[CUDA] Various int8 fix (cublas, cutlass, etc) #10596
Merged
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
masahi
force-pushed
the
cutlass-int8-fix
branch
from
4879857 to
12989d6
Compare
jwfromm
approved these changes
Mar 14, 2022
Contributor
jwfromm
left a comment
jwfromm
left a comment
There was a problem hiding this comment.
Those tabs can be dangerous. Thanks for the fix Masa, LGTM!
shingjan
pushed a commit
to shingjan/tvm
that referenced
this pull request
Mar 16, 2022
* [CUTLASS] avoid tile size 256 for int8 + align1 case * allow selecting int8 dense strategy for vulkan * fixed cublas batch matmul for int8 * fixed int8 dense tensorcore strategy * add cutlass conv align1 + int8 case * support int8 mixed precision cublas bmm * black
pfk-beta
pushed a commit
to pfk-beta/tvm
that referenced
this pull request
Apr 11, 2022
* [CUTLASS] avoid tile size 256 for int8 + align1 case * allow selecting int8 dense strategy for vulkan * fixed cublas batch matmul for int8 * fixed int8 dense tensorcore strategy * add cutlass conv align1 + int8 case * support int8 mixed precision cublas bmm * black
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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Outdated suggestions cannot be applied.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.
denseon int8 is never selected by the op strategy. This is because it sits in the else branch oftvm/python/tvm/relay/op/strategy/cuda.py
Lines 838 to 843 in 5c0ea30
batch_matmulThe first two fixes above make it possible to run int8
bert-baseend to end via autotvm or cublas on tensorcore (cutlass already works without error).@Laurawly @vinx13 @junrushao1994 @comaniac