Fast path slice contiguous constant #137
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
smjleo
added a commit
that referenced
this pull request
Oct 18, 2024
* Simplify path infra (#128) * Simplify path infra * fmt * add utils * Bump internals (#130) * Bump internals * fix * fixup * fix * get out hlomodule from wrapper module * construct analysis with test gpu specs and broken dlopen because some random dependency found its way in * fix build * actually compute non-zero costs * use analytical cost model * dedup dependency for gpu remove debug output * Convolution support * adding more multi patterns * Attempt gpu ci fix (#125) * Try local python usage * Attempt gpu ci fix * fix * fix * fix * Update llama.py * try cuda 12 pip * fix * fix libdevice * cleanup * Force gpu * Update gpu_pipeline.yml * Update gpu_pipeline.yml * Update gpu_pipeline.yml * Update test_utils.py * Update gpu_pipeline.yml * Update gpu_pipeline.yml * Update gpu_pipeline.yml * Update gpu_pipeline.yml * Update gpu_pipeline.yml * Update gpu_pipeline.yml * Update gpu_pipeline.yml * Update gpu_pipeline.yml * Update gpu_pipeline.yml * Update gpu_pipeline.yml * Update test_utils.py --------- Co-authored-by: William Moses <wsmoses@cyclops.juliacomputing.io> * formatting and fix multi pattern rewrite * add debug outputs * try infering device description (broken on cpu, maybe it works on gpu) * remove debug output for device desc * clean up cost model, and make it depend on platform * Adding resnet test and eqsat after jvp * hugging face transformers vision transformer and resnet added * Fast path slice contiguous constant (#137) * Fast path slice contiguous constant * fmt * remove unneeded code in resnet * fixed conv rewrite * fixed next multi-pattern for conv * turn off fusion costs by default * build python 3.11 instead * add batched matmul multi-pattern rewrite * merging rewrites * Adding back deleted * bert huggingface added * add matmul multi-pattern rewrite with different lhs contracting * Add jaxmd tests (#136) * Add jaxmd tests * fix * fix * fix * fix * fixup * fmt * More ops * fix * fixup * fixup * fix * cleanup * fix * fix * fix * fmt * fix 312 * add reduce test * red * now as test * fix * cleanup * jaxmd disable AD * remove extra print * no tpu * Scatter activity test * fix * fix * fix * cleanup * cleanup * dynamic slice fwd * dynamic update slice fwd * Fix * remove pyc * add pyc * fix * fix * fix * fix * fix * fastsmpl * fix * cleanup * cleanup * fix * final fix * fix broken asserts * fix contracting 2 rewrite * adding mistral test * Transpose batch (#138) * ci: touch lock file * try fixing ci * turn off visibility checks * try removing dependency? * try adding redzone allocator * changes * try adding config cuda * bump internals * bump again * use python 3.12 * remove crosstool in cuda config * Maxtext (#139) * Maxtext * fixup * fix * try fix * fix * fix * cleanup * Update test-requirements.txt * cleanup * don't build rocm * bring back crosstool * add a bunch of bazel flags * turn off tf nvcc clang * device description fix * update cuda/cudnn versions * try printing libs * revert * does nvrtc work as deps? * expose multi pattern rules * put python in path * add ortools as dep * add eqsat env var for test * ortools where are you??? * descent into madness * please * Update StableHLOAutoDiffOpInterfaceImpl.cpp * a * cleanup, and hopefully make it build on cyclops again * get results csv from the right place * add eqsat to maxtext test * csv location * get csv --------- Co-authored-by: William Moses <gh@wsmoses.com> Co-authored-by: jbachurski <kbachurski@gmail.com> Co-authored-by: Arya Vohra <arya.vohra04@gmail.com> Co-authored-by: William Moses <wsmoses@cyclops.juliacomputing.io>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.