Views
No views yet
StreamingSVD: Consistent, Dynamic, and Extendable Image-Guided Long Video Generation Roberto Henschel, Levon Khachatryan, Daniil Hayrapetyan, Hayk Poghosyan, Vahram Tadevosyan, Zhangyang Wang, Shant Navasardyan, Humphrey Shi [arXiv 2403.14773] · [Project page] · [Code]
| Size | Dtype | |
|---|---|---|
| Original | ~12.5 GB | F32 |
| This repo | ~3.2 GB | INT8 |
1from safetensors.torch import load_file
2import torch
3
4tensors = load_file("model.int8.safetensors")
5weights = {}
6for k, v in tensors.items():
7 if k.endswith(".scale"):
8 continue
9 scale_key = k + ".scale"
10 if scale_key in tensors:
11 scale = tensors[scale_key]
12 shape = [-1] + [1] * (v.dim() - 1)
13 weights[k] = v.float() * scale.view(shape)
14 else:
15 weights[k] = v1@article{henschel2024streamingt2v,
2 title={StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text},
3 author={Henschel, Roberto and Khachatryan, Levon and Hayrapetyan, Daniil and Poghosyan, Hayk
4 and Tadevosyan, Vahram and Wang, Zhangyang and Navasardyan, Shant and Shi, Humphrey},
5 journal={arXiv preprint arXiv:2403.14773},
6 year={2024}
7}