* Metal shaders for efficient self attention on large sequences
Updated fast attention: GEMM-ified with Steel primitives
Uses flash attention 1 for scale correction
* more compiler silencing
* Address rebase issues
* Templatize kernel instantiation, revise cpu bindings
* Safer writes to output
* Permit batch size > 1
* Numerical fixes for sdpa self attention
* Re-enable test, remove unused variable
* add benchmarking script
* Disable sdpa prior to perf tuning, and simplify tests for per-patch CI
* Not sure if this is correct
* Format
* Edit tests
* Add negative test
* Format
* add one more test
---------
Co-authored-by: Awni Hannun <awni@apple.com>
* add synchronize function
* fix linux
* fix linux
* fix and fix docs
* fix test
* try synchronize in stream destroy
* synchronize works for both cpu and gpu
* more async eval
* fix rebase
* try correct async eval
* fix async
* more tests for async eval
* use shared events for synchronization
* comment + cleanup
* with autorelease pool
* fix no metal build
* fix compile
* fix patch
* don't eval if asyn evale'd
* don't use is_evaled
* comments
* more multi stream tests
* try and cleanup use of is_evaled
* use a status flag
* std and expm1
* actually add expm1
* fix linux
* fix vjp
* relax tol for linux test
* Add it to the compilable primitives
---------
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>
* add implicit conversion of list to array for equality constraint
* add tests for array equality
* add test for tuple and array equality
* return False if __eq__ arg is list or tuple
* write tests for equality
* update the rule of comparison for __ge__/__gt__/__lt__/__le__
* add a helper function for detecting mlx.core.array
* return true in case fo inequality
* debug minor issue regarding detecting mlx array
* add tests for inequality comparisons
* add name for contribution
* reformat files using pre-commit
* update tests for float
* update tests for inequality
* raise exception in case of invalid comparisons
* use isinstance instead of string comparison
* replace "is_convirtable_to_array" with previous logic
* remove throwing exceptions for other operations
* just a comment
* minor changes for efficiency
* optimize a utils function
* change the function name
* Update ACKNOWLEDGMENTS.md
---------
Co-authored-by: Awni Hannun <awni.hannun@gmail.com>
* add numeric type hierarchy and issubdtype as well as a set_dtype method to nn.Module with predicate
numeric type hierarchy and issubtype is compatible to the [numpy hierarchy](220f0ab2c5/numpy/_core/numerictypes.py (L42)).
Closes#285.
* nits in docs
* unify type category checking
* nits in docs
* nits in docs
* more docs nits
* fix callable type
---------
Co-authored-by: Awni Hannun <awni@apple.com>
* some small overhead improvements
* use result_type in rms_norm
* remove release force
* fix + use non-vector version
* revert compile change
* fix ops
* a little more overhead
* a little more cleanup and overhead
* fast rmsnorm
* no rms gpu
* kernel
* fix shared mem
* looped rms and donation in softmax
* Make the squaring in float32 to avoid underflow
* Fix the default StreamOrDevice for rope and rms_norm in fast
* nits
---------
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>
* Update mlx_set_item to handle regular slices without expanding
* Refactor ellipsis handling
* Route mlx_set_item to slice_update where possible
* Update mlx_scatter_args_slice
* Don't route to gather if no array indices
* mostly builds
* most tests pass
* fix circle build
* add back buffer protocol
* includes
* fix for py38
* limit to cpu device
* include
* fix stubs
* move signatures for docs
* stubgen + docs fix
* doc for compiled function, comments
* Fast Inference SDPA op
Implements metal shaders for:
o = mx.fast_inference_sdpa(queries, keys, values, scale, mask)
Supports fp16, fp32 dtypes; assumes d_k = 128.
Generic op support / prompt encoding supported via mlx primitives.
Metal implementation is for the inference use case only.
Majority of performance benefits appears to results from GQA & reduced
bandwidth requirements; there is approximate performance parity for the
MHA use case (from some measurements on M3 Max).
* Flush shared memory to zero before unprotected reads for (scores @ values)
* Move to fast:: namespace, address reviewer comments
... also attempt to revert formatter auto-change for files not relevant
to this change
* Shared memory flush to top of kernel
* Resolve compiler warnings
* Update python/src/fast.cpp
Co-authored-by: Awni Hannun <awni.hannun@gmail.com>
* Update python/src/fast.cpp
Co-authored-by: Awni Hannun <awni.hannun@gmail.com>
* Update python/src/fast.cpp
Co-authored-by: Awni Hannun <awni.hannun@gmail.com>
* Update python/src/fast.cpp
Co-authored-by: Awni Hannun <awni.hannun@gmail.com>
* Update docstring per PR feedback
* Softmax in higher precision, ...
* route to fallback for more use cases - batch size > 1, head_dim other
than 128, etc.
* Address linux build failure
* Address other reviewer comments
* Remove extraneous eval_cpu function per review
---------
Co-authored-by: Atila Orhon <64497909+atiorh@users.noreply.github.com>
Co-authored-by: Awni Hannun <awni.hannun@gmail.com>
Co-authored-by: atila <atiorh@icloud.com>
* refactor tree utils
* fix compile + tree code refactor
* Add an extra test
* add a few missing activations to docs
* hash structure
* Encode the full argument structure
---------
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>
* shapeless compilation for some graphs
* update compile benchmark
* default compile a few activations
* buffer donation
* bugfix
* shapeless fix
* update tests to work for cpu and gpu fusion
* test kwargs
* add kwargs to compile
* Recompile when python arguments change
* no compile for tanh
* some constant tests
---------
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>
* Add a few LR schedulers
* Move parents's constructor call to the top
* Fix docstring
* refactor optimizers into two files
* add docs
* nit
* Fix Callable type annotation for python 3.8
---------
Co-authored-by: Awni Hannun <awni@apple.com>
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>
* extensions start
* rope custom op
* fix build
* docs + rope benchmark
* fix test
* Add a Metal kernel for RoPE
* Fix position of traditional
* transform tests
* Move rope computation to float and fix tests
* Fix the test and a typo
* change to fast
* fix no metal build
---------
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>
* Simple kernel generation
* Remove the generate kernel from graph_utils
* fix multi-output with compile
* fuse with stopgrad
* v1 input, output capture in compile
* cleanup tree update with visitor update
* nit
* remove todo
* state for model, optional explicit init and more pure optimizer steps
* move learning rate to state
* add lr to opt state, some fixes in capture
* fix optim
* update tuple of containers as well
* fix stream for compiled output
* rng state for compile
* nit
* updates and comments
---------
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>
* Implement custom_vjp and checkpointing
* Add a dependency management primitive
* Change the eval order to deep branches first
* Add graph depth tracking to the array
