Merge improvements of 0.7.1b release into main #46
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
There are multiple compatible variants of gfx942, e.g., gfx942:sramecc-:xnack- and gfx942:sramecc+:xnack-
Major changes: 1. Fix numerical errors from scaling input tensors with log_2(e) as preprocessing. Fudge factors are adjusted accordingly 2. Adopt techniques from forward kernel to specialize inner loops of the bwd kernel as well. 3. Update the tuning database for MI200/300 accordingly Minor changes: 1. `pyaotriton` now includes `$ORIGIN` in its `DT_RUNPATH` 2. `install` target now installs `pyaotriton` to `$CMAKE_INSTALL_PREFIX/lib` 3. `mptune` now stores testing results' batch size, making the timing results more informative 4. `performance_*.py` scripts now read `USE_TFLOPS`, `D_HEADS`, and `N_CTX` env vars, allowing changing the testing size without editing the code 5. `test/test_backward.py` now displays target fudge factors for fudge factor adjustment 6. `tune_flash.py` now shrinks batch size to 2 when both sequence lengths > 4096, to not exceed the VRAM limit. 7. Fix a problem of `sancheck_lut_tensor` in `class FlashKernel(KernelDescription)`, which did not handle single element LUT tensor correctly. 8. `v2python/table_tool.py` now ignores `inputs$BATCH` column Notes: 1. The fudge factors in use assume PyTorch <= 2.4. See pytorch/pytorch#135590 for detailed discussion why PyTorch 2.5 cannot be used for testing. PyTorch 2.6 will include a new interface to fix the problem.
xinyazhang
changed the title
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Outdated suggestions cannot be applied.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.