Wan2.2 TI2V 5B NVFP4 for NML Imagine
This is NML's deterministic NVFP4 conversion of Wan-AI/Wan2.2-TI2V-5B@921dbaf3f1674a56f47e83fb80a34bac8a8f203e. Eligible linear and embedding contractions in the Wan video denoiser and UMT5-XXL prompt encoder use NML's last-axis weight-v2 NVFP4 representation. Biases, normalization, adaptive modulation, relative-position embeddings, and patch convolution use the official BF16 inference precision; the convolutional Wan VAE remains F32.
The repository includes exact source, tensor-disposition, conversion, and output-file manifests. It supports both text-to-video and image-to-video through NML's imagine CLI. Source PyTorch pickle checkpoints are converted into safetensors and are not redistributed. Recipe: nml-wan2.2-ti2v-5b-nvfp4-v1.
- Downloads last month
- 24
Model tree for narendra747/wan2.2-ti2v-5b-nvfp4
Base model
Wan-AI/Wan2.2-TI2V-5B
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js