Youneng Bao, Yulong Cheng, Yiping Liu, Yichen Yang, Peng Qin, Mu Li, Yongsheng Liang: DynaQuant: Dynamic Mixed-Precision Quantization for Learned Image Compression https://arxiv.org/abs/2511.07903 https://arxiv.org/pdf/2511.07903 https://arxiv.org/html/2511.07903
November 12, 2025 at 6:35 AM
Youneng Bao, Yulong Cheng, Yiping Liu, Yichen Yang, Peng Qin, Mu Li, Yongsheng Liang: DynaQuant: Dynamic Mixed-Precision Quantization for Learned Image Compression https://arxiv.org/abs/2511.07903 https://arxiv.org/pdf/2511.07903 https://arxiv.org/html/2511.07903
Sunghyun Wee, Suyoung Kim, Hyeonjin Kim, Kyomin Hwang, Nojun Kwak: Alignment-Aware Quantization for LLM Safety https://arxiv.org/abs/2511.07842 https://arxiv.org/pdf/2511.07842 https://arxiv.org/html/2511.07842
November 12, 2025 at 6:29 AM
Sunghyun Wee, Suyoung Kim, Hyeonjin Kim, Kyomin Hwang, Nojun Kwak: Alignment-Aware Quantization for LLM Safety https://arxiv.org/abs/2511.07842 https://arxiv.org/pdf/2511.07842 https://arxiv.org/html/2511.07842
- Relu^2 activation function
- FSDP + TP + SP
- Int6 gradient communication
- Quantization Aware Training (QAT)
blog.character.ai/technical/in...
- FSDP + TP + SP
- Int6 gradient communication
- Quantization Aware Training (QAT)
blog.character.ai/technical/in...
Inside Kaiju - building conversational models at scale
What made Character.ai's early models so engaging? Before open-source models became the norm, our team built Kaiju - a family of in-house LLMs designed to power millions of fast, expressive conversati...
blog.character.ai
November 12, 2025 at 5:39 AM
- Relu^2 activation function
- FSDP + TP + SP
- Int6 gradient communication
- Quantization Aware Training (QAT)
blog.character.ai/technical/in...
- FSDP + TP + SP
- Int6 gradient communication
- Quantization Aware Training (QAT)
blog.character.ai/technical/in...
[2025-11-12] 📚 Updates in #MQ
(1) GDNSQ: Gradual Differentiable Noise Scale Quantization for Low-bit Neural Networks
(2) Learning Quantized Continuous Controllers for Integer Hardware
🔍 More at researchtrend.ai/communities/MQ
(1) GDNSQ: Gradual Differentiable Noise Scale Quantization for Low-bit Neural Networks
(2) Learning Quantized Continuous Controllers for Integer Hardware
🔍 More at researchtrend.ai/communities/MQ
November 12, 2025 at 3:10 AM
Getting some interesting results with #CHOPs for #camerashake. The main idea here is to re-use the positional noise for rotation to simulate the camera's inertia. #Houdini #animation
November 12, 2025 at 3:03 AM
Getting some interesting results with #CHOPs for #camerashake. The main idea here is to re-use the positional noise for rotation to simulate the camera's inertia. #Houdini #animation
November 12, 2025 at 12:48 AM
#de-quantization#Tags: hyperbolic geometry, negative curvature, public participation art, sculpture
November 12, 2025 at 12:22 AM
#de-quantization#Tags: hyperbolic geometry, negative curvature, public participation art, sculpture
Looking Back on Our Shared Digital History: “The Web We’ve Built” Mini-Doc | Internet Archive Blogs
blog.archive.org
November 11, 2025 at 9:01 PM
- Software squeezes more from old silicon: Quantization, pruning, MoE, compiler/driver gains, and better schedulers improve tokens-per-watt on older gear—again making it worth running longer.
November 11, 2025 at 7:34 PM
- Software squeezes more from old silicon: Quantization, pruning, MoE, compiler/driver gains, and better schedulers improve tokens-per-watt on older gear—again making it worth running longer.
Qm/mm Molecular Dynamics with Few-Mode Quantization Simulates Light-Matter Interactions at the Nanoscale
Read more:
https://quantumzeitgeist.com/molecular-dynamics-few-mode-quantization-simulates-light-matter/
Read more:
https://quantumzeitgeist.com/molecular-dynamics-few-mode-quantization-simulates-light-matter/
Qm/mm Molecular Dynamics With Few-Mode Quantization Simulates Light-Matter Interactions At The Nanoscale
Researchers have developed a computational method that accurately models how light interacts with molecules positioned near nanoscale structures, revealing that strong coupling between light and matter remains possible even with molecular complexity and disorder, and demonstrating how this interaction shapes energy transfer between molecules.
quantumzeitgeist.com
November 11, 2025 at 4:53 PM
Qm/mm Molecular Dynamics with Few-Mode Quantization Simulates Light-Matter Interactions at the Nanoscale
Read more:
https://quantumzeitgeist.com/molecular-dynamics-few-mode-quantization-simulates-light-matter/
Read more:
https://quantumzeitgeist.com/molecular-dynamics-few-mode-quantization-simulates-light-matter/
Zhaoyang Wang, Dong Wang
GABFusion: Rethinking Feature Fusion for Low-Bit Quantization of Multi-Task Networks
https://arxiv.org/abs/2511.05898
GABFusion: Rethinking Feature Fusion for Low-Bit Quantization of Multi-Task Networks
https://arxiv.org/abs/2511.05898
November 11, 2025 at 2:44 PM
Zhaoyang Wang, Dong Wang
GABFusion: Rethinking Feature Fusion for Low-Bit Quantization of Multi-Task Networks
https://arxiv.org/abs/2511.05898
GABFusion: Rethinking Feature Fusion for Low-Bit Quantization of Multi-Task Networks
https://arxiv.org/abs/2511.05898
chat, what are we thinking: quantization or batch size?
November 11, 2025 at 12:15 PM
chat, what are we thinking: quantization or batch size?
🎓 𝐂𝐞𝐥𝐞𝐛𝐫𝐚𝐭𝐢𝐧𝐠 𝐀𝐜𝐚𝐝𝐞𝐦𝐢𝐜 𝐄𝐱𝐜𝐞𝐥𝐥𝐞𝐧𝐜𝐞 𝐚𝐭 𝐑-𝐏𝐎𝐃𝐈𝐃! 🎉
Proud to share that Joris van de Weg successfully completed his M.Sc. Thesis at Delft University of Technology!
📘 Thesis Title:
Adaptive Compression of Deep Learning Models for Edge Inference via Bayesian Decomposition and Quantization Gates
Proud to share that Joris van de Weg successfully completed his M.Sc. Thesis at Delft University of Technology!
📘 Thesis Title:
Adaptive Compression of Deep Learning Models for Edge Inference via Bayesian Decomposition and Quantization Gates
November 11, 2025 at 11:06 AM
🎓 𝐂𝐞𝐥𝐞𝐛𝐫𝐚𝐭𝐢𝐧𝐠 𝐀𝐜𝐚𝐝𝐞𝐦𝐢𝐜 𝐄𝐱𝐜𝐞𝐥𝐥𝐞𝐧𝐜𝐞 𝐚𝐭 𝐑-𝐏𝐎𝐃𝐈𝐃! 🎉
Proud to share that Joris van de Weg successfully completed his M.Sc. Thesis at Delft University of Technology!
📘 Thesis Title:
Adaptive Compression of Deep Learning Models for Edge Inference via Bayesian Decomposition and Quantization Gates
Proud to share that Joris van de Weg successfully completed his M.Sc. Thesis at Delft University of Technology!
📘 Thesis Title:
Adaptive Compression of Deep Learning Models for Edge Inference via Bayesian Decomposition and Quantization Gates
Amit LeVi, Raz Lapid, Rom Himelstein, Yaniv Nemcovsky, Ravid Shwartz Ziv, Avi Mendelson
You Had One Job: Per-Task Quantization Using LLMs' Hidden Representations
https://arxiv.org/abs/2511.06516
You Had One Job: Per-Task Quantization Using LLMs' Hidden Representations
https://arxiv.org/abs/2511.06516
November 11, 2025 at 10:43 AM
Amit LeVi, Raz Lapid, Rom Himelstein, Yaniv Nemcovsky, Ravid Shwartz Ziv, Avi Mendelson
You Had One Job: Per-Task Quantization Using LLMs' Hidden Representations
https://arxiv.org/abs/2511.06516
You Had One Job: Per-Task Quantization Using LLMs' Hidden Representations
https://arxiv.org/abs/2511.06516
Mrinal Kanti Roychowdhury: Optimal Quantization on Spherical Surfaces: Continuous and Discrete Models - A Beginner-Friendly Expository Study https://arxiv.org/abs/2511.05099 https://arxiv.org/pdf/2511.05099 https://arxiv.org/html/2511.05099
November 11, 2025 at 6:39 AM
Mrinal Kanti Roychowdhury: Optimal Quantization on Spherical Surfaces: Continuous and Discrete Models - A Beginner-Friendly Expository Study https://arxiv.org/abs/2511.05099 https://arxiv.org/pdf/2511.05099 https://arxiv.org/html/2511.05099
Hongjun Wang, Jiyuan Chen, Xuan Song, Yinqiang Zheng: HarmoQ: Harmonized Post-Training Quantization for High-Fidelity Image https://arxiv.org/abs/2511.05868 https://arxiv.org/pdf/2511.05868 https://arxiv.org/html/2511.05868
November 11, 2025 at 6:35 AM
Hongjun Wang, Jiyuan Chen, Xuan Song, Yinqiang Zheng: HarmoQ: Harmonized Post-Training Quantization for High-Fidelity Image https://arxiv.org/abs/2511.05868 https://arxiv.org/pdf/2511.05868 https://arxiv.org/html/2511.05868
Yui Tatsumi, Ziyue Zeng, Hiroshi Watanabe: Training-Free Adaptive Quantization for Variable Rate Image Coding for Machines https://arxiv.org/abs/2511.05836 https://arxiv.org/pdf/2511.05836 https://arxiv.org/html/2511.05836
November 11, 2025 at 6:35 AM
Yui Tatsumi, Ziyue Zeng, Hiroshi Watanabe: Training-Free Adaptive Quantization for Variable Rate Image Coding for Machines https://arxiv.org/abs/2511.05836 https://arxiv.org/pdf/2511.05836 https://arxiv.org/html/2511.05836
Zhao, Li, Liu, Lu, Wang, Yang, Jiang, Guan: QUARK: Quantization-Enabled Circuit Sharing for Transformer Acceleration by Exploiting Common Patterns in Nonlinear Operations https://arxiv.org/abs/2511.06767 https://arxiv.org/pdf/2511.06767 https://arxiv.org/html/2511.06767
November 11, 2025 at 6:35 AM
Zhao, Li, Liu, Lu, Wang, Yang, Jiang, Guan: QUARK: Quantization-Enabled Circuit Sharing for Transformer Acceleration by Exploiting Common Patterns in Nonlinear Operations https://arxiv.org/abs/2511.06767 https://arxiv.org/pdf/2511.06767 https://arxiv.org/html/2511.06767
Medjadji, Alawadi, Awaysheh, Leduc, Kubler, Le Traon: FedSparQ: Adaptive Sparse Quantization with Error Feedback for Robust & Efficient Federated Learning https://arxiv.org/abs/2511.05591 https://arxiv.org/pdf/2511.05591 https://arxiv.org/html/2511.05591
November 11, 2025 at 6:32 AM
Medjadji, Alawadi, Awaysheh, Leduc, Kubler, Le Traon: FedSparQ: Adaptive Sparse Quantization with Error Feedback for Robust & Efficient Federated Learning https://arxiv.org/abs/2511.05591 https://arxiv.org/pdf/2511.05591 https://arxiv.org/html/2511.05591
Zhaoyang Wang, Dong Wang: GABFusion: Rethinking Feature Fusion for Low-Bit Quantization of Multi-Task Networks https://arxiv.org/abs/2511.05898 https://arxiv.org/pdf/2511.05898 https://arxiv.org/html/2511.05898
November 11, 2025 at 6:30 AM
Zhaoyang Wang, Dong Wang: GABFusion: Rethinking Feature Fusion for Low-Bit Quantization of Multi-Task Networks https://arxiv.org/abs/2511.05898 https://arxiv.org/pdf/2511.05898 https://arxiv.org/html/2511.05898
Amit LeVi, Raz Lapid, Rom Himelstein, Yaniv Nemcovsky, Ravid Shwartz Ziv, Avi Mendelson: You Had One Job: Per-Task Quantization Using LLMs' Hidden Representations https://arxiv.org/abs/2511.06516 https://arxiv.org/pdf/2511.06516 https://arxiv.org/html/2511.06516
November 11, 2025 at 6:30 AM
Amit LeVi, Raz Lapid, Rom Himelstein, Yaniv Nemcovsky, Ravid Shwartz Ziv, Avi Mendelson: You Had One Job: Per-Task Quantization Using LLMs' Hidden Representations https://arxiv.org/abs/2511.06516 https://arxiv.org/pdf/2511.06516 https://arxiv.org/html/2511.06516
[2025-11-11] 📚 Updates in #FedML
(1) FedSparQ: Adaptive Sparse Quantization with Error Feedback for Robust & Efficient Federated Learning
(2) FedHUG: Federated Heterogeneous Unsupervised Generalization for Remote Physiological Measurements
🔍 More at researchtrend.ai/communities/FedML
(1) FedSparQ: Adaptive Sparse Quantization with Error Feedback for Robust & Efficient Federated Learning
(2) FedHUG: Federated Heterogeneous Unsupervised Generalization for Remote Physiological Measurements
🔍 More at researchtrend.ai/communities/FedML
November 11, 2025 at 4:02 AM
[2025-11-11] 📚 Updates in #MQ
(1) SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models
(2) FedSparQ: Adaptive Sparse Quantization with Error Feedback for Robust & Efficient Federated Learning
🔍 More at researchtrend.ai/communities/MQ
(1) SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models
(2) FedSparQ: Adaptive Sparse Quantization with Error Feedback for Robust & Efficient Federated Learning
🔍 More at researchtrend.ai/communities/MQ
November 11, 2025 at 4:02 AM
Moonshot AI's Kimi-K2-Thinking's INT4 QAT explanation
"INT4 QAT is weight-only with fake-quantization: we keep the original BF16 weights in memory, during the forward pass we on-the-fly quantize them to INT4 and immediately de-quantize back to BF16 for the actual computation.
"INT4 QAT is weight-only with fake-quantization: we keep the original BF16 weights in memory, during the forward pass we on-the-fly quantize them to INT4 and immediately de-quantize back to BF16 for the actual computation.
November 10, 2025 at 11:38 PM
Moonshot AI's Kimi-K2-Thinking's INT4 QAT explanation
"INT4 QAT is weight-only with fake-quantization: we keep the original BF16 weights in memory, during the forward pass we on-the-fly quantize them to INT4 and immediately de-quantize back to BF16 for the actual computation.
"INT4 QAT is weight-only with fake-quantization: we keep the original BF16 weights in memory, during the forward pass we on-the-fly quantize them to INT4 and immediately de-quantize back to BF16 for the actual computation.