Fix (llm): correct handling of attention mask shape #652
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
Giuseppe5
force-pushed
the
attetion_mask_llm
branch
from
8d48e24
to
d8a2bca
Compare
volcacius
pushed a commit
that referenced
this pull request
Jul 17, 2023
volcacius
added a commit
that referenced
this pull request
Jul 17, 2023
* Examples: WIP LLM block quantization * Add support for block zero-point * Add torch-mlir custom op support * Add test linear Signed-off-by: Alessandro Pappalardo <[email protected]> * Update to custom matmul export Signed-off-by: Alessandro Pappalardo <[email protected]> * Fix errors Signed-off-by: Alessandro Pappalardo <[email protected]> * Fix output shape of custom op Signed-off-by: Alessandro Pappalardo <[email protected]> * Add lowering to torch_mlir for single layer Signed-off-by: Alessandro Pappalardo <[email protected]> * Some cleanups * WIP llm flow Signed-off-by: Alessandro Pappalardo <[email protected]> * Fix (examples/llm): typo in custom quant matmul op (#607) * Test act equalization support * Initial end to end flow * Initial support for QuantMHA on OPT * Fix act equalization * Typos in prints * Reorganize validate * Add initial per row quantizers * Add per row input quantization support * Support group quant slicing * Adopt SliceTensor for block weight partial quant * Add float16 support * Fix scale type name * Add support for LN affine merging * WIP currently broken * Clean up weight eq support * Set weight narrow range always to False * Add fx act equalization, fixes for float16 support * Fix validate * Fix backport imports * Fix example export Signed-off-by: Alessandro Pappalardo <[email protected]> * Fix value_trace call in ln affine merging * Add per tensor/row/group dynamic scale support, some dtype improvements * Fix (llm): correct handling of attention mask shape (#652) * ALways export in fp32 base dtype on CPU * Export improvements * Fix errors after latest PR --------- Signed-off-by: Alessandro Pappalardo <[email protected]> Co-authored-by: jinchen62 <[email protected]> Co-authored-by: Giuseppe Franco <[email protected]>
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.