Repository navigation

All pull requests

Arm backend: Lower Aten argmax + INT32 cast -> tosa.ARGMAX
#24156AdrianLundell opened Oct 26, 2026Collaborator
1
Fix -Wunreachable-code-break findings in fbcode/executorch, xplat/executorch
#24155r-barnes opened Oct 26, 2026Contributor
4
Add a backend-free runner_lib for the QNN LLM runner
#24154billmguo opened Oct 26, 2026Contributor
3
Quantize expand_as through the existing shared-qparams expand path
#24153pssrawat opened Oct 26, 2026Contributor
3
exir: Normalize input layouts before lowering view_copy
#24152zingo opened Oct 26, 2026Collaborator
1
Bump PyTorch and TorchAO pins for dev20261025
#24147pytorchupdatebot opened Oct 26, 2026Collaborator
3
[PyBindings] Torch-free follow-ups: reused inputs, error reasons, bfloat16 shape, dtype names, tests, docs
#24146shoumikhin opened Oct 26, 2026Contributor
43
Remove redundant join ownership bookkeeping from batching shutdown
#24145meyering opened Oct 26, 2026Contributor
3
Add skip_h2d_for_constants to keep constants in device memory
#24144shoumikhin opened Oct 26, 2026Contributor
22
Stop installing test-only packages with the executorch package
#24143shoumikhin opened Oct 25, 2026Contributor
10
Fix export script path in backends-mediatek.md
#24142shyhyawJou opened Oct 25, 2026
6
Export yolo26 in strict mode on current ultralytics and raise the pin
#24141Sonica-B opened Oct 25, 2026
7
Give the shape and layout plumbing the qparams of the tensor it moves
#24140pssrawat opened Oct 24, 2026Contributor
2
Migrate runtime consumers to std::string_view
#24139r-barnes opened Oct 24, 2026Contributor
4
Update the TensorRT backend docs page
#24136shoumikhin opened Oct 24, 2026Contributor
39
Bump undici from 6.28.1 to 6.29.1 in /examples/demo-apps/react-native/rnllama
#24135dependabot[bot] opened Oct 24, 2026Bot
2
Bump setuptools from 80.10.2 to 84.1.0 in /backends/arm
#24134dependabot[bot] opened Oct 24, 2026Bot
2
Bump fast-uri from 3.0.3 to 3.1.9 in /examples/demo-apps/react-native/rnllama
#24133dependabot[bot] opened Oct 24, 2026Bot
2
NXP backend: Mark NXP Neutron backend as production ready
#24130robert-kalmar opened Oct 23, 2026Collaborator
5
Arm backend: Keep int64 tensors under the int64 extension
#24128abm-77 opened Oct 23, 2026
66
Let a backend share one delegate across calls of the same lowered module
#24127shoumikhin opened Oct 23, 2026Contributor
18
Implement the portable kernel for avg_pool3d
#24126yenhao-huang opened Oct 23, 2026Contributor
1
Vulkan: lower transposed conv1d onto the conv2d transposed path
#24124msluszniak opened Oct 23, 2026Contributor
1
Implement the portable kernel for avg_pool1d
#24122yenhao-huang opened Oct 23, 2026Contributor
6
Substitute on qparams, not the input fake tensor (#24121)
#24121pssrawat opened Oct 23, 2026Contributor
2

Navigation Menu

Repository navigation

All pull requests

Arm backend: Lower Aten argmax + INT32 cast -> tosa.ARGMAX

Fix `-Wunreachable-code-break` findings in `fbcode/executorch`, `xplat/executorch`

Add a backend-free `runner_lib` for the QNN LLM runner

Quantize `expand_as` through the existing shared-qparams `expand` path

exir: Normalize input layouts before lowering view_copy

Bump PyTorch and TorchAO pins for dev20261025

[PyBindings] Torch-free follow-ups: reused inputs, error reasons, bfloat16 shape, dtype names, tests, docs

Remove redundant join ownership bookkeeping from batching shutdown

Add skip_h2d_for_constants to keep constants in device memory

Stop installing test-only packages with the executorch package

Fix export script path in backends-mediatek.md

Export yolo26 in strict mode on current ultralytics and raise the pin

Give the shape and layout plumbing the qparams of the tensor it moves

Migrate runtime consumers to `std::string_view`

Update the TensorRT backend docs page

Bump undici from 6.28.1 to 6.29.1 in /examples/demo-apps/react-native/rnllama

Bump setuptools from 80.10.2 to 84.1.0 in /backends/arm

Bump fast-uri from 3.0.3 to 3.1.9 in /examples/demo-apps/react-native/rnllama

NXP backend: Mark NXP Neutron backend as production ready

Arm backend: Keep int64 tensors under the int64 extension

Let a backend share one delegate across calls of the same lowered module

Implement the portable kernel for avg_pool3d

Vulkan: lower transposed conv1d onto the conv2d transposed path

Implement the portable kernel for avg_pool1d

Substitute on qparams, not the input fake tensor (#24121)

Pull requests