Skip to content
New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

#

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

Jump to bottom

Feature (quantizer): Adding accumulator-aware weight quantization #567

Merged
merged 31 commits into from
Apr 5, 2023
Merged

Feature (quantizer): Adding accumulator-aware weight quantization #567

merged 31 commits into from
Apr 5, 2023

Conversation

i-colbert
Copy link
Collaborator

Adding experimental accumulator-aware weight quantization based on Quantized Neural Networks for Low-Precision Accumulation with Guaranteed Overflow Avoidance.

@i-colbert i-colbert marked this pull request as ready for review March 30, 2023 03:37
@i-colbert i-colbert requested a review from volcacius March 30, 2023 03:37
@i-colbert i-colbert requested review from volcacius and removed request for volcacius March 30, 2023 23:38
i-colbert and others added 24 commits April 4, 2023 10:00
@i-colbert
Adding DecoupledRescalingIntQuantWithInput
b28f8e4
@i-colbert
Updating QuantWeightMixin to handle quant_input if required
9e1e2d9
@i-colbert
Updating base QuantLayer to pass quant_input to weight_quant
0fdeda9
@i-colbert
Adding new weight quant proxy to pass input bit-width and sign
d8a3fc7
@i-colbert
Adding input-aware decoupled weight quant to top-level import
97e2782
@i-colbert
Updating quant weight to use cached quant input
698a50a
@i-colbert
Adding accumulator-aware pre-clipping scaling factor
c599750
@i-colbert
Adding new weight quant
a5be8e6
@i-colbert
Adding pre-scaling to top-level import
7915b0b
@i-colbert
Updating quant injector
c3c9112
@i-colbert
Adding int8 weight quantizer to fixed_point.py
ed66c60
@i-colbert
Adding int8 a2q to quantizer testing
648de8f
@i-colbert
Adding pytest skip for case that A2Q does not support.
94b4211
@i-colbert
Minor cosmetic changes to pre-scaling impl
8e42223
@i-colbert
Adding accumulator bit-width bounds to nn.utils
a6986e1
@i-colbert
Updating function with notes and comments
66e7048
@i-colbert
Adding re-parameterization of default case_model fixture
2e7920c
@i-colbert
Adding tests for A2Q
029500d
@i-colbert
Pre-commit fixes (all but onnx export)
173ac27
@volcacius @i-colbert
Remove caching on the forward pass
64064c3
@volcacius @i-colbert
Add NotImplementedError for metadata retrieval over new decoupled proxy
f1d743d
@volcacius @i-colbert
Fix QuantScaleBias quant_weight
d6711ea
@i-colbert
Pre-commit fixes
f9967b6
@i-colbert
Fixing rebase merge
1731d49
i-colbert added 3 commits April 4, 2023 10:02
@i-colbert
Fixing first set of errors
0adeefd 
Can't target case_id deconstruction when a2q and base test cases are mismatched
@i-colbert
Skipping batch quant for a2q cases
d4e2e46
@i-colbert
Fixing a2q pytest cases
3898a65
@volcacius volcacius requested review from volcacius and removed request for volcacius April 4, 2023 18:06
@i-colbert i-colbert requested review from volcacius and removed request for volcacius April 4, 2023 18:21
@volcacius volcacius merged commit 55d4ee2 into Xilinx:dev Apr 5, 2023
# for free to join this conversation on GitHub. Already have an account? # to comment
Reviewers

@volcacius volcacius Awaiting requested review from volcacius

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@i-colbert @volcacius