Cosmos3 (64B) INT4 Quants Bring Local Image and Video Gen to Mac and CUDA
Formal-Swordfish-228 · reddit · 2026-09-09
A developer released INT4 quantized builds of NVIDIA's 64B-parameter Cosmos3 supporting CUDA and Apple Silicon MLX, enabling local text-to-image and image-to-video.
- Code: gtrg55/cosmos3-quant-mlx-cuda on GitHub
- Weights: JuliaML/Cosmos3-Super-Text2Image-4Step-INT4-G64-BF16 on HF, 4-step generation
- Benchmarked at 5 minutes per clip on an M4 Max 128GB
- Includes a quality comparison against Grok
More from Infra
- Ubicloud publishes "Fixing Tool-Call Accuracy in vLLM" — petrusenko_max · 2026-10-26
- Meta approves nonvoting Class C shares as 2026 capex guidance hits $135B–$145B — thione · 2026-10-26
- Unsloth Ships MiniMax-M3.1-Flash Quants: 2-bit Retains 91.5% Accuracy, Runs in 150GB RAM — danielhanchen · 2026-10-26
- Engineer with thousands of fabricated devices pours cold water on photonic crystal modulator yields — jwt0625 · 2026-10-26
- Anthropic S-1 reveals 48 cents of compute cost per dollar of Q2 revenue — nmasc_ · 2026-10-26
- Persistent AI joins UR network to route private AI inference with encrypted, IP-hidden traffic — 0xSammy · 2026-10-26