Guillaume Lagrange
e18a3c2e30
refactor(backends)!: remove associated element types & replace with device defaults ( #5000 )
...
* Remove multi backend router
* Remove associated element types; replaced by device defaults
* Fix wgpu devices
* Fix burn-store
* Fix cubecl backend extensions + device configs
* Missing fixes
* Fix clippy
* Fix wgpu backend feature gate in dispatch
* Fix wgpu/vulkan features for xtask tests
* Fix merge
2026-05-25 09:04:44 -04:00
Nathaniel Simard
8037309704
perf(core)!: improve compile times via opaque inner types to break dependency chain ( #4977 )
...
* Initial commit
* Improve compilation time
* Refactor device
* Device blob
* Improvements
* Autodiff fix
* Burn-std no cubecl dep
* WIP
* Autodiff
* Sanitize outside of generic
* Fix display compilation
* Clippy
* Improve device selection
* WIP
* Fix examples wip
* Fix fmt
* Cleanup
* Fmt
* Fix clippy
* Fix autodiff
* Fix test device
* Fix docs
* Fix tests
* Fmt
* Fix display
* Miri fix
* Improve comments
* Fix device
* Fix docs
2026-05-21 14:08:30 -04:00
Guillaume Lagrange
73b72bc92a
refactor(extension): feature gate Tensor::from/into_primitive and add from/into_bridge w/ TensorKindId validation ( #4961 )
...
* Dispatch default_backend should not be enabled for wgpu_only
* Add extension feature with from/into primitive
* Fix usage in test
* Fix doc
2026-05-15 16:29:55 -04:00
Guillaume Lagrange
d30a82ae78
refactor(tensor): add BridgeTensor to bridge high-level tensor API with dispatch ( #4958 )
...
* Replace tensor primitive type
* Fix into_scalar calls
* Fix clippy
* Fix unused + bounds
* Fix extensions
* Rename to BridgeTensor
* Cargo fmt
2026-05-15 10:59:33 -04:00
Guillaume Lagrange
27964e0160
Refactor tensor kind traits ( #4957 )
...
* Remove generics from tensor kind traits and Elem associated type
* Move backend extension, re-export no types from burn_backend and fix into_scalar missing types
* Remove into/from primitive usage in burn-optim
* Fix backtrace
* Fix float_sort_with_indices int dtype
* Fix feature gated
* Remove dead code
* Add tensor primitive note
* Another note
* High level kind
* Working but still pub traits
* Fix into_scalar doc example
* Fix argsort out dtype
* Restrict ops traits to pub(crate)
* Add other backends for burn-vision
2026-05-14 14:02:44 -04:00
Nathaniel Simard
0c763dcd26
Refactor/isolate burn backend deps ( #4954 )
...
* Wip Tensor module moving
* WIP
* WIP
* Fix
* Fix dependencies
* Cleanup
* Fixes
* Fix docs
* Remove registered reference
* Fix refs
* set_default_dtypes is in burn-tensor
---------
Co-authored-by: Guillaume Lagrange <lagrange.guillaume.1@gmail.com >
2026-05-13 16:26:14 -04:00
Guillaume Lagrange
dbf03c516b
[Breaking] Remove Tensor backend generic and add high-level Device struct ( #4717 )
...
* Remove backend generic from `Tensor` and add high-level `Device` struct
* Fix rfft
* WIP modules
* Cleanup flags
* Carry autodiff intent via checkpointing field
* Fix ctc loss test shape
* Fix imports
* Fix burn-collective & burn-communication
* Working core / nn / optim
* Removed from burn-rl
* Removed from burn-train
* Update lock
* Working mnist
* More examples
* Fix int/bool to_device with float AD device + add enumerate devices
* Working text-classification
* Fixed MultiGradientsParams usage + added gradient checkpointing to trainer
* Updated text-generation example
* Update comment
* Fix merge
* Remove backend generics from merge
* Update lock
* Remove from burn-store
* Fix burn-store dtype tests
* Remove circular dep to burn-tensor from burn-flex
* Cleanup
* Fix burn-store tests
* Fix seed
* Fix imports
* More examples
* Add `backend_extension` proc macro
* Working vision extension
* Missing merge changes
* More merge fixes
* tree
* Whoops
* Cleanup
* Not feature gated
* Fix wgpu kernel example
* Small change
* Update lock
* Cleanup
* Cargo fmt
* Clippy
* Remove burn-collective
* Cleanup
* Fix burn-store import
* Fallback to wgpu for workspace builds & blanket fusion impl
* Add `default_backend` fallback for burn-dispatch
* Gate unused var autodiff
* Fix no-std
* Refactor new tests
* Fix more tests
* Adjust ctc_loss backward f16 tolerance
* Fix test device for burn-core
* Fix test device
* Fix some docs
* Default to ndarray for now due to burn-flex CAS limitation
* Fix docs + ndarray default
* Add portable_atomic_util::Arc in burn-autodiff
* Add optim feature flag because it requires autodiff (fix no-std; not supported on all targets)
* Add Gradients high-level wrapper
* Fix types
* Fix test device
* Fix docs
* Fix burn-train auto merge
* Fix clippy
* WIP burn device override
* Fix dtype usage
* Remove test-metal
* Fix CI test flags
* Remove burn-train vision on metal
* Small note
* Missing metal feature for macos CI
* Fix docs
* Remove backend ops note
* Change set_default_dtypes to &mut self
* Fix mut device
* Fix burn-vision backend features
* Exclude burn-backend-tests from workspace tests (explicitly handled)
2026-05-13 10:03:12 -04:00
Dilshod Tadjibaev
9a0ba1d81c
Centralize internal burn-* deps in [workspace.dependencies] ( #4876 )
...
Consumers now say `workspace = true` instead of repeating
`path = "../burn-X", version = "=0.21.0-pre.3"` 180+ times.
Workspace deps pin `default-features = false` because Cargo forbids
consumers from overriding `default-features` when inheriting. Consumers
that previously enabled defaults now opt in with `features = ["default"]`.
2026-04-27 08:13:24 -04:00
Genna Wingert
fce1a57211
refactor: Vector size generic ( #4624 )
2026-03-13 08:36:34 -04:00
Nathaniel Simard
3c6b710dca
Refactor/device handle ( #4593 )
2026-03-11 18:46:49 -04:00
Genna Wingert
66733ae0eb
refactor: Metadata type/strides refactor ( #4534 )
...
* Remove R::supported_line_sizes
* refactor: Metadata optimization
* Revert temp fix
* Rename `ShapeError` to `MetadataError`
* Cleanup
* Bump cubecl and cubek rev
* Fix doc test
* Bump cubecl rev
2026-02-19 10:07:37 -05:00
Florian Liao
51811d5563
chore: fix typos caught by xtask ( #4406 )
2026-01-30 07:38:12 -05:00
Genna Wingert
7b32614f16
refactor: Migrate to usize indexing ( #4273 )
...
* Migrate to `usize` indexing
* Fix merge
* Update cubecl
* Update cubek
2026-01-08 12:53:31 -05:00
Nathaniel Simard
19db09c968
Feat/error handling cubecl ( #4076 )
2025-11-27 11:05:09 -05:00
Nathaniel Simard
cc3ee1e0ef
Chore/update dtypes ( #3998 )
2025-11-10 11:09:30 -05:00
Guillaume Lagrange
01912f80e2
Refactor Shape manipulations ( #3845 )
...
* Shape shape shape
* Add shape indexing/deref/as_slice...
* Add shape methods and refactor to use them
* Add shape cat + more refactor
* Add shape slice
* Cargo fmt
* More fmt
* Fix clippy
* Remove dead code
* Matmul output
* Remove comment
2025-10-09 09:37:17 -04:00
Nathaniel Simard
742e5ae1d1
Fix/autodiff/multi threads ( #3793 )
2025-10-03 11:38:46 -04:00
Crutcher Dunnavant
f5198e5e6f
Chain lint inheritance [was: Disable new default clippy tests] ( #3200 )
2025-05-20 08:23:11 -04:00
Guillaume Lagrange
17e3167a70
Improve test tolerance assertions ( #3024 )
...
* wip new tolerance
* use the new assert_approx_eq
* fix or tag broken tests
* Fix more tests
* All tests pass cuda/wgpu
* Format
* Fix other modules tests tolerance
* Fix onnx import tolerances
* Remove broken comments
* Expected values are not defined with enough precision anyway
* Less strict tolerance for tch
* Fix docstring
* More tolerance
* Revert to 1e-1 abs (previous tolerance, probably set for tch)
* Tolerances againnnn
* Fix review
* relax some tolerances
---------
Co-authored-by: maxtremblay <t.maxime@pm.me >
2025-04-16 13:33:05 -04:00
Guillaume Lagrange
303a83e909
Update to edition 2024 ( #2931 )
...
* Upgrade to edition 2024
* Explicit unsafe
* Fix keywords and unnecessery ref modifiers
* Fix lifetime
* Update lock
* Cargo fmt
* Update CI prev
2025-03-20 10:28:51 -04:00
Nathaniel Simard
d9e41460ff
Refactor burn jit => burn-cubecl ( #2809 )
2025-02-13 12:39:29 -05:00
nathaniel
3ed38cfcb9
Rename crate
2025-02-13 11:59:09 -05:00
Maxime Tremblay
29c383b87d
Replace return with terminate ( #2742 )
...
* replace return with terminate
* bump cubecl
* cargo fmt
2025-01-27 09:57:28 -05:00
Genna Wingert
42e7c1f225
[Feat] 8-bit bool for JitBackend ( #2526 )
2024-11-29 09:21:13 -05:00
Guillaume Lagrange
9a2b841665
Add float cast op for JIT backend ( #2511 )
...
* Fix backend FloatElem docstring
* Remove elem type generic from JitTensor (float ops still using default dtype)
* Use execute_with_dtype macro for float ops
* Add cast for fusion
* Add warning for type promotion
* Add cast to backend router
* Add Primitive trait to correctly display the tensor dtype
* Fix primtiive associated type
* Fix jit bool tensor display
* Fix CI
* Fix clippy
* Whoops
* Fix candle bool tensor display
* Add shape to primitive trait
* Add primitive tests
* Rename trait to TensorMetadata
* Add missing changes to example
* Fix split from merge
2024-11-21 13:19:33 -05:00
Genna Wingert
d5e8e3185c
Add documentation for custom cubecl kernels, update some outdated docs ( #2404 )
2024-10-25 13:22:23 -04:00
