Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add cleanup transformation sorting inputs of commutative operations #85

Merged

Conversation

iksnagreb
Copy link
Contributor

@iksnagreb iksnagreb commented Nov 14, 2023

Adds the cleanup transformation requested in #84

Closes #84

iksnagreb and others added 20 commits November 20, 2023 16:10
This is required for supporting scaled dot-product attention, which has
3 dimensional tensors with batch/head, sequence and embedding dimension.
…ata_layouts

Introduce 3D layout annotations defaulting to NWC
…x/tf2onnx-1.16.1

Try newer tf2onnx version to get CI back in green
…ature/tensor_stats

Range analysis improvements and better input shape override
maltanar and others added 27 commits August 9, 2024 09:58
…eature/readme_dev_updates

Updates to dev docs: pip upgrade, linting
…p-pool-with-one-worker

Avoiding mp.Pool in case of using only 1 worker for easier pdb debugging
…t_conv_lowering_convmnist

Improve/test conv lowering convmnist
random input generated with seed=42 was causing a major difference
in Conv_13_out0 for no apparent reason (probably float / numerical
related)
…eature/convlower_qnt

Preserve weight quantizer while lowering convolutions
…eature/test_a2q_nets

Add easy fetch of accumulator-aware quantized (A2Q) CIFAR-10 models for testing
@maltanar
Copy link
Collaborator

LGTM, ready to merge. Thanks @iksnagreb !

@maltanar maltanar merged commit e02f701 into fastmachinelearning:main Sep 12, 2024
5 checks passed
@maltanar maltanar added this to the v0.4.0 milestone Dec 20, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Add cleanup transformation sorting inputs of commutative operations
9 participants