By the numbers: 7B Parameter Model Weight Sizes
f16/bf16: ~15GB
Q4_K_M: (standard default)
Smaller quants: Range from 3GB to 15GB

#batchinference #selfhosting #gpuoptimization #llm #quantization #modelweights

Moonshot AI introduced Kimi K3.1 Pro on October 24 for coding and other complex tasks, and said it had reduced the model’s ability to carry out cyberattacks before a planned release of its weights.

#MoonshotAI #KimiK31Pro #OpenWeightAI #Cybersecurity #AICoding #ModelWeights
[AI]

Sharing access control within a single set of model weights aims to make security compose across principals while utility may not, tested against held-out splits on a nested-factorisation mechanism. The authors…

#cybersecurity #AIafety #LLMsecurity #modelweights
https://arxiv.org/abs/2609.00445

Deep neural networks across diverse tasks converge to shared low-dimensional weight subspaces, with spectral analysis of 1200+ models showing majority variance in just a few principal directions.

#AIresearch #DeepLearning #ModelWeights
https://arxiv.org/abs/2512.05117

The top Democrat on the House China committee sent letters to five AI CEOs demanding records of theft attempts and details of their security around model weights.

#AiSecurity #China #Congress #ExportControls #ModelWeights #RoKhanna

Deepseek just dropped V4 Pro, adds flash‑aware tricks and holds back the full weights. Pricing tweaks, new agent tools, and fresh benchmarks—what this means for open‑source LLMs. Dive in for the full scoop! #DeepseekV4Pro #OpenSourceLLM #ModelWeights

🔗 aidailypost.com/news/deepsee...

Baseten is flipping the AI game—offering a platform where you keep full control of model weights, bypassing the big hyperscalers. Think training, inference, even speculative decoding without handing over your data. Curious? Dive in. #Baseten #ModelWeights #Hyperscalers

🔗

How much is a few extra years of life worth checking for? youtu.be/cx_CKLb8Yl4?... #Cancer #HealthcareData #ModelWeights #ENDPAC #CNN