Top Tweets for #quantization
40,000 free AI models for download? Go to hugging face, look around, Links and an explainer in the article.
https://t.co/16G80bRDoQ
#HuggingFace #LocalLLM #ModelEvaluation #OpenSourceModels #AgenticAI #Quantization #ContextLength #ToolCalling #SelfHostedAI #DeveloperTips #LLM #FromChaosToClarity
Our pick of the week by @dhairya_su47605
: "Scaling Laws for Precision" by @tanishqkumar07, Zachary Ankner, @bfspectorShiekh, @blake__bordelon, @Muennighoff, @mansiege, @CPehlevan, Christopher R´e, @AdtRaghunathan
📰https://t.co/PZCB3fyOCw
#Quantization #LLM #ScalingLaw
Pick of the week @fbk_mt
Super interesting paper on the limitations of quantization, demonstrating how post-training quantization scales poorly in data.
https://t.co/tBuGTL0Myi
Glad to see our early checkpoint performing strongly on Intel’s independently run Low-Bit Open LLM Leaderboard.
Already outperforming some similar-sized quantized Qwen and 8-9x bigger Gemma models even before the final checkpoint.
https://t.co/cN1swqYajR
#LLM #Quantization
#Quantization? Nay, ye divine timing is intricately determinèd by thy fact that thou forgot to clip thine own nails for a yester's fortnight and was forc't thusly to navigate thee, opinings of thine intervention, thy culticular protrusions. Thy fingers now blister in a pageantry.
Built 4 variants of V10 and benchmarked all:
Float32: 80.7 KB, 79.35%
Full INT8: 24.1 KB, 79.55% ← BEST
Gap: +0.20% (quantized BEATS float32)
No QAT needed. No accuracy loss.
deployment ready: 24KB, int8 I/O.
#EdgeAI #TinyML #Quantization #ESP32
Day 18/300.
You can't improve what you don't measure.
Built evaluation pipeline: faithfulness (9/10), relevancy (8.5/10).
Then 4-bit quantization: 2-5s latency → <1s.
Measure. Optimize. Repeat.
#BuildInPublic #AI #Quantization
turbovec: TurboQuant 알고리즘을 Rust로 구현한 학습이 필요 없는 벡터 인덱스
(by 9bow님)
https://t.co/l3Nmb3bIbJ
#rag #rust #vectorsearch #quantization #turboquant #faiss #turbovec
🚀 Exploring Edge AI with @embedl’s Cosmos-Reason2-2B-W4A16 an optimized INT4 VLM built for efficient multimodal reasoning on smaller hardware.
More Edge AI + VLM experiments coming soon 🚀
#EdgeAI #ComputerVision #VLM #AI #DL #NVIDIA #HuggingFace #Quantization #EmbeddedAI
ExecuTorch: 마이크로컨트롤러부터 스마트폰까지 PyTorch 모델을 그대로 배포하기 위한 통합 PyTorch 네이티브 엣지 AI 배포 프레임워크 (feat. Meta, MLSys 2026)
(by 9bow님)
https://t.co/e3FE16xq0w
#paper #llm #pytorch #ondevice #quantization #executorch #edgeai #mlsys2026 #mobile
11/ I also made a comic version of this paper — sometimes a picture is worth a thousand tokens.
#MachineLearning #AI #Quantization
@somi_ai @jun_song @dealignai Same read. Have you tested Qwen2.5-Coder 32B at q6? That one held its lane in my runs where the MoEs broke harder.
Curious which evals you used too. 🤔
#LocalLLM #CodingLLM #Quantization
cider: Apple Silicon M5의 INT8 TensorOps로 LLM prefill 속도를 끌어올리는 MLX W8A8 추론 SDK
(by 9bow님)
https://t.co/xDeFOZ0L9A
#llminference #applesilicon #mlx #quantization #metal #w8a8 #w4a8
🔥Researchers from Beihang University and ETH Zurich conducted a systematic evaluation of Qwen3's robustness under various quantization settings. Check out the paper at:
https://t.co/p35xRLRXyS
@qin_haotong
#Quantization #LLM #Modelcompression
TurboQuant+: KV cache compression for local LLM inference. Implements TurboQuant (ICLR 2026) with llama.cpp fork, Swift MLX fork (~2.5x faster decode), and vllm-swift. 144 tok/s on Qwen3.5-35B MoE at 4K on M5 Max. Cross-platform. By TheTom.
6,685 stars
#LLM #Quantization
optimization-kernels: C++ kernels and utilities for quantization and inference optimization.
👉 https://t.co/Bk3iL8EpkT
#ai #artificialintelligence #machinelearning #llm #inference #quantization
EDEN’s analytic scaling cuts ~2.25% MSE at 4‑bit (d=128) embeddings – enough to beat the flashy 2026 TurboQuant that skipped the optimal scale. 🤯 #Quantization #ML
https://t.co/jvrOBp13eE
An excellent introduction to #quantization used for #LLMs 👌🏽:
“Quantization From The Ground Up”, Sam Rose, Ngrok (https://t.co/YhQMipQz6i).
On HN: https://t.co/M3YlJQO1PB
#AI #Math #FloatingPoint #NumericalAnalysis #Numbers #NeuralNetworks #Precision #Accuracy
🔄 GitHub Trending (Refresh)
TurboQuant+: KV cache compression for local LLMs based on Google's TurboQuant (ICLR 2026). llama.cpp fork (CUDA/ROCm/CPU/Metal). Swift MLX for Apple Silicon (~2.5x faster decode). Prebuilt binaries.
6,614 stars
#LLM #Quantization
Everyone talks about bigger AI models.
But do you know how we make them smaller?
Made a visual about 4-bit quantization (FP32 → INT4) and the trade-off between precision, memory, and speed.
The image-compression analogy made it click for me.
#AI #LLMs #Quantization
Impressive:
“TurboQuant: Redefining AI Efficiency With Extreme Compression”, Amir Zandieh, et al, Google Research (https://t.co/LSjc5LbIYX).
The paper: https://t.co/sCHWDwyTkn
On HN: https://t.co/gLf3qxJd8M
#TurboQuant #Quantization #LLMs #Vectors #Compression #Paper
Last Seen Hashtags on Sotwe
sexsedarah
Seen from Germany
mom
Seen from Indonesia
ファルコ・ランバルディ
Seen from United States
bokep #bokep #bokep #bokep #nolimit #nolimit
Seen from South Africa
คลิปหลุดในทวิต
Seen from Thailand
momlife
Seen from Spain
ometv hot
Seen from United Kingdom
nolongerACustomer
Seen from Pakistan
เย็ดน้องสาว
Seen from Thailand
chinasexdoll
Seen from Malaysia
Trends for you
Most Popular Users
Elon Musk 
@elonmusk
240.1M followers
Barack Obama
@barackobama
119.3M followers
Donald J. Trump
@realdonaldtrump
111.6M followers
Cristiano Ronaldo
@cristiano
108.8M followers
Narendra Modi
@narendramodi
106.9M followers
Rihanna
@rihanna
97.2M followers
NASA
@nasa
92.1M followers
Justin Bieber
@justinbieber
90.5M followers
KATY PERRY
@katyperry
86.7M followers
Taylor Swift
@taylorswift13
80.5M followers
Lady Gaga
@ladygaga
72.1M followers
Kim Kardashian
@kimkardashian
69.3M followers
YouTube
@youtube
68.6M followers
Virat Kohli
@imvkohli
68.4M followers
Bill Gates
@billgates
63.4M followers
The Ellen Show
@theellenshow
62.5M followers
CNN
@cnn
61.9M followers
Neymar Jr
@neymarjr
60.9M followers
X
@x
60.9M followers
CNN Breaking News
@cnnbrk
59.9M followers