* fix extract_dataset + tests
* add CI
* sops.gz itself is same as master
* yml + gzip -c + ge
* don't commit that
* bump limit to 1000
* axis=7
* test_tiny
* Don't use numpy inside hlb_cifar10 training loop
* Lint it
* jit it
* Drop the last half-batch
* Use gather for random_crop and reuse perms
* Wrap train_cifar in FUSE_ARANGE context
* No need to pass FUSE_ARANGE=1 to hlb_cifar10.py
* Add cutmix to jittable augmentations
* Remove .contiguous() from fetch_batches
* Fix indexing boundary
---------
Co-authored-by: Irwin1138 <irwin1139@gmail.com>
* change clang -march flag to -mcpu with fp16 disassembly test
* fix
* add capstone to macos dependencies
* just check no cast in test
* rm import
* woops
* lets check
* move check
* llvm init before cpu chcek
* try this
* bump autogen llvm version
* also update libclang?
* revert
* add comment
* skip llvm test and add comment
* linter
* fix
* add early verbose demo test
* is this how to write tests :s
* is definition drift even a thing? gemini says it is
* clean up
* better
* even better
* try add to CI
* doesn't work quite yet
* much more work to be done
* whoops
* partition the test heh
* skipif
* some nits for better names
* add webgpu test for onnxrunner
* fix reference links
* flush for now
* don't run linearize().uop tests in get_action_space test
this part takes 2 minutes in CI and has nothing to do with action space. also not sure if the "for some reason" comment is still relevant
* -n=auto test/models
* move view left to the outer graph
* global view right
* dont need that one
* remove comment
* test kernelize
* simple
* split onnx, test sdxl null
* fix testing
* ugh, wrong one
* Update test.yml
* move high level stuff to unit tests [pr]
* process replay on unit tests
* fix pr, less compute
* set omp num threads
* set 200MB buffer size limit
* delete junk
* fix tests
* faster
* move test_indexing to unit
* faster
* remove del spam from CI
* more
* preconstruct default buffer spec
* ignore those errors
* check exception
* more exception check
* skip stuff
* smaller tests mean faster tests
* a few more
* skip a few slow tests
* use a venv for python packages
* create venv
* no user, it's in venv
* ignore venv
* venv
* new cache key
* try that
* this
* version the python cache
* First version, caught a bug?
* Nicely print failure to reproduce
* Remove that
* Put the assert back
* Change fuzzing to use testing_unit so it has z3
* Test key to match
* Add rule
* Add test
* Add test for edge case 0
* Merge patterns
* update comment
* consistent whitespace
* whitespace
* add condition
* add test
* update comment
* use Variable
* fuzzer using z3_renderer
* Cleaned up printing and debugging
* working new fuzzer
* change some comments and printing
* more formatting
* fuzz failures in seperate file
* fix fstring
* more tests
* naming
* remove added line
* remove comment
* print number of skipped expressions
* use self.assertEqual
---------
Co-authored-by: chenyu <chenyu@fastmail.com>
This mockgpu sqtt emulation will just ignore basically everything and end
up with a 0x1000 size trace full of zeroes, but just testing for things
like register rename is better than nothing i guess
* Add mmapeak implementation for 7900 XTX
* Change identation
* Use a template instead of multiple assebly files
* Fix output formatting
* Reduce register file bank conflicts
* More accurate measurement for quick instructions
* Add support for gfx1201
* RDNA4 wmma requires less VGRPs
* RDNA4 does not have s_cmpk instructions
* Add v_wmma_i32_16x16x32_iu4 for gfx1201
* Add sparse wmma instructions
* split to tinybox red MLPerf Benchmark
---------
Co-authored-by: Panagiotis Kourouklidis <panagiotis.kourouklidis@gmail.com>
hlb cifar is fast so added it, can add bert too if you think it's ok
6 real gpus to test multigraph and transfers + accuracy validation
should probably be added to tinystats too, i don't know how though
Co-authored-by: chenyu <chenyu@fastmail.com>
