SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1712.05877 · 2017

Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference

Hartwig Adam, Bo Chen, Andrew Howard, Menglong Zhu, Dmitry Kalenichenko, Benoit Jacob, Skirmantas Kligys, Matthew Tang

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 7 functions out of this paper's own repositories and ran 7 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

FunctionStatusWhere it lives
BaseQuantizer Ran linyang-zhh/FQ-ViT/models/ptq/quantizer/uniform.py
code served (permissive licence) · get_code("6c8c5613a41846a2")
FullyConnected Ran jameszampa/ECE-570-Implementation/python/integer_inference.py
pointer only (licence: NONE) · get_code("40771743ce834305")
MultiplyByQuantizedMultiplierSmallerThanOne Ran jameszampa/ECE-570-Implementation/python/integer_inference.py
pointer only (licence: NONE) · get_code("9c24f357578474e5")
QuantizationFunction Ran MaximIntegratedAI/ai8x-training/ai8x.py
code served (permissive licence) · get_code("fbe27ee7dc10aa67")
RoundingDivideByPOT Ran jameszampa/ECE-570-Implementation/python/integer_inference.py
pointer only (licence: NONE) · get_code("f80d39ab0932d302")
SaturatingRoundingDoublingHighMul Ran jameszampa/ECE-570-Implementation/python/integer_inference.py
pointer only (licence: NONE) · get_code("98ec162d7259fc3b")
UniformQuantizer Ran linyang-zhh/FQ-ViT/models/ptq/quantizer/uniform.py
code served (permissive licence) · get_code("476098355ebe4558")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

The rising popularity of intelligent mobile devices and the daunting computational cost of deep learning-based models call for efficient and accurate on-device inference schemes. We propose a quantization scheme that allows inference to be carried out using integer-only arithmetic, which can be implemented more efficiently than floating point inference on commonly available integer-only hardware. We also co-design a training procedure to preserve end-to-end model accuracy post quantization. As a result, the proposed quantization scheme improves the tradeoff between accuracy and on-device latency. The improvements are significant even on MobileNets, a model family known for run-time efficiency, and are demonstrated in ImageNet classification and COCO detection on popular CPUs.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("1712.05877")
get_code_for_paper("1712.05877")
have("1712.05877")

Connect an agent — have() is free.