community Reddit LocalLLaMA 2026-09-09T14:21 正文 thin 0 words

SOTA ImageGen Locally NVIDIA Cosmos3(64B) INT4 quants CUDA/MLX

英文标题:SOTA ImageGen Locally NVIDIA Cosmos3(64B) INT4 quants CUDA/MLX

Cosmos3 INT4 T2I + I2V on Apple Silicon — code, weights and a Grok comparison GitHub - https://github.com/gtrg55/cosmos3-quant-mlx-cuda HF weights - https://huggingface.co/JuliaML/Cosmos3-Super-Text2Image-4Step-INT4-G64-BF16 Single clip took approximately 5m on M4 MAX 128 GB Mac Cosmos3 - a 64
打开原文相关 2 · 机会 2

中文

Cosmos3 INT4 T2I + I2V on Apple Silicon — code, weights and a Grok comparison GitHub - https://github.com/gtrg55/cosmos3-quant-mlx-cuda HF weights - https://huggingface.co/JuliaML/Cosmos3-Super-Text2Image-4Step-INT4-G64-BF16 Single clip took approximately 5m on M4 MAX 128 GB Mac Cosmos3 - a 64

中文全文翻译正在补齐,可先查看下方英文原文。

English / 原文

Cosmos3 INT4 T2I + I2V on Apple Silicon — code, weights and a Grok comparison GitHub - https://github.com/gtrg55/cosmos3-quant-mlx-cuda HF weights - https://huggingface.co/JuliaML/Cosmos3-Super-Text2Image-4Step-INT4-G64-BF16 Single clip took approximately 5m on M4 MAX 128 GB Mac Cosmos3 - a 64