Optimized compression for FP8 modes #3748
Draft
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
github-actions
bot
added
the
NNCF OpenVINO
| _f16_to_f8e4m3_bits_vec = np.vectorize(_f16_to_f8e4m3_bits_scalar, otypes=[np.uint8]) | ||
|
|
||
|
|
||
| def fp32_to_fp8e4m3_values(x: np.ndarray) -> np.ndarray: |
Collaborator
Author
There was a problem hiding this comment.
The reference conversion implementation looks a bit ugly unfortunately. It also can't be done within our nncf.Tensor framework yet, because _f16_to_f8e4m3_bits_scalar has to be vectorized with NumPy.
The reason behind all this, is that fp32 ->f8e4m3 conversion is done through fp16, i.e. fp32 -> fp16 -> f8e4m3. It is implemented this way on OpenVINO (and OneDNN) side because it is a more hardware friendly way. Please see the following links:
- OpenVINO implementation: https://github.com/openvinotoolkit/openvino/blame/master/src/core/src/type/float8_e4m3.cpp
- PR with addition of 2-step conversion: Implement 2-step conversion from fp32 to fp8 openvino#28501
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.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.
Changes
Reason for changes
UX improvement. On phi3-mini-4k-instruct for FP8_E4M3 mode, I get about 17x time reduction: 191 sec. -> 11 sec.
Related tickets
Tests