* fix(onnx): unwrap list/tuple value in Pad op
* add regression test for Pad list value
* remove trailing whitespace
* use _resolve_const for Pad constant_value
* use less flaky metric for whisper long transcription
* multiline long transcription 3 reference
* fix reference transcript
see https://homepage.ntu.edu.tw/~karchung/miniconversations/MC.htm
sanitized for whisper
* try lower wer threshold
* add test for wer metric
* extract TRANSCRIPTION_3_ALT
* rename test
* rename
* add tests for high WER difference
* move tests
* sync metric
* rtoposort is fast, can replace rangeify with this
* fast rangeify
* work
* fast rangeify works for mnist
* should work
* progress
* pad fix
* FAST
* tests passing
* don't delete those shape ops
* put in rangeify map
* ending ranges fix
* tests
* mstack/mselect no hacks
* move to indexing.py
* touch up tests + add comments
* disable failing test
* actually make the file readable
* failing
* error
* move device tests to test/device
* test speedups
* test device
* linalg to unit
* upd
* so pytest just works
* more divide and skip
* speed
* test devectorize
* add pillow
* BOOM
* cache extra/huggingface/models/
* why max buffer size is not 0
* override MAX_BUFFER_SIZE
* less models
* remove more models and change cache dir to already cached dir
* only metal
* less is more?
* remove check ops
* why is this not setting the ENVVAR
* ughhhhh just test in models
* only cpu and gpu
* only cpu actually
* just override it idk
* final
* move extra dependencies up top
* simplification
* fix print
* make README better
* revert ops_disk fix for now
* clean up test_onnx
* remove testing fashion clip model cuz sloooowwwwww
* actually let METAL run this
* fix comment mistake
* fix download path in run_models
* does this work?
* cleanup setup and teardown
* contextvar like this?
* prove model is cached
* do I need to increment DOWNLOAD_CACHE_VERSION?
* see if cached with incremented DOWNLOAD_CACHE_VERSION
* use warnings to see if the model exists
* revert DOWNLOAD_CACHE_VERSION stuff and clean up
* add retry to download
* nit
* file path as input and have parse be in OnnxRunner.__init__
* modelproto_to_onnxrunner -> modelproto_to_runner
* whoops, fix import
* oh flakiness again, is it because it's getting gc-ed?
* small changes
* CI flaky so just move compile4 fix in
* copy typing of onnx_load
* actually can just import onnx_load instead of onnx.load
* fix external_benchmark_openpilot
* fix onnx_runner test to use onnx_helper
* rerun CI
* try run_modelproto
* spam CI a few times
* revert run_modelproto since that's flaky also
* no external onnx_load usage except onnx.py
* cursor tab complete is evil. Snuck a darn sorted in. But does order change result? Why?
* model_benchmark 193s -> 80s, add OnnxRunner.to()...
* minimize diff and clean up
* device can be None, weird but eh
---------
Co-authored-by: chenyu <chenyu@fastmail.com>
* 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
* this
* clean up
* more clean ups and improve debug msg
* more correct training toggler
* remove manual training toggling
* change some variable names
* actually just add the training toggle for LIMIT envvar too
* more refinement
* __call__ and OnnxRunner
* fix half pylint, other half is importing from onnx while this file is onnx.py, figure out later
* ahhhh found another mistake
* remove limit from __call__
---------
Co-authored-by: chenyu <chenyu@fastmail.com>
* validate that FC exists before loading pretrained weights
* add test case for ResNet pretrained model without FC layer
* remove extra newline
* rename test case
* reraise exception if not handled by check
the test was useless because it was looking at the jit graph counts. wrap with JIT=2 for now.
if it's stable we could consider making kernel count strict, which helps change like #7940
