#HuggingFace

Tested ZONOS2 on an 8GB RTX 4060 Laptop (WSL2): the 15.3GB bf16 weights run via Windows system-memory fallback, a KV-cache override, and the CUDA toolkit at ~1/20 realtime. Plus a Japanese name-accent gotcha with A/B audio.

Laxhar's U1 trainer needs 32GB+ GPU, bf16 only — 4bit broke gen tower. Prefix offload keeps ~20GB peak. 8-step LoRA stack, A3B MoE compat, official training code gap.

40GB+ VRAM for a 3B model. VBench 85.11 beats dedicated 14B video generators. RunPod GPU costs from $2.2/session. The 'unified' model still ships as two checkpoint files.

Sentence Transformers v5.4 adds multimodal support. Eight embedding models and four rerankers including Qwen3-VL and NVIDIA Nemotron can now be used through a unified API.