-
Notifications
You must be signed in to change notification settings - Fork 91
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
#8563: sweep split_query_key_value_and_split_heads, split and concat #8610
Merged
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
sjameelTT
requested review from
eyonland,
arakhmati,
cfjchu and
xanderchin
as code owners
May 17, 2024 16:04
sjameelTT
force-pushed
the
sjameel/sweeps
branch
7 times, most recently
from
May 24, 2024 18:45
ef8ed30
to
11570d3
Compare
sjameelTT
changed the title
#8563: sweep split_query_key_value_and_split_heads
#8563: sweep split_query_key_value_and_split_heads, split and concat
May 27, 2024
sjameelTT
force-pushed
the
sjameel/sweeps
branch
2 times, most recently
from
May 30, 2024 17:21
bfe9a36
to
0a1cdac
Compare
arakhmati
approved these changes
Jun 5, 2024
- Batch_size, and cores_h are together in a tuple to minimize permutations that are expected to fail - (Num_q_heads, num_kv_heads, cores_w) and (seq_len_q, seq_len_kv) are also in a tuple for the same reason - If all required combinations were tested with all data types and memory configurations then we would have 24192 as opposed to the current 2918 - PCC will be low for the interleaved version since the sharded and interleaved versions both expect the logical QKV tensor to be concatenated along different dimensions - TODO: add expected failure cases too for configurations that shouldn't be supported
- concat tests reworked to be able to track the dimensions for each test case (makes for easier debugging)
- use ttnn.experimental.tensor.split_dim_two_chunks_tiled instead of the ttnn split for now since split is not implemented and just a wrapper - add some known working configs since the current split is hardcoded to split in half - TODO: ttnn.split should be implemented, potentially call the ttnn.experimental versino - TODO: remove known working config cases when ttnn.split is implemented
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.
Add new sweep tests for ttnn.transformers.split_query_key_value_and_split_heads, split and concat
split_query_key_value_and_split_heads:
split:
concat: