[mlir-gen] Add mlir builders for llama3.1 and tests #13
Conversation
This file contains hidden or 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
Adds a simple end-to-end example demonstrating programatic transform schedule creation, MLIR JIT compilation, execution, and numerical verification of the result. Additionally, 'utils' submodule is added with basic tools to simplify creation of ctype arguments in format accepted by jitted function.
Member
|
Should this be in |
Author
|
The e2e should be, yup, but this is mostly tests and getters. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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Outdated suggestions cannot be applied.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.
Putting up this dirty draft for early feedback/questions. I'm putting together some tests to run a e2e llama3.1 going through linalg on tensors. The goal is to generate some nice linalg that would be optimization friendly. At the moment, there are just functional blocks and pieces that are just smoke-tested. These include naive implementations for rotary embeddings, feed forward, rms, and a bunch of other small snippets that are useful to implement the model. These are already enough to put an attention block together. It'd be nice to test it against the original implementation, but that'd require fairscale as a dependency. For now I only added pytest and kept the pipeline as simple as possible. I also reused the example with the schedule, so now it is a part of every test.