Add custom block size for neural network workflow #2891
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
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #2891 +/- ##
==========================================
- Coverage 55.82% 55.81% -0.02%
==========================================
Files 539 539
Lines 62582 62754 +172
Branches 8599 8607 +8
==========================================
+ Hits 34936 35025 +89
- Misses 25877 25951 +74
- Partials 1769 1778 +9 ☔ View full report in Codecov by Sentry. |
added 4 commits
July 26, 2024 21:16
There was the assumption that ModelSession should be able to handle multiple inputs and outputs, and each one of those with multiple shapes. This logic was left partially implemented, with a lot of assertions whenever ModelSession was used. At the same time, the interface of tiktorch server doesn't support multiple tensors and multiple shapes. The feature of supporting this should be revised, and not be left partially implemented. To be able to continue with other features, ModelSession has been refactored to just work with the simple case of single tensors and shapes.
ModelSession's input shape is more evidently classified either as explicit or parameterized.
thodkatz
force-pushed
the
add-nn-custom-block-size
branch
from
8717f5d to
19dc023
Compare
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Outdated suggestions cannot be applied.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.
Attempts to close #2701.
Before
After
We used to determine the block size by a fixed sensible value. But there were some use cases like #2071, that this value wasn't the best fit. So, I have attempted to expose the configuration of the block size in the neural network applet.
There are at least two important factors when determining the block size. Tiling artifacts caused by normalization issues from the model (relevant PR), and hardware constraints e.g. GPU memory usage.
Another example of tiling artifacts:
The current PR doesn't address the hardware constraints issue. It is possible to use a very large shape that can lead to an out of memory exception. Then the user should attempt to choose a smaller shape. But this process maybe shouldn't eventually be exposed to the user at all?