---
title: '‘low-precision NN’ directory'
description: "Bibliography for directory ai/nn/sparsity/low-precision, most recent first: 1 related tag, 124 annotations, & 26 links (parent)."
thumbnail: /doc/ai/nn/fully-connected/2024-zhao-figure1-llmshavemuchrougherdecisionboundariesthanmlpsorsvmsordecisiontrees.png
thumbnail-text: 'Figure 1: Decision boundaries of LLMs and traditional machine learning models on a linearly separable binary classification task. The background colors represent the model’s predictions, while the points represent the in-context or training examples. LLMs exhibit non-smooth decision boundaries compared to the classical models. See Appendix E for model hyperparameters.'
thumbnail-css: "outline"
created: 2019-12-27
modified: 2026-09-20
status: in progress
previous: /doc/ai/nn/sparsity/knowledge-distillation/index
next: /doc/ai/nn/sparsity/pruning/index
confidence: log
importance: 0
css-extension: dropcaps-not
index: True
backlink: False
...
# See Also
::: {#see-alsos .directory-indexes .columns}
- [Parent ('*`NN sparsity`{=html}*' tag)](/doc/ai/nn/sparsity/index "Link to parent directory 'doc/ai/nn/sparsity/' (ascending)"){#_FBGJ03Sa
.link-annotated .link-tag .directory-indexes-upwards rel="tag"}
- [*`AI hardware`{=html}*](/doc/ai/scaling/hardware/index "‘AI hardware’ directory"){#_SnXXGYvz
.link-annotated .link-tag .directory-indexes-sideways rel="tag"}
:::
# Links {.display-pop-not}
## `“Score Centering Stabilizes Off-Policy Reinforcement Learning”, Marek & Ryabinin 2026`{=html} {#marek-ryabinin-2026-section}
**[`Score Centering Stabilizes Off-policy Reinforcement Learning`{=html}](https://arxiv.org/abs/2609.20807 "'Score Centering Stabilizes Off-policy Reinforcement Learning', Marek & Ryabinin 2026"){.link-modified-recently
.link-annotated .id-not .include-annotation link-icon="𝛘"
link-icon-type="text" link-icon-color="#b31b1b"}**
## `“Soulplayer-C64: A Real 25k-Parameter Transformer Running on a Commodore 64!”, gizmo64k 2026`{=html} {#gizmo64k-2026-section}
**[`soulplayer-c64: A real 25k-parameter transformer running on a Commodore 64!`{=html}](https://github.com/gizmo64k/soulplayer-c64 "'soulplayer-c64: A real 25k-parameter transformer running on a Commodore 64!', gizmo64k 2026"){.link-annotated
.id-not .include-annotation link-icon="github" link-icon-type="svg"}**
## `“EGGROLL: Evolution Strategies at the Hyperscale”, Sarkar et al 2025`{=html} {#sarkar-et-al-2025-section}
**[`EGGROLL: Evolution Strategies at the Hyperscale`{=html}](https://arxiv.org/abs/2511.16652 "'EGGROLL: Evolution Strategies at the Hyperscale', Sarkar et al 2025"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“1 Bit Is All We Need: Binary Normalized Neural Networks”, Cabral et al 2025`{=html} {#cabral-et-al-2025-section}
**[`1 bit is all we need: binary normalized neural networks`{=html}](https://arxiv.org/abs/2509.07025 "'1 bit is all we need: binary normalized neural networks', Cabral et al 2025"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Compiling a Differentiable Logic Gate Neural Net to C for a 1,744× Speedup”, Clayton 2025`{=html} {#clayton-2025-section}
**[`Compiling a Differentiable Logic Gate Neural Net to C for a 1,744× speedup`{=html}](https://slightknack.dev/blog/difflogic/ "'Compiling a Differentiable Logic Gate Neural Net to C for a 1,744× speedup', Clayton 2025"){.link-annotated-partial
.id-not .include-annotation}**
## `“Insights into DeepSeek-V3: Scaling Challenges and Reflections on Hardware for AI Architectures”, Zhao et al 2025`{=html} {#zhao-et-al-2025-section}
**[`Insights into DeepSeek-V3: Scaling Challenges and Reflections on Hardware for AI Architectures`{=html}](https://arxiv.org/abs/2505.09343#deepseek "'Insights into DeepSeek-V3: Scaling Challenges and Reflections on Hardware for AI Architectures', Zhao et al 2025"){.link-annotated
.id-not .include-annotation link-icon="deepseek" link-icon-type="svg"
link-icon-color="#4d6bfe"}**
## `“BitNet B1.58 2B4T Technical Report”, Ma et al 2025`{=html} {#ma-et-al-2025-section}
**[`BitNet b1.58 2B4T Technical Report`{=html}](https://arxiv.org/abs/2504.12285#microsoft "'BitNet b1.58 2B4T Technical Report', Ma et al 2025"){.link-annotated
.id-not .include-annotation link-icon="MS"
link-icon-type="text,sans,italic"}**
## `“MVDRAM: Enabling GeMV Execution in Unmodified DRAM for Low-Bit LLM Acceleration”, Kubo et al 2025`{=html} {#kubo-et-al-2025-section}
**[`MVDRAM: Enabling GeMV Execution in Unmodified DRAM for Low-Bit LLM Acceleration`{=html}](https://arxiv.org/abs/2503.23817 "'MVDRAM: Enabling GeMV Execution in Unmodified DRAM for Low-Bit LLM Acceleration', Kubo et al 2025"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Compute-Optimal LLMs Provably Generalize Better With Scale”, Finzi et al 2025`{=html} {#finzi-et-al-2025-section}
**[`Compute-Optimal LLMs Provably Generalize Better with Scale`{=html}](https://openreview.net/forum?id=MF7ljU8xcf "'Compute-Optimal LLMs Provably Generalize Better with Scale', Finzi et al 2025"){.link-annotated
.id-not .include-annotation link-icon="OR" link-icon-type="text,sans"
link-icon-color="#8c1b13"}**
## `“μnit Scaling: Simple and Scalable FP8 LLM Training”, Narayan et al 2025`{=html} {#narayan-et-al-2025-section}
**[`μnit Scaling: Simple and Scalable FP8 LLM Training`{=html}](https://arxiv.org/abs/2502.05967 "'μnit Scaling: Simple and Scalable FP8 LLM Training', Narayan et al 2025"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“2:4 Sparse Llama: Smaller Models for Efficient GPU Inference”, Kurtić et al 2024`{=html} {#kurtić-et-al-2024-section}
**[`2:4 Sparse Llama: Smaller Models for Efficient GPU Inference`{=html}](https://neuralmagic.com/blog/24-sparse-llama-smaller-models-for-efficient-gpu-inference/ "'2:4 Sparse Llama: Smaller Models for Efficient GPU Inference', Kurtić et al 2024"){.link-annotated-partial
.id-not .include-annotation}**
## `“Model Equality Testing: Which Model Is This API Serving?”, Gao et al 2024`{=html} {#gao-et-al-2024-1-section}
**[`Model Equality Testing: Which Model Is This API Serving?`{=html}](https://arxiv.org/abs/2410.20247 "'Model Equality Testing: Which Model Is This API Serving?', Gao et al 2024"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Addition Is All You Need for Energy-Efficient Language Models”, Luo & Sun 2024`{=html} {#luo-sun-2024-section}
**[`Addition is All You Need for Energy-efficient Language Models`{=html}](https://arxiv.org/abs/2410.00907 "'Addition is All You Need for Energy-efficient Language Models', Luo & Sun 2024"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Robust Training of Neural Networks at Arbitrary Precision and Sparsity”, Ye et al 2024`{=html} {#ye-et-al-2024-section}
**[`Robust Training of Neural Networks at Arbitrary Precision and Sparsity`{=html}](https://arxiv.org/abs/2409.09245#deepmind "'Robust Training of Neural Networks at Arbitrary Precision and Sparsity', Ye et al 2024"){.link-annotated
.id-not .include-annotation link-icon="deepmind" link-icon-type="svg"
link-icon-color="#4185f4"}**
## `“A Visual Guide to Quantization”, Grootendorst 2024`{=html} {#grootendorst-2024-section}
**[`A Visual Guide to Quantization`{=html}](https://newsletter.maartengrootendorst.com/p/a-visual-guide-to-quantization "'A Visual Guide to Quantization', Grootendorst 2024"){.link-annotated-partial
.id-not .include-annotation}**
## `“OpenDiLoCo: An Open-Source Framework for Globally Distributed Low-Communication Training”, Jaghouar et al 2024`{=html} {#jaghouar-et-al-2024-section}
**[`OpenDiLoCo: An Open-Source Framework for Globally Distributed Low-Communication Training`{=html}](https://arxiv.org/abs/2407.07852 "'OpenDiLoCo: An Open-Source Framework for Globally Distributed Low-Communication Training', Jaghouar et al 2024"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Probing the Decision Boundaries of In-Context Learning in Large Language Models”, Zhao et al 2024`{=html} {#zhao-et-al-2024-2-section}
**[`Probing the Decision Boundaries of In-context Learning in Large Language Models`{=html}](https://arxiv.org/abs/2406.11233 "'Probing the Decision Boundaries of In-context Learning in Large Language Models', Zhao et al 2024"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Nemotron-4 340B Technical Report”, Adler et al 2024`{=html} {#adler-et-al-2024-section}
**[`Nemotron-4 340B Technical Report`{=html}](https://arxiv.org/abs/2406.11704#nvidia "'Nemotron-4 340B Technical Report', Adler et al 2024"){.link-annotated
.id-not .include-annotation link-icon="n"
link-icon-type="text,sans,italic" link-icon-color="#77ba00"}**
## `“Scalable Matmul-Free Language Modeling”, Zhu et al 2024`{=html} {#zhu-et-al-2024-2-section}
**[`Scalable Matmul-free Language Modeling`{=html}](https://arxiv.org/abs/2406.02528 "'Scalable Matmul-free Language Modeling', Zhu et al 2024"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“BOLD: Boolean Logic Deep Learning”, Nguyen et al 2024`{=html} {#nguyen-et-al-2024-section}
**[`BOLD: Boolean Logic Deep Learning`{=html}](https://arxiv.org/abs/2405.16339 "'BOLD: Boolean Logic Deep Learning', Nguyen et al 2024"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Neural Networks (MNIST Inference) on the ‘3¢’ Microcontroller”, cpldcpu 2024`{=html} {#cpldcpu-2024-section}
**[`Neural Networks (MNIST inference) on the ‘3¢’ Microcontroller`{=html}](https://cpldcpu.wordpress.com/2024/05/02/machine-learning-mnist-inference-on-the-3-cent-microcontroller/ "'Neural Networks (MNIST inference) on the ‘3¢’ Microcontroller', cpldcpu 2024"){.link-annotated-partial
.id-not .include-annotation}**
## `“How Good Are Low-Bit Quantized LLaMA-3 Models? An Empirical Study”, Huang et al 2024`{=html} {#huang-et-al-2024-4-section}
**[`How Good Are Low-bit Quantized LLaMA-3 Models? An Empirical Study`{=html}](https://arxiv.org/abs/2404.14047 "'How Good Are Low-bit Quantized LLaMA-3 Models? An Empirical Study', Huang et al 2024"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Physics of Language Models: Part 3.3, Knowledge Capacity Scaling Laws”, Allen-Zhu & Li 2024`{=html} {#allen-zhu-li-2024-section}
**[`Physics of Language Models: Part 3.3, Knowledge Capacity Scaling Laws`{=html}](https://arxiv.org/abs/2404.05405#facebook "'Physics of Language Models: Part 3.3, Knowledge Capacity Scaling Laws', Allen-Zhu & Li 2024"){.link-annotated
.id-not .include-annotation link-icon="facebook" link-icon-type="svg"
link-icon-color="#1877f2"}**
## `“LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models”, Zheng et al 2024`{=html} {#zheng-et-al-2024-section}
**[`LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models`{=html}](https://arxiv.org/abs/2403.13372 "'LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models', Zheng et al 2024"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Decoding Compressed Trust: Scrutinizing the Trustworthiness of Efficient LLMs Under Compression”, Hong et al 2024`{=html} {#hong-et-al-2024-1-section}
**[`Decoding Compressed Trust: Scrutinizing the Trustworthiness of Efficient LLMs Under Compression`{=html}](https://arxiv.org/abs/2403.15447 "'Decoding Compressed Trust: Scrutinizing the Trustworthiness of Efficient LLMs Under Compression', Hong et al 2024"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“The Era of 1-Bit LLMs: All Large Language Models Are in 1.58 Bits”, Ma et al 2024`{=html} {#ma-et-al-2024-3-section}
**[`The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits`{=html}](https://arxiv.org/abs/2402.17764#microsoft "'The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits', Ma et al 2024"){.link-annotated
.id-not .include-annotation link-icon="MS"
link-icon-type="text,sans,italic"}**
## `“FP6-LLM: Efficiently Serving Large Language Models Through FP6-Centric Algorithm-System Co-Design”, Xia et al 2024`{=html} {#xia-et-al-2024-1-section}
**[`FP6-LLM: Efficiently Serving Large Language Models Through FP6-Centric Algorithm-System Co-Design`{=html}](https://arxiv.org/abs/2401.14112#microsoft "'FP6-LLM: Efficiently Serving Large Language Models Through FP6-Centric Algorithm-System Co-Design', Xia et al 2024"){.link-annotated
.id-not .include-annotation link-icon="MS"
link-icon-type="text,sans,italic"}**
## `“Beyond Chinchilla-Optimal: Accounting for Inference in Language Model Scaling Laws”, Sardana & Frankle 2023`{=html} {#sardana-frankle-2023-section}
**[`Beyond Chinchilla-Optimal: Accounting for Inference in Language Model Scaling Laws`{=html}](https://arxiv.org/abs/2401.00448 "'Beyond Chinchilla-Optimal: Accounting for Inference in Language Model Scaling Laws', Sardana & Frankle 2023"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“TinyGPT-V: Efficient Multimodal Large Language Model via Small Backbones”, Yuan et al 2023`{=html} {#yuan-et-al-2023-1-section}
**[`TinyGPT-V: Efficient Multimodal Large Language Model via Small Backbones`{=html}](https://arxiv.org/abs/2312.16862 "'TinyGPT-V: Efficient Multimodal Large Language Model via Small Backbones', Yuan et al 2023"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“LLM-FP4: 4-Bit Floating-Point Quantized Transformers”, Liu et al 2023`{=html} {#liu-et-al-2023-06-section}
**[`LLM-FP4: 4-Bit Floating-Point Quantized Transformers`{=html}](https://arxiv.org/abs/2310.16836 "'LLM-FP4: 4-Bit Floating-Point Quantized Transformers', Liu et al 2023"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Training Transformers With 4-Bit Integers”, Xi et al 2023`{=html} {#xi-et-al-2023-section}
**[`Training Transformers with 4-bit Integers`{=html}](https://arxiv.org/abs/2306.11987 "'Training Transformers with 4-bit Integers', Xi et al 2023"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“SpQR: A Sparse-Quantized Representation for Near-Lossless LLM Weight Compression”, Dettmers et al 2023`{=html} {#dettmers-et-al-2023-section}
**[`SpQR: A Sparse-Quantized Representation for Near-Lossless LLM Weight Compression`{=html}](https://arxiv.org/abs/2306.03078 "'SpQR: A Sparse-Quantized Representation for Near-Lossless LLM Weight Compression', Dettmers et al 2023"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Binary and Ternary Natural Language Generation”, Liu et al 2023`{=html} {#liu-et-al-2023-13-section}
**[`Binary and Ternary Natural Language Generation`{=html}](https://arxiv.org/abs/2306.01841#facebook "'Binary and Ternary Natural Language Generation', Liu et al 2023"){.link-annotated
.id-not .include-annotation link-icon="facebook" link-icon-type="svg"
link-icon-color="#1877f2"}**
## `“AWQ: Activation-Aware Weight Quantization for LLM Compression and Acceleration”, Lin et al 2023`{=html} {#lin-et-al-2023-7-section}
**[`AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration`{=html}](https://arxiv.org/abs/2306.00978 "'AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration', Lin et al 2023"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Big-PERCIVAL: Exploring the Native Use of 64-Bit Posit Arithmetic in Scientific Computing”, Mallasén et al 2023`{=html} {#mallasén-et-al-2023-section}
**[`Big-PERCIVAL: Exploring the Native Use of 64-Bit Posit Arithmetic in Scientific Computing`{=html}](https://arxiv.org/abs/2305.06946 "'Big-PERCIVAL: Exploring the Native Use of 64-Bit Posit Arithmetic in Scientific Computing', Mallasén et al 2023"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Int-4 LLaMa Is Not Enough—Int-3 and Beyond: More Compression, Easier to Build Apps on LLMs That Run Locally”, nolano.org 2023`{=html} {#nolanoorg-2023-section}
**[`Int-4 LLaMa is not enough—Int-3 and beyond: More compression, easier to build apps on LLMs that run locally`{=html}](https://nolanoorg.substack.com/p/int-4-llama-is-not-enough-int-3-and "'Int-4 LLaMa is not enough—Int-3 and beyond: More compression, easier to build apps on LLMs that run locally', nolano.org 2023"){.link-annotated
.id-not .include-annotation link-icon="substack" link-icon-type="svg"
link-icon-color="#ff6719"}**
## `“SpikeGPT: Generative Pre-Trained Language Model With Spiking Neural Networks”, Zhu et al 2023`{=html} {#zhu-et-al-2023-3-section}
**[`SpikeGPT: Generative Pre-trained Language Model with Spiking Neural Networks`{=html}](https://arxiv.org/abs/2302.13939 "'SpikeGPT: Generative Pre-trained Language Model with Spiking Neural Networks', Zhu et al 2023"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“BMT: Binarized Neural Machine Translation”, Zhang et al 2023`{=html} {#zhang-et-al-2023-18-section}
**[`BMT: Binarized Neural Machine Translation`{=html}](https://arxiv.org/abs/2302.04907#google "'BMT: Binarized Neural Machine Translation', Zhang et al 2023"){.link-annotated
.id-not .include-annotation link-icon="alphabet" link-icon-type="svg"
link-icon-color="#4285f4"}**
## `“Self-Compressing Neural Networks”, Cséfalvay & Imber 2023`{=html} {#cséfalvay-imber-2023-section}
**[`Self-Compressing Neural Networks`{=html}](https://arxiv.org/abs/2301.13142 "'Self-Compressing Neural Networks', Cséfalvay & Imber 2023"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Who Says Elephants Can’t Run: Bringing Large Scale MoE Models into Cloud Scale Production”, Kim et al 2022`{=html} {#kim-et-al-2022-3-section}
**[`Who Says Elephants Can’t Run: Bringing Large Scale MoE Models into Cloud Scale Production`{=html}](https://arxiv.org/abs/2211.10017#microsoft "'Who Says Elephants Can’t Run: Bringing Large Scale MoE Models into Cloud Scale Production', Kim et al 2022"){.link-annotated
.id-not .include-annotation link-icon="MS"
link-icon-type="text,sans,italic"}**
## `“SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models”, Xiao et al 2022`{=html} {#xiao-et-al-2022-1-section}
**[`SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models`{=html}](https://arxiv.org/abs/2211.10438 "'SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models', Xiao et al 2022"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Efficiently Scaling Transformer Inference”, Pope et al 2022`{=html} {#pope-et-al-2022-section}
**[`Efficiently Scaling Transformer Inference`{=html}](https://arxiv.org/abs/2211.05102#google "'Efficiently Scaling Transformer Inference', Pope et al 2022"){.link-annotated
.id-not .include-annotation link-icon="alphabet" link-icon-type="svg"
link-icon-color="#4285f4"}**
## `“GPTQ: Accurate Post-Training Quantization for Generative Pre-Trained Transformers”, Frantar et al 2022`{=html} {#frantar-et-al-2022-section}
**[`GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers`{=html}](https://arxiv.org/abs/2210.17323 "'GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers', Frantar et al 2022"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Fast DistilBERT on CPUs”, Shen et al 2022`{=html} {#shen-et-al-2022-section}
**[`Fast DistilBERT on CPUs`{=html}](https://arxiv.org/abs/2211.07715#intel "'Fast DistilBERT on CPUs', Shen et al 2022"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Broken Neural Scaling Laws”, Caballero et al 2022`{=html} {#caballero-et-al-2022-section}
**[`Broken Neural Scaling Laws`{=html}](https://arxiv.org/abs/2210.14891 "'Broken Neural Scaling Laws', Caballero et al 2022"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“GLM-130B: An Open Bilingual Pre-Trained Model”, Zeng et al 2022`{=html} {#zeng-et-al-2022-1-section}
**[`GLM-130B: An Open Bilingual Pre-trained Model`{=html}](https://arxiv.org/abs/2210.02414#zhipu "'GLM-130B: An Open Bilingual Pre-trained Model', Zeng et al 2022"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“FP8 Formats for Deep Learning”, Micikevicius et al 2022`{=html} {#micikevicius-et-al-2022-section}
**[`FP8 Formats for Deep Learning`{=html}](https://arxiv.org/abs/2209.05433#nvidia "'FP8 Formats for Deep Learning', Micikevicius et al 2022"){.link-annotated
.id-not .include-annotation link-icon="n"
link-icon-type="text,sans,italic" link-icon-color="#77ba00"}**
## `“LLM.int8(): 8-Bit Matrix Multiplication for Transformers at Scale”, Dettmers et al 2022`{=html} {#dettmers-et-al-2022-section}
**[`LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale`{=html}](https://arxiv.org/abs/2208.07339#facebook "'LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale', Dettmers et al 2022"){.link-annotated
.id-not .include-annotation link-icon="facebook" link-icon-type="svg"
link-icon-color="#1877f2"}**
## `“Is Integer Arithmetic Enough for Deep Learning Training?”, Ghaffari et al 2022`{=html} {#ghaffari-et-al-2022-section}
**[`Is Integer Arithmetic Enough for Deep Learning Training?`{=html}](https://arxiv.org/abs/2207.08822 "'Is Integer Arithmetic Enough for Deep Learning Training?', Ghaffari et al 2022"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“On-Device Training Under 256KB Memory”, Lin et al 2022`{=html} {#lin-et-al-2022-smolml-section}
**[`On-Device Training Under 256KB Memory`{=html}](https://arxiv.org/abs/2206.15472 "'On-Device Training Under 256KB Memory', Lin et al 2022"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“How to Train Accurate BNNs for Embedded Systems?”, Putter & Corporaal 2022`{=html} {#putter-corporaal-2022-section}
**[`How to train accurate BNNs for embedded systems?`{=html}](https://arxiv.org/abs/2206.12322 "'How to train accurate BNNs for embedded systems?', Putter & Corporaal 2022"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Director: Deep Hierarchical Planning from Pixels”, Hafner et al 2022`{=html} {#hafner-et-al-2022-section}
**[`Director: Deep Hierarchical Planning from Pixels`{=html}](https://arxiv.org/abs/2206.04114#google "'Director: Deep Hierarchical Planning from Pixels', Hafner et al 2022"){.link-annotated
.id-not .include-annotation link-icon="alphabet" link-icon-type="svg"
link-icon-color="#4285f4"}**
## `“8-Bit Numerical Formats for Deep Neural Networks”, Noune et al 2022`{=html} {#noune-et-al-2022-section}
**[`8-bit Numerical Formats for Deep Neural Networks`{=html}](https://arxiv.org/abs/2206.02915#graphcore "'8-bit Numerical Formats for Deep Neural Networks', Noune et al 2022"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“XTC: Extreme Compression for Pre-Trained Transformers Made Simple and Efficient”, Wu et al 2022`{=html} {#wu-et-al-2022-08-section}
**[`XTC: Extreme Compression for Pre-trained Transformers Made Simple and Efficient`{=html}](https://arxiv.org/abs/2206.01859#microsoft "'XTC: Extreme Compression for Pre-trained Transformers Made Simple and Efficient', Wu et al 2022"){.link-annotated
.id-not .include-annotation link-icon="MS"
link-icon-type="text,sans,italic"}**
## `“ZeroQuant: Efficient and Affordable Post-Training Quantization for Large-Scale Transformers”, Yao et al 2022`{=html} {#yao-et-al-2022-2-section}
**[`ZeroQuant: Efficient and Affordable Post-Training Quantization for Large-Scale Transformers`{=html}](https://arxiv.org/abs/2206.01861#microsoft "'ZeroQuant: Efficient and Affordable Post-Training Quantization for Large-Scale Transformers', Yao et al 2022"){.link-annotated
.id-not .include-annotation link-icon="MS"
link-icon-type="text,sans,italic"}**
## `“Matryoshka Representations for Adaptive Deployment”, Kusupati et al 2022`{=html} {#kusupati-et-al-2022-section}
**[`Matryoshka Representations for Adaptive Deployment`{=html}](https://arxiv.org/abs/2205.13147 "'Matryoshka Representations for Adaptive Deployment', Kusupati et al 2022"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“PLAID: An Efficient Engine for Late Interaction Retrieval”, Santhanam et al 2022`{=html} {#santhanam-et-al-2022-2-section}
**[`PLAID: An Efficient Engine for Late Interaction Retrieval`{=html}](https://arxiv.org/abs/2205.09707 "'PLAID: An Efficient Engine for Late Interaction Retrieval', Santhanam et al 2022"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Maximizing Communication Efficiency for Large-Scale Training via 0/1 Adam”, Lu et al 2022`{=html} {#lu-et-al-2022-7-section}
**[`Maximizing Communication Efficiency for Large-scale Training via 0/1 Adam`{=html}](https://arxiv.org/abs/2202.06009#microsoft "'Maximizing Communication Efficiency for Large-scale Training via 0/1 Adam', Lu et al 2022"){.link-annotated
.id-not .include-annotation link-icon="MS"
link-icon-type="text,sans,italic"}**
## `“Is Programmable Overhead Worth The Cost? How Much Do We Pay for a System to Be Programmable? It Depends upon Who You Ask”, Bailey 2022`{=html} {#bailey-2022-section}
**[`Is Programmable Overhead Worth The Cost? How much do we pay for a system to be programmable? It depends upon who you ask`{=html}](https://semiengineering.com/is-programmable-overhead-worth-the-cost/ "'Is Programmable Overhead Worth The Cost? How much do we pay for a system to be programmable? It depends upon who you ask', Bailey 2022"){.link-annotated
.id-not .include-annotation}**
## `“Boosted Dense Retriever”, Lewis et al 2021`{=html} {#lewis-et-al-2021-section}
**[`Boosted Dense Retriever`{=html}](https://arxiv.org/abs/2112.07771#facebook "'Boosted Dense Retriever', Lewis et al 2021"){.link-annotated
.id-not .include-annotation link-icon="facebook" link-icon-type="svg"
link-icon-color="#1877f2"}**
## `“FQ-ViT: Fully Quantized Vision Transformer without Retraining”, Lin et al 2021`{=html} {#lin-et-al-2021-4-section}
**[`FQ-ViT: Fully Quantized Vision Transformer without Retraining`{=html}](https://arxiv.org/abs/2111.13824 "'FQ-ViT: Fully Quantized Vision Transformer without Retraining', Lin et al 2021"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“𝜇NCA: Texture Generation With Ultra-Compact Neural Cellular Automata”, Mordvintsev & Niklasson 2021`{=html} {#mordvintsev-niklasson-2021-section}
**[`𝜇NCA: Texture Generation with Ultra-Compact Neural Cellular Automata`{=html}](https://arxiv.org/abs/2111.13545#google "'𝜇NCA: Texture Generation with Ultra-Compact Neural Cellular Automata', Mordvintsev & Niklasson 2021"){.link-annotated
.id-not .include-annotation link-icon="alphabet" link-icon-type="svg"
link-icon-color="#4285f4"}**
## `“Prune Once for All: Sparse Pre-Trained Language Models”, Zafrir et al 2021`{=html} {#zafrir-et-al-2021-section}
**[`Prune Once for All: Sparse Pre-Trained Language Models`{=html}](https://arxiv.org/abs/2111.05754 "'Prune Once for All: Sparse Pre-Trained Language Models', Zafrir et al 2021"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“8-Bit Optimizers via Block-Wise Quantization”, Dettmers et al 2021`{=html} {#dettmers-et-al-2021-section}
**[`8-bit Optimizers via Block-wise Quantization`{=html}](https://arxiv.org/abs/2110.02861 "'8-bit Optimizers via Block-wise Quantization', Dettmers et al 2021"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Understanding and Overcoming the Challenges of Efficient Transformer Quantization”, Bondarenko et al 2021`{=html} {#bondarenko-et-al-2021-section}
**[`Understanding and Overcoming the Challenges of Efficient Transformer Quantization`{=html}](https://arxiv.org/abs/2109.12948 "'Understanding and Overcoming the Challenges of Efficient Transformer Quantization', Bondarenko et al 2021"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Quantization Backdoors to Deep Learning Commercial Frameworks”, Ma et al 2021`{=html} {#ma-et-al-2021-section}
**[`Quantization Backdoors to Deep Learning Commercial Frameworks`{=html}](https://arxiv.org/abs/2108.09187 "'Quantization Backdoors to Deep Learning Commercial Frameworks', Ma et al 2021"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Efficient Deep Learning: A Survey on Making Deep Learning Models Smaller, Faster, and Better”, Menghani 2021`{=html} {#menghani-2021-section}
**[`Efficient Deep Learning: A Survey on Making Deep Learning Models Smaller, Faster, and Better`{=html}](https://arxiv.org/abs/2106.08962 "'Efficient Deep Learning: A Survey on Making Deep Learning Models Smaller, Faster, and Better', Menghani 2021"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“A Winning Hand: Compressing Deep Networks Can Improve Out-Of-Distribution Robustness”, Diffenderfer et al 2021`{=html} {#diffenderfer-et-al-2021-section}
**[`A Winning Hand: Compressing Deep Networks Can Improve Out-Of-Distribution Robustness`{=html}](https://arxiv.org/abs/2106.09129 "'A Winning Hand: Compressing Deep Networks Can Improve Out-Of-Distribution Robustness', Diffenderfer et al 2021"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Ten Lessons From Three Generations Shaped Google’s TPUv4i”, Jouppi et al 2021`{=html} {#jouppi-et-al-2021-section}
**[`Ten Lessons From Three Generations Shaped Google’s TPUv4i`{=html}](/doc/ai/scaling/hardware/2021-jouppi.pdf "'Ten Lessons From Three Generations Shaped Google’s TPUv4i', Jouppi et al 2021"){.link-annotated
.id-not .include-annotation link-icon="pdf" link-icon-type="svg"
link-icon-color="#f40f02"}**
## `“High-Performance, Distributed Training of Large-Scale Deep Learning Recommendation Models (DLRMs)”, Mudigere et al 2021`{=html} {#mudigere-et-al-2021-section}
**[`High-performance, Distributed Training of Large-scale Deep Learning Recommendation Models (DLRMs)`{=html}](https://arxiv.org/abs/2104.05158#facebook "'High-performance, Distributed Training of Large-scale Deep Learning Recommendation Models (DLRMs)', Mudigere et al 2021"){.link-annotated
.id-not .include-annotation link-icon="facebook" link-icon-type="svg"
link-icon-color="#1877f2"}**
## `“Deep Residual Learning in Spiking Neural Networks”, Fang et al 2021`{=html} {#fang-et-al-2021-5-section}
**[`Deep Residual Learning in Spiking Neural Networks`{=html}](https://arxiv.org/abs/2102.04159 "'Deep Residual Learning in Spiking Neural Networks', Fang et al 2021"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“1-Bit Adam: Communication Efficient Large-Scale Training With Adam’s Convergence Speed”, Tang et al 2021`{=html} {#tang-et-al-2021-1bitadam-section}
**[`1-bit Adam: Communication Efficient Large-Scale Training with Adam’s Convergence Speed`{=html}](https://arxiv.org/abs/2102.02888#microsoft "'1-bit Adam: Communication Efficient Large-Scale Training with Adam’s Convergence Speed', Tang et al 2021"){.link-annotated
.id-not .include-annotation link-icon="MS"
link-icon-type="text,sans,italic"}**
## `“ES-ENAS: Blackbox Optimization over Hybrid Spaces via Combinatorial and Continuous Evolution”, Song et al 2021`{=html} {#song-et-al-2021-2-section}
**[`ES-ENAS: Blackbox Optimization over Hybrid Spaces via Combinatorial and Continuous Evolution`{=html}](https://arxiv.org/abs/2101.07415#google "'ES-ENAS: Blackbox Optimization over Hybrid Spaces via Combinatorial and Continuous Evolution', Song et al 2021"){.link-annotated
.id-not .include-annotation link-icon="alphabet" link-icon-type="svg"
link-icon-color="#4285f4"}**
## `“Switch Transformers: Scaling to Trillion Parameter Models With Simple and Efficient Sparsity”, Fedus et al 2021`{=html} {#fedus-et-al-2021-section}
**[`Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity`{=html}](https://arxiv.org/abs/2101.03961#google "'Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity', Fedus et al 2021"){.link-annotated
.id-not .include-annotation link-icon="alphabet" link-icon-type="svg"
link-icon-color="#4285f4"}**
## `“A Primer in BERTology: What We Know about How BERT Works”, Rogers et al 2020`{=html} {#rogers-et-al-2020-section}
**[`A Primer in BERTology: What we know about how BERT works`{=html}](https://arxiv.org/abs/2002.12327 "'A Primer in BERTology: What we know about how BERT works', Rogers et al 2020"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“RegDeepDanbooru: Yet Another Deep Danbooru Project”, zyddnys 2020`{=html} {#zyddnys-2020-section}
**[`RegDeepDanbooru: Yet another Deep Danbooru project`{=html}](https://github.com/zyddnys/RegDeepDanbooru "'RegDeepDanbooru: Yet another Deep Danbooru project', zyddnys 2020"){.link-annotated
.id-not .include-annotation link-icon="github" link-icon-type="svg"}**
## `“TernaryBERT: Distillation-Aware Ultra-Low Bit BERT”, Zhang et al 2020`{=html} {#zhang-et-al-2020-07-section}
**[`TernaryBERT: Distillation-aware Ultra-low Bit BERT`{=html}](https://arxiv.org/abs/2009.12812 "'TernaryBERT: Distillation-aware Ultra-low Bit BERT', Zhang et al 2020"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“HOBFLOPS CNNs: Hardware Optimized Bitslice-Parallel Floating-Point Operations for Convolutional Neural Networks”, Garland & Gregg 2020`{=html} {#garland-gregg-2020-section}
**[`HOBFLOPS CNNs: Hardware Optimized Bitslice-Parallel Floating-Point Operations for Convolutional Neural Networks`{=html}](https://arxiv.org/abs/2007.06563#facebook "'HOBFLOPS CNNs: Hardware Optimized Bitslice-Parallel Floating-Point Operations for Convolutional Neural Networks', Garland & Gregg 2020"){.link-annotated
.id-not .include-annotation link-icon="facebook" link-icon-type="svg"
link-icon-color="#1877f2"}**
## `“Bayesian Bits: Unifying Quantization and Pruning”, Baalen et al 2020`{=html} {#baalen-et-al-2020-section}
**[`Bayesian Bits: Unifying Quantization and Pruning`{=html}](https://arxiv.org/abs/2005.07093 "'Bayesian Bits: Unifying Quantization and Pruning', Baalen et al 2020"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“General Purpose Text Embeddings from Pre-Trained Language Models for Scalable Inference”, Du et al 2020`{=html} {#du-et-al-2020-section}
**[`General Purpose Text Embeddings from Pre-trained Language Models for Scalable Inference`{=html}](https://arxiv.org/abs/2004.14287#facebook "'General Purpose Text Embeddings from Pre-trained Language Models for Scalable Inference', Du et al 2020"){.link-annotated
.id-not .include-annotation link-icon="facebook" link-icon-type="svg"
link-icon-color="#1877f2"}**
## `“Lite Transformer With Long-Short Range Attention”, Wu et al 2020`{=html} {#wu-et-al-2020-3-section}
**[`Lite Transformer with Long-Short Range Attention`{=html}](https://arxiv.org/abs/2004.11886 "'Lite Transformer with Long-Short Range Attention', Wu et al 2020"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Training With Quantization Noise for Extreme Model Compression”, Fan et al 2020`{=html} {#fan-et-al-2020-2-section}
**[`Training with Quantization Noise for Extreme Model Compression`{=html}](https://arxiv.org/abs/2004.07320#facebook "'Training with Quantization Noise for Extreme Model Compression', Fan et al 2020"){.link-annotated
.id-not .include-annotation link-icon="facebook" link-icon-type="svg"
link-icon-color="#1877f2"}**
## `“Moniqua: Modulo Quantized Communication in Decentralized SGD”, Lu & Sa 2020`{=html} {#lu-sa-2020-section}
**[`Moniqua: Modulo Quantized Communication in Decentralized SGD`{=html}](https://arxiv.org/abs/2002.11787 "'Moniqua: Modulo Quantized Communication in Decentralized SGD', Lu & Sa 2020"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Train Large, Then Compress: Rethinking Model Size for Efficient Training and Inference of Transformers”, Li et al 2020`{=html} {#li-et-al-2020-5-section}
**[`Train Large, Then Compress: Rethinking Model Size for Efficient Training and Inference of Transformers`{=html}](https://arxiv.org/abs/2002.11794 "'Train Large, Then Compress: Rethinking Model Size for Efficient Training and Inference of Transformers', Li et al 2020"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“L2L: Training Large Neural Networks With Constant Memory Using a New Execution Algorithm”, Pudipeddi et al 2020`{=html} {#pudipeddi-et-al-2020-section}
**[`L2L: Training Large Neural Networks with Constant Memory using a New Execution Algorithm`{=html}](https://arxiv.org/abs/2002.05645#microsoft "'L2L: Training Large Neural Networks with Constant Memory using a New Execution Algorithm', Pudipeddi et al 2020"){.link-annotated
.id-not .include-annotation link-icon="MS"
link-icon-type="text,sans,italic"}**
## `“SWAT: Sparse Weight Activation Training”, Raihan & Aamodt 2020`{=html} {#raihan-aamodt-2020-section}
**[`SWAT: Sparse Weight Activation Training`{=html}](https://arxiv.org/abs/2001.01969 "'SWAT: Sparse Weight Activation Training', Raihan & Aamodt 2020"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“QUARL: Quantized Reinforcement Learning (ActorQ)”, Lam et al 2019`{=html} {#lam-et-al-2019-section}
**[`QUARL: Quantized Reinforcement Learning (ActorQ)`{=html}](https://arxiv.org/abs/1910.01055#google "'QUARL: Quantized Reinforcement Learning (ActorQ)', Lam et al 2019"){.link-annotated
.id-not .include-annotation link-icon="alphabet" link-icon-type="svg"
link-icon-color="#4285f4"}**
## `“SCaNN: Accelerating Large-Scale Inference With Anisotropic Vector Quantization”, Guo et al 2019`{=html} {#guo-et-al-2019-2-section}
**[`SCaNN: Accelerating Large-Scale Inference with Anisotropic Vector Quantization`{=html}](https://arxiv.org/abs/1908.10396#google "'SCaNN: Accelerating Large-Scale Inference with Anisotropic Vector Quantization', Guo et al 2019"){.link-annotated
.id-not .include-annotation link-icon="alphabet" link-icon-type="svg"
link-icon-color="#4285f4"}**
## `“And the Bit Goes Down: Revisiting the Quantization of Neural Networks”, Stock et al 2019`{=html} {#stock-et-al-2019-section}
**[`And the Bit Goes Down: Revisiting the Quantization of Neural Networks`{=html}](https://arxiv.org/abs/1907.05686 "'And the Bit Goes Down: Revisiting the Quantization of Neural Networks', Stock et al 2019"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Surrogate Gradient Learning in Spiking Neural Networks”, Neftci et al 2019`{=html} {#neftci-et-al-2019-section}
**[`Surrogate Gradient Learning in Spiking Neural Networks`{=html}](https://arxiv.org/abs/1901.09948 "'Surrogate Gradient Learning in Spiking Neural Networks', Neftci et al 2019"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Rethinking Floating Point for Deep Learning”, Johnson 2018`{=html} {#johnson-2018-section}
**[`Rethinking floating point for deep learning`{=html}](https://arxiv.org/abs/1811.01721 "'Rethinking floating point for deep learning', Johnson 2018"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Learning Recurrent Binary/Ternary Weights”, Ardakani et al 2018`{=html} {#ardakani-et-al-2018-section}
**[`Learning Recurrent Binary/Ternary Weights`{=html}](https://arxiv.org/abs/1809.11086 "'Learning Recurrent Binary/Ternary Weights', Ardakani et al 2018"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Rethinking Numerical Representations for Deep Neural Networks”, Hill et al 2018`{=html} {#hill-et-al-2018-2-section}
**[`Rethinking Numerical Representations for Deep Neural Networks`{=html}](https://arxiv.org/abs/1808.02513 "'Rethinking Numerical Representations for Deep Neural Networks', Hill et al 2018"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Highly Scalable Deep Learning Training System With Mixed-Precision: Training ImageNet in 4 Minutes”, Jia et al 2018`{=html} {#jia-et-al-2018-section}
**[`Highly Scalable Deep Learning Training System with Mixed-Precision: Training ImageNet in 4 Minutes`{=html}](https://arxiv.org/abs/1807.11205#tencent "'Highly Scalable Deep Learning Training System with Mixed-Precision: Training ImageNet in 4 Minutes', Jia et al 2018"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Quantization Mimic: Towards Very Tiny CNN for Object Detection”, Wei et al 2018`{=html} {#wei-et-al-2018-section}
**[`Quantization Mimic: Towards Very Tiny CNN for Object Detection`{=html}](https://arxiv.org/abs/1805.02152 "'Quantization Mimic: Towards Very Tiny CNN for Object Detection', Wei et al 2018"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Training ImageNet in 3 Hours for $25; and CIFAR-10 for $0.26”, Howard 2018`{=html} {#howard-2018-section}
**[`Training ImageNet in 3 hours for $25; and CIFAR-10 for $0.26`{=html}](https://www.fast.ai/2018/04/30/dawnbench-fastai/ "'Training ImageNet in 3 hours for $25; and CIFAR-10 for $0.26', Howard 2018"){.link-annotated
.id-not .include-annotation link-icon="Fai" link-icon-type="text,tri"
link-icon-color="#3399f3"}**
## `“High-Accuracy Low-Precision Training”, Sa et al 2018`{=html} {#sa-et-al-2018-section}
**[`High-Accuracy Low-Precision Training`{=html}](https://arxiv.org/abs/1803.03383 "'High-Accuracy Low-Precision Training', Sa et al 2018"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Training Wide Residual Networks for Deployment Using a Single Bit for Each Weight”, McDonnell 2018`{=html} {#mcdonnell-2018-section}
**[`Training wide residual networks for deployment using a single bit for each weight`{=html}](https://arxiv.org/abs/1802.08530 "'Training wide residual networks for deployment using a single bit for each weight', McDonnell 2018"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Universal Deep Neural Network Compression”, Choi et al 2018`{=html} {#choi-et-al-2018-section}
**[`Universal Deep Neural Network Compression`{=html}](https://arxiv.org/abs/1802.02271 "'Universal Deep Neural Network Compression', Choi et al 2018"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Deep Gradient Compression: Reducing the Communication Bandwidth for Distributed Training”, Lin et al 2017`{=html} {#lin-et-al-2017-1-section}
**[`Deep Gradient Compression: Reducing the Communication Bandwidth for Distributed Training`{=html}](https://arxiv.org/abs/1712.01887 "'Deep Gradient Compression: Reducing the Communication Bandwidth for Distributed Training', Lin et al 2017"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Shift: A Zero FLOP, Zero Parameter Alternative to Spatial Convolutions”, Wu et al 2017`{=html} {#wu-et-al-2017-1-section}
**[`Shift: A Zero FLOP, Zero Parameter Alternative to Spatial Convolutions`{=html}](https://arxiv.org/abs/1711.08141 "'Shift: A Zero FLOP, Zero Parameter Alternative to Spatial Convolutions', Wu et al 2017"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Training Simplification and Model Simplification for Deep Learning: A Minimal Effort Back Propagation Method”, Sun et al 2017`{=html} {#sun-et-al-2017-1-section}
**[`Training Simplification and Model Simplification for Deep Learning: A Minimal Effort Back Propagation Method`{=html}](https://arxiv.org/abs/1711.06528 "'Training Simplification and Model Simplification for Deep Learning: A Minimal Effort Back Propagation Method', Sun et al 2017"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Compressing Word Embeddings via Deep Compositional Code Learning”, Shu & Nakayama 2017`{=html} {#shu-nakayama-2017-section}
**[`Compressing Word Embeddings via Deep Compositional Code Learning`{=html}](https://arxiv.org/abs/1711.01068 "'Compressing Word Embeddings via Deep Compositional Code Learning', Shu & Nakayama 2017"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Learning Discrete Weights Using the Local Reparameterization Trick”, Shayer et al 2017`{=html} {#shayer-et-al-2017-section}
**[`Learning Discrete Weights Using the Local Reparameterization Trick`{=html}](https://arxiv.org/abs/1710.07739 "'Learning Discrete Weights Using the Local Reparameterization Trick', Shayer et al 2017"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“TensorQuant—A Simulation Toolbox for Deep Neural Network Quantization”, Loroch et al 2017`{=html} {#loroch-et-al-2017-section}
**[`TensorQuant—A Simulation Toolbox for Deep Neural Network Quantization`{=html}](https://arxiv.org/abs/1710.05758 "'TensorQuant—A Simulation Toolbox for Deep Neural Network Quantization', Loroch et al 2017"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Mixed Precision Training”, Micikevicius et al 2017`{=html} {#micikevicius-et-al-2017-section}
**[`Mixed Precision Training`{=html}](https://arxiv.org/abs/1710.03740#nvidia "'Mixed Precision Training', Micikevicius et al 2017"){.link-annotated
.id-not .include-annotation link-icon="n"
link-icon-type="text,sans,italic" link-icon-color="#77ba00"}**
## `“BitNet: Bit-Regularized Deep Neural Networks”, Raghavan et al 2017`{=html} {#raghavan-et-al-2017-section}
**[`BitNet: Bit-Regularized Deep Neural Networks`{=html}](https://arxiv.org/abs/1708.04788 "'BitNet: Bit-Regularized Deep Neural Networks', Raghavan et al 2017"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Beating Floating Point at Its Own Game: Posit Arithmetic”, Gustafson & Yonemoto 2017`{=html} {#gustafson-yonemoto-2017-section}
**[`Beating Floating Point at its Own Game: Posit Arithmetic`{=html}](/doc/ai/nn/sparsity/low-precision/2017-gustafson.pdf "'Beating Floating Point at its Own Game: Posit Arithmetic', Gustafson & Yonemoto 2017"){.link-annotated
.id-not .include-annotation link-icon="pdf" link-icon-type="svg"
link-icon-color="#f40f02"}**
## `“Bolt: Accelerated Data Mining With Fast Vector Compression”, Blalock & Guttag 2017`{=html} {#blalock-guttag-2017-section}
**[`Bolt: Accelerated Data Mining with Fast Vector Compression`{=html}](https://arxiv.org/abs/1706.10283 "'Bolt: Accelerated Data Mining with Fast Vector Compression', Blalock & Guttag 2017"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Google’s Neural Machine Translation System: Bridging the Gap between Human and Machine Translation”, Wu et al 2016`{=html} {#wu-et-al-2016-1-section}
**[`Google’s Neural Machine Translation System: Bridging the Gap between Human and Machine Translation`{=html}](https://arxiv.org/abs/1609.08144#google "'Google’s Neural Machine Translation System: Bridging the Gap between Human and Machine Translation', Wu et al 2016"){.link-annotated
.id-not .include-annotation link-icon="alphabet" link-icon-type="svg"
link-icon-color="#4285f4"}**
## `“Ternary Neural Networks for Resource-Efficient AI Applications”, Alemdar et al 2016`{=html} {#alemdar-et-al-2016-section}
**[`Ternary Neural Networks for Resource-Efficient AI Applications`{=html}](https://arxiv.org/abs/1609.00222 "'Ternary Neural Networks for Resource-Efficient AI Applications', Alemdar et al 2016"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Deep Neural Networks Are Robust to Weight Binarization and Other Non-Linear Distortions”, Merolla et al 2016`{=html} {#merolla-et-al-2016-section}
**[`Deep neural networks are robust to weight binarization and other non-linear distortions`{=html}](https://arxiv.org/abs/1606.01981 "'Deep neural networks are robust to weight binarization and other non-linear distortions', Merolla et al 2016"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Convolutional Networks for Fast, Energy-Efficient Neuromorphic Computing”, Esser et al 2016`{=html} {#esser-et-al-2016-section}
**[`Convolutional Networks for Fast, Energy-Efficient Neuromorphic Computing`{=html}](https://arxiv.org/abs/1603.08270 "'Convolutional Networks for Fast, Energy-Efficient Neuromorphic Computing', Esser et al 2016"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks”, Rastegari et al 2016`{=html} {#rastegari-et-al-2016-section}
**[`XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks`{=html}](https://arxiv.org/abs/1603.05279 "'XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks', Rastegari et al 2016"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Binarized Neural Networks: Training Deep Neural Networks With Weights and Activations Constrained to +1 or −1”, Courbariaux et al 2016`{=html} {#courbariaux-et-al-2016-section}
**[`Binarized Neural Networks: Training Deep Neural Networks with Weights and Activations Constrained to +1 or −1`{=html}](https://arxiv.org/abs/1602.02830 "'Binarized Neural Networks: Training Deep Neural Networks with Weights and Activations Constrained to +1 or −1', Courbariaux et al 2016"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“BinaryConnect: Training Deep Neural Networks With Binary Weights during Propagations”, Courbariaux et al 2015`{=html} {#courbariaux-et-al-2015-section}
**[`BinaryConnect: Training Deep Neural Networks with binary weights during propagations`{=html}](https://arxiv.org/abs/1511.00363 "'BinaryConnect: Training Deep Neural Networks with binary weights during propagations', Courbariaux et al 2015"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation”, Bengio et al 2013`{=html} {#bengio-et-al-2013-section}
**[`Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation`{=html}](https://arxiv.org/abs/1308.3432 "'Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation', Bengio et al 2013"){.link-annotated
.id-not .include-annotation link-icon="𝛘" link-icon-type="text"
link-icon-color="#b31b1b"}**
## `“Efficient Supervised Learning in Networks With Binary Synapses”, Baldassi et al 2007`{=html} {#baldassi-et-al-2007-section}
**[`Efficient supervised learning in networks with binary synapses`{=html}](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1904114/ "'Efficient supervised learning in networks with binary synapses', Baldassi et al 2007"){.link-annotated
.id-not .include-annotation link-icon="nlm-ncbi" link-icon-type="svg"
link-icon-color="#20558a"}**
## `“A Self-Optimizing, Non-Symmetrical Neural Net for Content Addressable Memory and Pattern Recognition”, Lapedes & Farber 1986`{=html} {#lapedes-farber-1986-section}
**[`A self-optimizing, non-symmetrical neural net for content addressable memory and pattern recognition`{=html}](/doc/ai/nn/rnn/1986-lapedes.pdf "'A self-optimizing, non-symmetrical neural net for content addressable memory and pattern recognition', Lapedes & Farber 1986"){.link-annotated
.id-not .include-annotation link-icon="pdf" link-icon-type="svg"
link-icon-color="#f40f02"}**
## `“Lossless Model Compression Experiment”`{=html} {#_xpQaZzSY-section}
**[`Lossless Model Compression Experiment`{=html}](https://brianbell-x.github.io/weight-compression/ "Lossless Model Compression Experiment"){.link-annotated-partial
.id-not .include-annotation}**
## `“Binary Vector Embeddings Are so Cool”`{=html} {#_i02soBXu-section}
**[`Binary vector embeddings are so cool`{=html}](https://emschwartz.me/binary-vector-embeddings-are-so-cool/ "Binary vector embeddings are so cool"){.link-annotated-partial
.id-not .include-annotation}**
## `“Building a Vector Database in 2GB for 36 Million Wikipedia Passages”`{=html} {#_oc4YkaVm-section}
**[`Building a vector database in 2GB for 36 million Wikipedia passages`{=html}](https://gpt3experiments.substack.com/p/building-a-vector-database-in-2gb "Building a vector database in 2GB for 36 million Wikipedia passages"){.id-not
.include-annotation link-icon="substack" link-icon-type="svg"
link-icon-color="#ff6719"}**
## `“Research Log: Monet/PEER Sparse Experts”`{=html} {#_ANcBzH1V-section}
**[`Research Log: Monet/PEER sparse experts`{=html}](https://www.lesswrong.com/posts/uDrDbeM3CvWYunFkt/research-log-monet-peer-sparse-experts "Research Log: Monet/PEER sparse experts"){.link-annotated-partial
.id-not .include-annotation link-icon="LW" link-icon-type="text"
link-icon-color="#7faf83"}**
## `“FlashAttention-3: Fast and Accurate Attention With Asynchrony and Low-Precision”`{=html} {#_rAsKiX1f-section}
**[`FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision`{=html}](https://www.together.ai/blog/flashattention-3 "FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision"){.link-annotated-partial
.id-not .include-annotation}**
## Sort By Magic
::: {demo-type="sort-by-magic-preface"}
Annotations sorted by machine learning into [inferred \'tags\'](/design#future-tag-features){#gwern-design--future-tag-features
.link-annotated-partial}. This provides an alternative way to browse: instead of by *date* order, one can browse in *topic* order. The \'sorted\' list has been automatically clustered into multiple sections & auto-labeled for easier browsing.
Beginning with the newest annotation, it uses the embedding of each annotation to attempt to create a list of nearest-neighbor annotations, creating a progression of topics. For more details, see the link.
:::
### `score-centering` {.collapse title="Machine-generated tag name for the following cluster of links."}
[\[see previous entry\]](#marek-ryabinin-2026-section){.include
.include-even-when-collapsed}
### `deep-translation` {.collapse title="Machine-generated tag name for the following cluster of links."}
[\[see previous entry\]](#wu-et-al-2016-1-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#zyddnys-2020-section){.include
.include-even-when-collapsed}
### `retro-computing` {.collapse title="Machine-generated tag name for the following cluster of links."}
[\[see previous entry\]](#mordvintsev-niklasson-2021-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#gizmo64k-2026-section){.include
.include-even-when-collapsed}
### `stochastic-optimization evolutionary-strategies conditional-computation gradient-estimation hybrid-evolutionary algorithm-optimization` {.collapse title="Machine-generated tag name for the following cluster of links."}
[\[see previous entry\]](#bengio-et-al-2013-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#sarkar-et-al-2025-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#song-et-al-2021-2-section){.include
.include-even-when-collapsed}
### `efficient-arithmetic low-bit-acceleration energy-optimization posit-computation model-scaling` {.collapse title="Machine-generated tag name for the following cluster of links."}
[\[see previous entry\]](#kubo-et-al-2025-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#luo-sun-2024-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#gustafson-yonemoto-2017-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#mallasén-et-al-2023-section){.include
.include-even-when-collapsed}
### `retrieval-acceleration dense-retrieval vector-compression efficient-inference data-mining` {.collapse title="Machine-generated tag name for the following cluster of links."}
[\[see previous entry\]](#guo-et-al-2019-2-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#blalock-guttag-2017-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#lewis-et-al-2021-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#santhanam-et-al-2022-2-section){.include
.include-even-when-collapsed}
### `efficient-training` {.collapse title="Machine-generated tag name for the following cluster of links."}
[\[see previous entry\]](#zhao-et-al-2024-2-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#kusupati-et-al-2022-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#shu-nakayama-2017-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#du-et-al-2020-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#jaghouar-et-al-2024-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#zheng-et-al-2024-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#gao-et-al-2024-1-section){.include
.include-even-when-collapsed}
### `low-bit-models` {.collapse title="Machine-generated tag name for the following cluster of links."}
[\[see previous entry\]](#lapedes-farber-1986-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#baldassi-et-al-2007-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#neftci-et-al-2019-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#fang-et-al-2021-5-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#ye-et-al-2024-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#merolla-et-al-2016-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#courbariaux-et-al-2016-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#courbariaux-et-al-2015-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#rastegari-et-al-2016-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#mcdonnell-2018-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#cabral-et-al-2025-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#ma-et-al-2024-3-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#ma-et-al-2025-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#raghavan-et-al-2017-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#baalen-et-al-2020-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#dettmers-et-al-2021-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#dettmers-et-al-2022-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#frantar-et-al-2022-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#yao-et-al-2022-2-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#bondarenko-et-al-2021-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#wu-et-al-2022-08-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#zhang-et-al-2020-07-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#zafrir-et-al-2021-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#li-et-al-2020-5-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#rogers-et-al-2020-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#shen-et-al-2022-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#kim-et-al-2022-3-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#fedus-et-al-2021-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#zhu-et-al-2023-3-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#pope-et-al-2022-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#sardana-frankle-2023-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#finzi-et-al-2025-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#allen-zhu-li-2024-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#pudipeddi-et-al-2020-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#zhu-et-al-2024-2-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#xia-et-al-2024-1-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#dettmers-et-al-2023-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#lin-et-al-2023-7-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#xiao-et-al-2022-1-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#liu-et-al-2023-06-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#xi-et-al-2023-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#ghaffari-et-al-2022-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#noune-et-al-2022-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#micikevicius-et-al-2022-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#narayan-et-al-2025-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#johnson-2018-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#hill-et-al-2018-2-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#garland-gregg-2020-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#micikevicius-et-al-2017-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#sa-et-al-2018-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#loroch-et-al-2017-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#stock-et-al-2019-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#fan-et-al-2020-2-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#choi-et-al-2018-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#cséfalvay-imber-2023-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#sun-et-al-2017-1-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#raihan-aamodt-2020-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#wu-et-al-2017-1-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#esser-et-al-2016-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#alemdar-et-al-2016-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#ardakani-et-al-2018-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#shayer-et-al-2017-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#nguyen-et-al-2024-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#diffenderfer-et-al-2021-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#hong-et-al-2024-1-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#huang-et-al-2024-4-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#nolanoorg-2023-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#zeng-et-al-2022-1-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#adler-et-al-2024-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#yuan-et-al-2023-1-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#wu-et-al-2020-3-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#liu-et-al-2023-13-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#zhang-et-al-2023-18-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#lin-et-al-2021-4-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#wei-et-al-2018-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#ma-et-al-2021-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#putter-corporaal-2022-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#lin-et-al-2022-smolml-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#lin-et-al-2017-1-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#lu-sa-2020-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#lu-et-al-2022-7-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#tang-et-al-2021-1bitadam-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#mudigere-et-al-2021-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#jia-et-al-2018-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#howard-2018-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#zhao-et-al-2025-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#jouppi-et-al-2021-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#bailey-2022-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#caballero-et-al-2022-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#menghani-2021-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#lam-et-al-2019-section){.include
.include-even-when-collapsed}
[\[see previous entry\]](#hafner-et-al-2022-section){.include
.include-even-when-collapsed}
## Wikipedia (7) {#titled-links-wikipedia .collapse}
1. **[`Bfloat16 floating-point format`{=html}](https://en.wikipedia.org/wiki/Bfloat16_floating-point_format "Bfloat16 floating-point format"){.id-not
.include-annotation link-icon="wikipedia" link-icon-type="svg"}**
::: aux-links-transclude-file
[`https://en.wikipedia.org/wiki/Bfloat16_floating-point_format`](https://en.wikipedia.org/wiki/Bfloat16_floating-point_format "Bfloat16 floating-point format"){.id-not
.include-content include-template="$annotationFileIncludeTemplate"}
:::
2. **[`Half-precision floating-point format`{=html}](https://en.wikipedia.org/wiki/Half-precision_floating-point_format "Half-precision floating-point format"){.id-not
.include-annotation link-icon="wikipedia" link-icon-type="svg"}**
::: aux-links-transclude-file
[`https://en.wikipedia.org/wiki/Half-precision_floating-point_format`](https://en.wikipedia.org/wiki/Half-precision_floating-point_format "Half-precision floating-point format"){.id-not
.include-content include-template="$annotationFileIncludeTemplate"}
:::
3. **[`Minifloat`{=html}](https://en.wikipedia.org/wiki/Minifloat "Minifloat"){.id-not
.include-annotation link-icon="wikipedia" link-icon-type="svg"}**
::: aux-links-transclude-file
[`https://en.wikipedia.org/wiki/Minifloat`](https://en.wikipedia.org/wiki/Minifloat "Minifloat"){.id-not
.include-content include-template="$annotationFileIncludeTemplate"}
:::
4. **[`Quantization (signal processing)`{=html}](https://en.wikipedia.org/wiki/Quantization_(signal_processing) "Quantization (signal processing)"){.link-annotated-partial
.id-not .include-annotation link-icon="wikipedia"
link-icon-type="svg"}**
5. **[`Random projection`{=html}](https://en.wikipedia.org/wiki/Random_projection "Random projection"){.link-annotated-partial
.id-not .include-annotation link-icon="wikipedia"
link-icon-type="svg"}**
6. **[`Unum (number format) § Unum III`{=html}](https://en.wikipedia.org/wiki/Unum_(number_format)#Unum_III "Unum (number format) § Unum III"){.id-not
.include-annotation link-icon="wikipedia" link-icon-type="svg"}**
::: aux-links-transclude-file
[`https://en.wikipedia.org/wiki/Unum_(number_format)#Unum_III`](https://en.wikipedia.org/wiki/Unum_(number_format)#Unum_III "Unum (number format) § Unum III"){.id-not
.include-content include-template="$annotationFileIncludeTemplate"}
:::
7. **[`Walsh function`{=html}](https://en.wikipedia.org/wiki/Walsh_function "Walsh function"){.id-not
.include-annotation link-icon="wikipedia" link-icon-type="svg"}**
::: aux-links-transclude-file
[`https://en.wikipedia.org/wiki/Walsh_function`](https://en.wikipedia.org/wiki/Walsh_function "Walsh function"){.id-not
.include-content include-template="$annotationFileIncludeTemplate"}
:::
# Miscellaneous
- **[`/doc/ai/nn/sparsity/low-precision/2023-xu-table1-enwik8textpredictionresultsusingspikegptandothertransformerrnnbaselines.jpg`{=html}](/doc/ai/nn/sparsity/low-precision/2023-xu-table1-enwik8textpredictionresultsusingspikegptandothertransformerrnnbaselines.jpg){.link-annotated-partial
.id-not .include-annotation link-icon="image"
link-icon-type="svg"}**
- **[`/doc/ai/nn/sparsity/low-precision/2022-pope-figure1-tpucostvssamplinglatencyofpalm540bmodelonatpucluster.png`{=html}](/doc/ai/nn/sparsity/low-precision/2022-pope-figure1-tpucostvssamplinglatencyofpalm540bmodelonatpucluster.png){.link-annotated-partial
.id-not .include-annotation link-icon="image"
link-icon-type="svg"}**
- **[`/doc/ai/nn/sparsity/low-precision/2021-fedus-figure1-switchmoetransformerscaling.png`{=html}](/doc/ai/nn/sparsity/low-precision/2021-fedus-figure1-switchmoetransformerscaling.png){.link-annotated-partial
.id-not .include-annotation link-icon="image"
link-icon-type="svg"}**
- **[`/doc/ai/nn/sparsity/low-precision/2021-fedus-figure13-switchtransformerknowledgevsreasoningscaling.jpg`{=html}](/doc/ai/nn/sparsity/low-precision/2021-fedus-figure13-switchtransformerknowledgevsreasoningscaling.jpg){.link-annotated-partial
.id-not .include-annotation link-icon="image"
link-icon-type="svg"}**
- **[`https://blog.pgvecto.rs/my-binary-vector-search-is-better-than-your-fp32-vectors`{=html}](https://blog.pgvecto.rs/my-binary-vector-search-is-better-than-your-fp32-vectors){.link-annotated-partial
.id-not .include-annotation}**
- **[`https://cpldcpu.wordpress.com/2024/04/24/implementing-neural-networks-on-the-10-cent-risc-v-mcu-without-multiplier/`{=html}](https://cpldcpu.wordpress.com/2024/04/24/implementing-neural-networks-on-the-10-cent-risc-v-mcu-without-multiplier/){.link-annotated-partial
.id-not .include-annotation}**
- **[`https://github.com/NolanoOrg/llama-int4-quant/`{=html}](https://github.com/NolanoOrg/llama-int4-quant/){.id-not
.include-annotation link-icon="github" link-icon-type="svg"}**
- **[`https://github.com/THUDM/GLM-130B/blob/main/doc/quantization.md`{=html}](https://github.com/THUDM/GLM-130B/blob/main/doc/quantization.md){.id-not
.include-annotation link-icon="github" link-icon-type="svg"}**
- **[`https://github.com/qwopqwop200/GPTQ-for-LLaMa`{=html}](https://github.com/qwopqwop200/GPTQ-for-LLaMa){.id-not
.include-annotation link-icon="github" link-icon-type="svg"}**
- **[`https://github.com/vitoplantamura/OnnxStream`{=html}](https://github.com/vitoplantamura/OnnxStream){.id-not
.include-annotation link-icon="github" link-icon-type="svg"}**
- **[`https://github.com/vitoplantamura/OnnxStream/tree/846da873570a737b49154e8f835704264864b0fe`{=html}](https://github.com/vitoplantamura/OnnxStream/tree/846da873570a737b49154e8f835704264864b0fe){.id-not
.include-annotation link-icon="github" link-icon-type="svg"}**
- **[`https://huggingface.co/blog/embedding-quantization`{=html}](https://huggingface.co/blog/embedding-quantization){.link-annotated-partial
.id-not .include-annotation link-icon="🤗" link-icon-type="text"
link-icon-color="#ffcc4d"}**
- **[`https://itnext.io/shrinking-a-language-detection-model-to-under-10-kb-b729bc25fd28`{=html}](https://itnext.io/shrinking-a-language-detection-model-to-under-10-kb-b729bc25fd28){.id-not
.include-annotation}**
- **[`https://justine.lol/matmul/`{=html}](https://justine.lol/matmul/){.link-annotated-partial
.id-not .include-annotation}**
- **[`https://lightning.ai/pages/community/lora-insights/`{=html}](https://lightning.ai/pages/community/lora-insights/){.id-not
.include-annotation}**
- **[`https://observablehq.com/@rreusser/half-precision-floating-point-visualized`{=html}](https://observablehq.com/@rreusser/half-precision-floating-point-visualized){.link-annotated-partial
.id-not .include-annotation}**
- **[`https://research.google/blog/quantization-for-fast-and-environmentally-sustainable-reinforcement-learning/`{=html}](https://research.google/blog/quantization-for-fast-and-environmentally-sustainable-reinforcement-learning/){.id-not
.include-annotation link-icon="alphabet" link-icon-type="svg"
link-icon-color="#4285f4"}**
- **[`https://txt.cohere.com/int8-binary-embeddings/`{=html}](https://txt.cohere.com/int8-binary-embeddings/){.link-annotated-partial
.id-not .include-annotation}**
- **[`https://www.reddit.com/r/LocalLLaMA/comments/1gsyp7q/humaneval_benchmark_of_exl2_quants_of_popular/`{=html}](https://www.reddit.com/r/LocalLLaMA/comments/1gsyp7q/humaneval_benchmark_of_exl2_quants_of_popular/){.link-annotated-partial
.id-not .include-annotation link-icon="reddit" link-icon-type="svg"
link-icon-color="#ff4500"}**
- **[`https://www.reddit.com/r/mlscaling/comments/146rgq2/chatgpt_is_running_quantized/`{=html}](https://www.reddit.com/r/mlscaling/comments/146rgq2/chatgpt_is_running_quantized/){.link-annotated-partial
.id-not .include-annotation link-icon="reddit" link-icon-type="svg"
link-icon-color="#ff4500"}**
- **[`https://x.com/NolanoOrg/status/1634027966651834370`{=html}](https://x.com/NolanoOrg/status/1634027966651834370 "NolanoOrg 2026"){.link-annotated-partial
.id-not .include-annotation link-icon="twitter" link-icon-type="svg"
link-icon-color="#1da1f2"}**
- **[`https://x.com/aidan_mclau/status/1822830757137596521`{=html}](https://x.com/aidan_mclau/status/1822830757137596521 "McLau 2026"){.id-not
.include-annotation link-icon="twitter" link-icon-type="svg"
link-icon-color="#1da1f2"}**
::: aux-links-transclude-file
[`https://x.com/aidan_mclau/status/1822830757137596521`](https://x.com/aidan_mclau/status/1822830757137596521 "McLau 2026"){.id-not
.include-content}
:::
- **[`https://x.com/davisblalock/status/1939956579698094166`{=html}](https://x.com/davisblalock/status/1939956579698094166 "Blalock 2026"){.id-not
.include-annotation link-icon="twitter" link-icon-type="svg"
link-icon-color="#1da1f2"}**
::: aux-links-transclude-file
[`https://x.com/davisblalock/status/1939956579698094166`](https://x.com/davisblalock/status/1939956579698094166 "Blalock 2026"){.id-not
.include-content}
:::
- **[`https://x.com/moyix/status/1582213498703990784`{=html}](https://x.com/moyix/status/1582213498703990784 "moyix 2026"){.link-annotated-partial
.id-not .include-annotation link-icon="twitter" link-icon-type="svg"
link-icon-color="#1da1f2"}**
- **[`https://x.com/thecharlieblake/status/1581913495670755328`{=html}](https://x.com/thecharlieblake/status/1581913495670755328 "thecharlieblake 2026"){.link-annotated-partial
.id-not .include-annotation link-icon="twitter" link-icon-type="svg"
link-icon-color="#1da1f2"}**
- **[`https://x.com/thiteanish/status/1635188333705043969`{=html}](https://x.com/thiteanish/status/1635188333705043969 "thiteanish 2026"){.link-annotated-partial
.id-not .include-annotation link-icon="twitter" link-icon-type="svg"
link-icon-color="#1da1f2"}**
# Bibliography {#link-bibliography-section}
1. `https://arxiv.org/abs/2511.16652`:
[`“EGGROLL: Evolution Strategies at the Hyperscale”`{=html}](https://arxiv.org/abs/2511.16652 "'EGGROLL: Evolution Strategies at the Hyperscale', Sarkar et al 2025"){#sarkar-et-al-2025
.link-annotated},
[Bidipta Sarkar, Mattie Fellows, Juan Agustin Duque[, Alistair Letcher, Antonio León Villares, Anya Sims, Dylan Cope, Jarek Liesen, Lukas Seier, Theo Wolf, Uljad Berdica, Alexander David Goldie, [Aaron Courville](https://scholar.google.com/citations?user=km6CP8cAAAAJ "Aaron Courville"){.id-not}, Karin Sevegnani, Shimon Whiteson, Jakob Nicolaus Foerster]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2511.16652.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2511.16652"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
2. `https://arxiv.org/abs/2505.09343#deepseek`:
[`“Insights into DeepSeek-V3: Scaling Challenges and Reflections on Hardware for AI Architectures”`{=html}](https://arxiv.org/abs/2505.09343#deepseek "'Insights into DeepSeek-V3: Scaling Challenges and Reflections on Hardware for AI Architectures', Zhao et al 2025"){#zhao-et-al-2025
.link-annotated},
[Chenggang Zhao, Chengqi Deng, Chong Ruan[, Damai Dai, Huazuo Gao, Jiashi Li, Liyue Zhang, Panpan Huang, Shangyan Zhou, Shirong Ma, [Liang Wenfeng](https://en.wikipedia.org/wiki/Liang_Wenfeng "Liang Wenfeng"){.id-not}, Ying He, Yuqing Wang, Yuxuan Liu, Y. X. Wei]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2505.09343%2523deepseek.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2505.09343#deepseek"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
3. `https://arxiv.org/abs/2406.11233`:
[`“Probing the Decision Boundaries of In-Context Learning in Large Language Models”`{=html}](https://arxiv.org/abs/2406.11233 "'Probing the Decision Boundaries of In-context Learning in Large Language Models', Zhao et al 2024"){#zhao-et-al-2024-2
.link-annotated},
[Siyan Zhao, Tung Nguyen, [Aditya Grover](https://scholar.google.com/citations?user=oOhnPUgAAAAJ "Aditya Grover"){.id-not}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2406.11233.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2406.11233"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
4. `https://arxiv.org/abs/2404.14047`:
[`“How Good Are Low-Bit Quantized LLaMA-3 Models? An Empirical Study”`{=html}](https://arxiv.org/abs/2404.14047 "'How Good Are Low-bit Quantized LLaMA-3 Models? An Empirical Study', Huang et al 2024"){#huang-et-al-2024-4
.link-annotated},
[[Wei Huang](https://en.wikipedia.org/wiki/Wei_Huang "Wei Huang"){.id-not}, Xudong Ma, Haotong Qin[, Xingyu Zheng, Chengtao Lv, Hong Chen, Jie Luo, Xiaojuan Qi, Xianglong Liu, Michele Magno]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2404.14047.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2404.14047"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
5. `https://arxiv.org/abs/2401.14112#microsoft`:
[`“FP6-LLM: Efficiently Serving Large Language Models Through FP6-Centric Algorithm-System Co-Design”`{=html}](https://arxiv.org/abs/2401.14112#microsoft "'FP6-LLM: Efficiently Serving Large Language Models Through FP6-Centric Algorithm-System Co-Design', Xia et al 2024"){#xia-et-al-2024-1
.link-annotated},
[Haojun Xia, Zhen Zheng, Xiaoxia Wu[, Shiyang Chen, Zhewei Yao, Stephen Youn, Arash Bakhtiari, Michael Wyatt, Donglin Zhuang, Zhongzhu Zhou, Olatunji Ruwase, Yuxiong He, Shuaiwen Leon Song]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2401.14112%2523microsoft.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2401.14112#microsoft"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
6. `https://arxiv.org/abs/2312.16862`:
[`“TinyGPT-V: Efficient Multimodal Large Language Model via Small Backbones”`{=html}](https://arxiv.org/abs/2312.16862 "'TinyGPT-V: Efficient Multimodal Large Language Model via Small Backbones', Yuan et al 2023"){#yuan-et-al-2023-1
.link-annotated}, [Zhengqing Yuan, Zhaoxu Li, Lichao Sun]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2312.16862.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2312.16862"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
7. `https://arxiv.org/abs/2310.16836`:
[`“LLM-FP4: 4-Bit Floating-Point Quantized Transformers”`{=html}](https://arxiv.org/abs/2310.16836 "'LLM-FP4: 4-Bit Floating-Point Quantized Transformers', Liu et al 2023"){#liu-et-al-2023-06
.link-annotated},
[Shih-yang Liu, Zechun Liu, Xijie Huang[, Pingcheng Dong, Kwang-Ting Cheng]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2310.16836.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2310.16836"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
8. `https://arxiv.org/abs/2305.06946`:
[`“Big-PERCIVAL: Exploring the Native Use of 64-Bit Posit Arithmetic in Scientific Computing”`{=html}](https://arxiv.org/abs/2305.06946 "'Big-PERCIVAL: Exploring the Native Use of 64-Bit Posit Arithmetic in Scientific Computing', Mallasén et al 2023"){#mallasén-et-al-2023
.link-annotated},
[David Mallasén, Alberto A. Del Barrio, Manuel Prieto-Matias]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2305.06946.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2305.06946"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
9. `https://nolanoorg.substack.com/p/int-4-llama-is-not-enough-int-3-and`:
[`“Int-4 LLaMa Is Not Enough—Int-3 and Beyond: More Compression, Easier to Build Apps on LLMs That Run Locally”`{=html}](https://nolanoorg.substack.com/p/int-4-llama-is-not-enough-int-3-and "'Int-4 LLaMa is not enough—Int-3 and beyond: More compression, easier to build apps on LLMs that run locally', nolano.org 2023"){#nolanoorg-2023
.link-annotated}, [nolano.org]{.author .cite-author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Fnolanoorg.substack.com%252Fp%252Fint-4-llama-is-not-enough-int-3-and.html "Directory-tag link-bibliography for link https://nolanoorg.substack.com/p/int-4-llama-is-not-enough-int-3-and"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
10. `https://arxiv.org/abs/2302.13939`:
[`“SpikeGPT: Generative Pre-Trained Language Model With Spiking Neural Networks”`{=html}](https://arxiv.org/abs/2302.13939 "'SpikeGPT: Generative Pre-trained Language Model with Spiking Neural Networks', Zhu et al 2023"){#zhu-et-al-2023-3
.link-annotated},
[Rui-Jie Zhu, Qihang Zhao, Jason K. Eshraghian]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2302.13939.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2302.13939"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
11. `https://arxiv.org/abs/2211.10438`:
[`“SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models”`{=html}](https://arxiv.org/abs/2211.10438 "'SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models', Xiao et al 2022"){#xiao-et-al-2022-1
.link-annotated},
[Guangxuan Xiao, Ji Lin, Mickael Seznec[, Julien Demouth, [Song Han](https://hanlab.mit.edu/songhan "Song Han—Associate Professor, MIT EECS"){.link-annotated-partial
.id-not}]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2211.10438.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2211.10438"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
12. `https://arxiv.org/abs/2211.05102#google`:
[`“Efficiently Scaling Transformer Inference”`{=html}](https://arxiv.org/abs/2211.05102#google "'Efficiently Scaling Transformer Inference', Pope et al 2022"){#pope-et-al-2022
.link-annotated},
[Reiner Pope, [Sholto Douglas](https://en.wikipedia.org/wiki/Sholto_Douglas "Sholto Douglas"){.id-not}, Aakanksha Chowdhery[, Jacob Devlin, [James Bradbury](https://x.com/jekbradbury "Bradbury 2026"){.id-not}, Anselm Levskaya, Jonathan Heek, Kefan Xiao, Shivani Agrawal, [Jeff Dean](https://en.wikipedia.org/wiki/Jeff_Dean "Jeff Dean"){.link-annotated-partial
.id-not}]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2211.05102%2523google.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2211.05102#google"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
13. `https://arxiv.org/abs/2210.17323`:
[`“GPTQ: Accurate Post-Training Quantization for Generative Pre-Trained Transformers”`{=html}](https://arxiv.org/abs/2210.17323 "'GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers', Frantar et al 2022"){#frantar-et-al-2022
.link-annotated},
[Elias Frantar, Saleh Ashkboos, [Torsten Hoefler](https://en.wikipedia.org/wiki/Torsten_Hoefler "Torsten Hoefler"){.id-not}, Dan Alistarh]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2210.17323.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2210.17323"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
14. `https://arxiv.org/abs/2210.02414#zhipu`:
[`“GLM-130B: An Open Bilingual Pre-Trained Model”`{=html}](https://arxiv.org/abs/2210.02414#zhipu "'GLM-130B: An Open Bilingual Pre-trained Model', Zeng et al 2022"){#zeng-et-al-2022-1
.link-annotated},
[Aohan Zeng, Xiao Liu, Zhengxiao Du[, Zihan Wang, Hanyu Lai, [Ming Ding](https://scholar.google.com/citations?user=Va50YzkAAAAJ "Ming Ding"){.id-not}, Zhuoyi Yang, [Yifan Xu](https://en.wikipedia.org/wiki/Yifan_Xu "Yifan Xu"){.id-not}, Wendi Zheng, Xiao Xia, Weng Lam Tam, Zixuan Ma, Yufei Xue, Jidong Zhai, Wenguang Chen, [Peng Zhang](https://sites.google.com/site/pengzhang27182/ "Peng Zhang’s homepage"){.id-not}, Yuxiao Dong, [Jie Tang](https://en.wikipedia.org/wiki/Jie_Tang "Jie Tang"){.id-not}]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2210.02414%2523zhipu.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2210.02414#zhipu"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
15. `https://arxiv.org/abs/2206.15472`:
[`“On-Device Training Under 256KB Memory”`{=html}](https://arxiv.org/abs/2206.15472 "'On-Device Training Under 256KB Memory', Lin et al 2022"){#lin-et-al-2022-smolml
.link-annotated},
[Ji Lin, Ligeng Zhu, Wei-Ming Chen[, Wei-Chen Wang, Chuang Gan, [Song Han](https://hanlab.mit.edu/songhan "Song Han—Associate Professor, MIT EECS"){.link-annotated-partial
.id-not}]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2206.15472.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2206.15472"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
16. `https://arxiv.org/abs/2206.04114#google`:
[`“Director: Deep Hierarchical Planning from Pixels”`{=html}](https://arxiv.org/abs/2206.04114#google "'Director: Deep Hierarchical Planning from Pixels', Hafner et al 2022"){#hafner-et-al-2022
.link-annotated},
[Danijar Hafner, Kuang-Huei Lee, Ian Fischer, [Pieter Abbeel](https://en.wikipedia.org/wiki/Pieter_Abbeel "Pieter Abbeel"){.id-not}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2206.04114%2523google.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2206.04114#google"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
17. `https://arxiv.org/abs/2206.01859#microsoft`:
[`“XTC: Extreme Compression for Pre-Trained Transformers Made Simple and Efficient”`{=html}](https://arxiv.org/abs/2206.01859#microsoft "'XTC: Extreme Compression for Pre-trained Transformers Made Simple and Efficient', Wu et al 2022"){#wu-et-al-2022-08
.link-annotated},
[Xiaoxia Wu, Zhewei Yao, Minjia Zhang[, Conglong Li, Yuxiong He]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2206.01859%2523microsoft.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2206.01859#microsoft"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
18. `https://arxiv.org/abs/2206.01861#microsoft`:
[`“ZeroQuant: Efficient and Affordable Post-Training Quantization for Large-Scale Transformers”`{=html}](https://arxiv.org/abs/2206.01861#microsoft "'ZeroQuant: Efficient and Affordable Post-Training Quantization for Large-Scale Transformers', Yao et al 2022"){#yao-et-al-2022-2
.link-annotated},
[Zhewei Yao, Reza Yazdani Aminabadi, Minjia Zhang[, Xiaoxia Wu, Conglong Li, Yuxiong He]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2206.01861%2523microsoft.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2206.01861#microsoft"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
19. `https://arxiv.org/abs/2205.13147`:
[`“Matryoshka Representations for Adaptive Deployment”`{=html}](https://arxiv.org/abs/2205.13147 "'Matryoshka Representations for Adaptive Deployment', Kusupati et al 2022"){#kusupati-et-al-2022
.link-annotated},
[Aditya Kusupati, Gantavya Bhatt, Aniket Rege[, Matthew Wallingford, [Aditya Sinha](https://en.wikipedia.org/wiki/Aditya_Sinha "Aditya Sinha"){.id-not}, Vivek Ramanujan, William Howard-Snyder, Kaifeng Chen, [Sham Kakade](https://en.wikipedia.org/wiki/Sham_Kakade "Sham Kakade"){.id-not}, Prateek Jain, [Ali Farhadi](https://en.wikipedia.org/wiki/Ali_Farhadi "Ali Farhadi"){.id-not}]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2205.13147.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2205.13147"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
20. `https://arxiv.org/abs/2202.06009#microsoft`:
[`“Maximizing Communication Efficiency for Large-Scale Training via 0/1 Adam”`{=html}](https://arxiv.org/abs/2202.06009#microsoft "'Maximizing Communication Efficiency for Large-scale Training via 0/1 Adam', Lu et al 2022"){#lu-et-al-2022-7
.link-annotated},
[Yucheng Lu, Conglong Li, Minjia Zhang[, Christopher De Sa, Yuxiong He]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2202.06009%2523microsoft.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2202.06009#microsoft"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
21. `https://semiengineering.com/is-programmable-overhead-worth-the-cost/`:
[`“Is Programmable Overhead Worth The Cost? How Much Do We Pay for a System to Be Programmable? It Depends upon Who You Ask”`{=html}](https://semiengineering.com/is-programmable-overhead-worth-the-cost/ "'Is Programmable Overhead Worth The Cost? How much do we pay for a system to be programmable? It depends upon who you ask', Bailey 2022"){#bailey-2022
.link-annotated}, [Brian Bailey]{.author .cite-author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Fsemiengineering.com%252Fis-programmable-overhead-worth-the-cost%252F.html "Directory-tag link-bibliography for link https://semiengineering.com/is-programmable-overhead-worth-the-cost/"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
22. `https://arxiv.org/abs/2111.13824`:
[`“FQ-ViT: Fully Quantized Vision Transformer without Retraining”`{=html}](https://arxiv.org/abs/2111.13824 "'FQ-ViT: Fully Quantized Vision Transformer without Retraining', Lin et al 2021"){#lin-et-al-2021-4
.link-annotated},
[Yang Lin, Tianyu Zhang, Peiqin Sun[, Zheng Li, Shuchang Zhou]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2111.13824.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2111.13824"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
23. `https://arxiv.org/abs/2111.05754`:
[`“Prune Once for All: Sparse Pre-Trained Language Models”`{=html}](https://arxiv.org/abs/2111.05754 "'Prune Once for All: Sparse Pre-Trained Language Models', Zafrir et al 2021"){#zafrir-et-al-2021
.link-annotated},
[Ofir Zafrir, Ariel Larey, Guy Boudoukh[, Haihao Shen, Moshe Wasserblat]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2111.05754.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2111.05754"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
24. `https://arxiv.org/abs/2110.02861`:
[`“8-Bit Optimizers via Block-Wise Quantization”`{=html}](https://arxiv.org/abs/2110.02861 "'8-bit Optimizers via Block-wise Quantization', Dettmers et al 2021"){#dettmers-et-al-2021
.link-annotated},
[Tim Dettmers, [Mike Lewis](https://scholar.google.com/citations?user=SnQnQicAAAAJ "Mike Lewis"){.id-not}, Sam Shleifer, [Luke Zettlemoyer](https://scholar.google.com/citations?user=UjpbO6IAAAAJ "Luke Zettlemoyer"){.id-not}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2110.02861.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2110.02861"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
25. `https://arxiv.org/abs/2109.12948`:
[`“Understanding and Overcoming the Challenges of Efficient Transformer Quantization”`{=html}](https://arxiv.org/abs/2109.12948 "'Understanding and Overcoming the Challenges of Efficient Transformer Quantization', Bondarenko et al 2021"){#bondarenko-et-al-2021
.link-annotated},
[Yelysei Bondarenko, Markus Nagel, Tijmen Blankevoort]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2109.12948.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2109.12948"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
26. `2021-jouppi.pdf`:
[`“Ten Lessons From Three Generations Shaped Google’s TPUv4i”`{=html}](/doc/ai/scaling/hardware/2021-jouppi.pdf "'Ten Lessons From Three Generations Shaped Google’s TPUv4i', Jouppi et al 2021"){#jouppi-et-al-2021
.link-annotated},
[[Norman P. Jouppi](https://en.wikipedia.org/wiki/Norman_P._Jouppi "Norman P. Jouppi"){.id-not}, Doe Hyun Yoon, Matthew Ashcraft[, Mark Gottscho, Thomas B. Jablin, [George Kurian](https://en.wikipedia.org/wiki/George_Kurian "George Kurian"){.id-not}, James Laudon, Sheng Li, Peter Ma, Xiaoyu Ma, Thomas Norrie, Nishant Patil, Sushma Prasad, Cliff Young, Zongwei Zhou, David Patterson]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/%252Fdoc%252Fai%252Fscaling%252Fhardware%252F2021-jouppi.pdf.html "Directory-tag link-bibliography for link /doc/ai/scaling/hardware/2021-jouppi.pdf"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
27. `https://arxiv.org/abs/2102.04159`:
[`“Deep Residual Learning in Spiking Neural Networks”`{=html}](https://arxiv.org/abs/2102.04159 "'Deep Residual Learning in Spiking Neural Networks', Fang et al 2021"){#fang-et-al-2021-5
.link-annotated},
[[Wei Fang](https://en.wikipedia.org/wiki/Wei_Fang "Wei Fang"){.id-not}, Zhaofei Yu, Yanqi Chen[, Tiejun Huang, Timothée Masquelier, Yonghong Tian]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2102.04159.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2102.04159"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
28. `https://arxiv.org/abs/2102.02888#microsoft`:
[`“1-Bit Adam: Communication Efficient Large-Scale Training With Adam’s Convergence Speed”`{=html}](https://arxiv.org/abs/2102.02888#microsoft "'1-bit Adam: Communication Efficient Large-Scale Training with Adam’s Convergence Speed', Tang et al 2021"){#tang-et-al-2021-1bitadam
.link-annotated},
[Hanlin Tang, Shaoduo Gan, Ammar Ahmad Awan[, Samyam Rajbhandari, Conglong Li, Xiangru Lian, Ji Liu, Ce Zhang, Yuxiong He]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2102.02888%2523microsoft.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2102.02888#microsoft"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
29. `https://arxiv.org/abs/2101.03961#google`:
[`“Switch Transformers: Scaling to Trillion Parameter Models With Simple and Efficient Sparsity”`{=html}](https://arxiv.org/abs/2101.03961#google "'Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity', Fedus et al 2021"){#fedus-et-al-2021
.link-annotated},
[[William Fedus](https://scholar.google.com/citations?user=-ZfwQOkAAAAJ "William Fedus"){.id-not}, [Barret Zoph](https://barretzoph.github.io/ "Barret Zoph"){.link-annotated-partial
.id-not}, [Noam Shazeer](https://en.wikipedia.org/wiki/Noam_Shazeer "Noam Shazeer"){.id-not}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2101.03961%2523google.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2101.03961#google"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
30. `https://arxiv.org/abs/2004.07320#facebook`:
[`“Training With Quantization Noise for Extreme Model Compression”`{=html}](https://arxiv.org/abs/2004.07320#facebook "'Training with Quantization Noise for Extreme Model Compression', Fan et al 2020"){#fan-et-al-2020-2
.link-annotated},
[Angela Fan, Pierre Stock, [Benjamin Graham](https://en.wikipedia.org/wiki/Benjamin_Graham "Benjamin Graham"){.id-not}[, Edouard Grave, [Remi Gribonval](https://en.wikipedia.org/wiki/Remi_Gribonval "Remi Gribonval"){.id-not}, Herve Jegou, [Armand Joulin](https://scholar.google.com/citations?user=kRJkDakAAAAJ "Armand Joulin"){.id-not}]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2004.07320%2523facebook.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2004.07320#facebook"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
31. `https://arxiv.org/abs/2001.01969`:
[`“SWAT: Sparse Weight Activation Training”`{=html}](https://arxiv.org/abs/2001.01969 "'SWAT: Sparse Weight Activation Training', Raihan & Aamodt 2020"){#raihan-aamodt-2020
.link-annotated}, [Md Aamir Raihan, Tor M. Aamodt]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F2001.01969.html "Directory-tag link-bibliography for link https://arxiv.org/abs/2001.01969"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
32. `https://arxiv.org/abs/1910.01055#google`:
[`“QUARL: Quantized Reinforcement Learning (ActorQ)”`{=html}](https://arxiv.org/abs/1910.01055#google "'QUARL: Quantized Reinforcement Learning (ActorQ)', Lam et al 2019"){#lam-et-al-2019
.link-annotated},
[Maximilian Lam, Sharad Chitlangia, Srivatsan Krishnan[, Zishen Wan, Gabriel Barth-Maron, [Aleksandra Faust](https://en.wikipedia.org/wiki/Aleksandra_Faust "Aleksandra Faust"){.id-not}, Vijay Janapa Reddi]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F1910.01055%2523google.html "Directory-tag link-bibliography for link https://arxiv.org/abs/1910.01055#google"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
33. `https://www.fast.ai/2018/04/30/dawnbench-fastai/`:
[`“Training ImageNet in 3 Hours for $25; and CIFAR-10 for $0.26”`{=html}](https://www.fast.ai/2018/04/30/dawnbench-fastai/ "'Training ImageNet in 3 hours for $25; and CIFAR-10 for $0.26', Howard 2018"){#howard-2018
.link-annotated}, [Jeremy Howard]{.author .cite-author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Fwww.fast.ai%252F2018%252F04%252F30%252Fdawnbench-fastai%252F.html "Directory-tag link-bibliography for link https://www.fast.ai/2018/04/30/dawnbench-fastai/"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
34. `https://arxiv.org/abs/1802.08530`:
[`“Training Wide Residual Networks for Deployment Using a Single Bit for Each Weight”`{=html}](https://arxiv.org/abs/1802.08530 "'Training wide residual networks for deployment using a single bit for each weight', McDonnell 2018"){#mcdonnell-2018
.link-annotated},
[[Mark D. McDonnell](https://en.wikipedia.org/wiki/Mark_D._McDonnell "Mark D. McDonnell"){.id-not}]{.author
.cite-author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F1802.08530.html "Directory-tag link-bibliography for link https://arxiv.org/abs/1802.08530"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
35. `https://arxiv.org/abs/1712.01887`:
[`“Deep Gradient Compression: Reducing the Communication Bandwidth for Distributed Training”`{=html}](https://arxiv.org/abs/1712.01887 "'Deep Gradient Compression: Reducing the Communication Bandwidth for Distributed Training', Lin et al 2017"){#lin-et-al-2017-1
.link-annotated},
[Yujun Lin, [Song Han](https://hanlab.mit.edu/songhan "Song Han—Associate Professor, MIT EECS"){.link-annotated-partial
.id-not}, Huizi Mao[, Yu Wang, [William J. Dally](https://en.wikipedia.org/wiki/William_J._Dally "William J. Dally"){.id-not}]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F1712.01887.html "Directory-tag link-bibliography for link https://arxiv.org/abs/1712.01887"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
36. `https://arxiv.org/abs/1711.08141`:
[`“Shift: A Zero FLOP, Zero Parameter Alternative to Spatial Convolutions”`{=html}](https://arxiv.org/abs/1711.08141 "'Shift: A Zero FLOP, Zero Parameter Alternative to Spatial Convolutions', Wu et al 2017"){#wu-et-al-2017-1
.link-annotated},
[Bichen Wu, Alvin Wan, Xiangyu Yue[, Peter Jin, Sicheng Zhao, Noah Golmant, Amir Gholaminejad, Joseph Gonzalez, [Kurt Keutzer](https://en.wikipedia.org/wiki/Kurt_Keutzer "Kurt Keutzer"){.id-not}]{.collapse}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F1711.08141.html "Directory-tag link-bibliography for link https://arxiv.org/abs/1711.08141"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}
37. `https://arxiv.org/abs/1603.05279`:
[`“XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks”`{=html}](https://arxiv.org/abs/1603.05279 "'XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks', Rastegari et al 2016"){#rastegari-et-al-2016
.link-annotated},
[Mohammad Rastegari, Vicente Ordonez, [Joseph Redmon](https://pjreddie.com/ "Survival Strategies for the Robot Rebellion"){.link-annotated-partial
.id-not}, [Ali Farhadi](https://en.wikipedia.org/wiki/Ali_Farhadi "Ali Farhadi"){.id-not}]{.author}
[[link-bibliography](/metadata/annotation/link-bibliography/https%253A%252F%252Farxiv.org%252Fabs%252F1603.05279.html "Directory-tag link-bibliography for link https://arxiv.org/abs/1603.05279"){.id-not
.include-even-when-collapsed}]{.collapse
.tag-index-link-bibliography-block}