125294428

Summary: Exported methods can reuse a constant name with different immutable tensors, causing `to_native` to reject the combined program. Namespace every immutable constant as `<method>/<key>`, including single-method programs, and update its graph references, including quantization metadata and nested graphs. Identical tensors still deduplicate when saving the `.ptn`. Keys mutated by any method use shared `state/<key>` names throughout the program, preserving state shared by readers and writers. Update non-persistent buffer bindings and mutation outputs consistently, and reserve the `state` method namespace. Reject differing values for shared mutable keys and collisions between final names. Authored with assistance from OpenAI Codex. Reviewed By: digantdesai Differential Revision: D125294428 Pull Request resolved: pytorch#23891

JakeStevens requested review from JacobSzwejbka, SS-JIA, kirklandsign and larryliu0820 as code owners October 15, 2026 13:52

meta-cla Bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Oct 15, 2026

meta-codesync Bot added the meta-exported label Oct 15, 2026

meta-codesync Bot changed the title ~~Namespace colliding lifted tensor constants in to_native~~ Namespace colliding lifted tensor constants in to_native (#23891) Oct 15, 2026

JakeStevens force-pushed the export-D125294428 branch from 6c40a75 to 7567aa6 Compare October 15, 2026 16:52

JakeStevens force-pushed the export-D125294428 branch from 7567aa6 to 91fd717 Compare October 15, 2026 16:55

JakeStevens force-pushed the export-D125294428 branch from 91fd717 to 75bf79f Compare October 15, 2026 18:16

JakeStevens force-pushed the export-D125294428 branch from 75bf79f to bd4eaab Compare October 15, 2026 18:20

meta-codesync Bot changed the title ~~Namespace colliding lifted tensor constants in to_native (#23891)~~ Namespace lifted tensor constants per method in to_native (#23891) Oct 15, 2026

JakeStevens force-pushed the export-D125294428 branch from bd4eaab to 0bac518 Compare October 15, 2026 19:58

digantdesai requested changes Oct 16, 2026

View reviewed changes

meta-codesync Bot changed the title ~~Namespace lifted tensor constants per method in to_native (#23891)~~ Namespace immutable constants per method in to_native (#23891) Oct 16, 2026

JakeStevens force-pushed the export-D125294428 branch from 0bac518 to 97051f4 Compare October 16, 2026 17:20

JakeStevens force-pushed the export-D125294428 branch 2 times, most recently from f807a6e to fb49267 Compare October 16, 2026 18:27

JakeStevens force-pushed the export-D125294428 branch from fb49267 to 27fff9b Compare October 16, 2026 18:45

digantdesai approved these changes Oct 27, 2026

View reviewed changes

JakeStevens force-pushed the export-D125294428 branch from 27fff9b to b29af96 Compare October 27, 2026 18:40

JakeStevens force-pushed the export-D125294428 branch from b29af96 to 5322a04 Compare October 27, 2026 20:57

JakeStevens added 2 commits October 27, 2026 13:57

JakeStevens force-pushed the export-D125294428 branch from 5322a04 to 53825b9 Compare October 27, 2026 20:58

meta-codesync Bot merged commit 8d7cce8 into pytorch:main Oct 28, 2026
412 of 415 checks passed

Navigation Menu

Repository navigation

Namespace immutable constants per method in `to_native` (#23891) - #23891

Namespace immutable constants per method in `to_native` (#23891)#23891
meta-codesync[bot] merged 2 commits into
pytorch:mainfrom
JakeStevens:export-D125294428

JakeStevens commented Oct 15, 2026 •

edited by meta-codesync Bot

Loading

Uh oh!

pytorch-bot Bot commented Oct 15, 2026 •

edited

Loading

Uh oh!

meta-codesync Bot commented Oct 15, 2026

Uh oh!

github-actions Bot commented Oct 15, 2026

Uh oh!

digantdesai left a comment

Uh oh!

digantdesai left a comment

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

JakeStevens commented Oct 15, 2026 • edited by meta-codesync Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot Bot commented Oct 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/23891

❌ 2 New Failures, 1 Unrelated Failure

Uh oh!

meta-codesync Bot commented Oct 15, 2026

Uh oh!

github-actions Bot commented Oct 15, 2026

This PR needs a release notes: label

Uh oh!

digantdesai left a comment

Choose a reason for hiding this comment

Uh oh!

digantdesai left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

JakeStevens commented Oct 15, 2026 •

edited by meta-codesync Bot

Loading

pytorch-bot Bot commented Oct 15, 2026 •

edited

Loading

This PR needs a `release notes:` label