AI BriefingNVIDIA Releases INT4 Quantized Version of Cosmos3 64B
Key point
The INT4 quantized version of the NVIDIA Cosmos3 64B model has been released for CUDA and MLX, enabling local execution.
Details
NVIDIA's Cosmos3 64B parameter model has been released in an INT4 quantized version, making it runnable on CUDA and Apple Silicon (MLX) environments.
The released resources are distributed via GitHub repositories and Hugging Face, supporting text-to-image (T2I) and image-to-video (I2V) generation capabilities. In particular, INT4-G64-BF16 weights with 4-Step inference optimization are provided.
In terms of performance, it is reported that generating a single clip takes approximately 5 minutes on an M4 Max 128GB Mac. This case demonstrates the possibility of running large-parameter models on local hardware, and developers can also perform comparative tests with other models such as Grok using the released code.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.