wan 1.3b gguf
简介
Wan 1.3B GGUF 是由 Calcuis 提供的轻量化视频生成模型量化版本。它将原本庞大的视频生成能力压缩至 1.3B 参数规模,并通过 GGUF 格式优化,使得普通开发者无需顶配 A100 显卡,在消费级 GPU 甚至部分大内存设备上即可尝试文本生成视频。虽然参数量较小,但它在保持基础动态效果的同时,极大降低了推理门槛。对于想要在本地部署视频工作流、或在 ComfyUI 等工具中快速迭代视频原型的用户来说,这是一个极佳的低成本入门选择。
核心亮点
- GGUF 量化大幅降低显存占用,适配消费级显卡
- 轻量化参数实现快速推理,适合本地视频原型开发
- Apache-2.0 协议,对商业应用和二次开发极友好
- 可无缝集成至支持 GGUF 的主流 AI 视频工作流
使用方法
安装依赖
# 安装 Hugging Face transformers
pip install transformers torch
SDK 使用
# 使用 transformers 加载模型
from transformers import AutoModel, AutoTokenizer
model = AutoModel.from_pretrained("calcuis/wan-1.3b-gguf")
tokenizer = AutoTokenizer.from_pretrained("calcuis/wan-1.3b-gguf")
Hugging Face 下载
我们推荐使用命令行或者 Hugging Face Hub SDK 来进行模型的下载。
操作指引:在下载前,请先通过如下命令安装 huggingface_hub:
操作指引
pip install -U huggingface_hub
命令行下载
下载完整模型库
下载完整模型库
huggingface-cli download calcuis/wan-1.3b-gguf
下载单个文件到指定本地文件夹(以下载 config.json 到当前路径下 ./dir 目录为例)
下载单个文件到指定本地文件夹(以下载 config.json 到当前路径下 ./dir 目录为例)
huggingface-cli download calcuis/wan-1.3b-gguf config.json --local-dir ./dir
SDK 下载
SDK 下载
# 模型下载
from huggingface_hub import snapshot_download
model_dir = snapshot_download('calcuis/wan-1.3b-gguf')
Git 下载
请确保 lfs 已经被正确安装
Git 下载
git lfs install
git clone https://huggingface.co/calcuis/wan-1.3b-gguf
如果您希望跳过 lfs 大文件下载,可以使用如下命令
跳过 LFS
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/calcuis/wan-1.3b-gguf
模型文件托管在 Hugging Face Hub,使用 HF CLI / SDK / Git 直接下载,不经过本站。
PyTorch / Transformers 使用
安装 Transformers
安装 Transformers
pip install -U transformers torch
模型加载和推理
模型加载和推理
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained('calcuis/wan-1.3b-gguf')
tokenizer = AutoTokenizer.from_pretrained('calcuis/wan-1.3b-gguf')
模型下载
我们推荐使用命令行或者 ModelScope SDK 来进行模型的下载。
操作指引:在下载前,请先通过如下命令安装 ModelScope:
操作指引
pip install modelscope
命令行下载
下载完整模型库
下载完整模型库
modelscope download --model calcuis/wan-1.3b-gguf
下载单个文件到指定本地文件夹(以下载 README.md 到当前路径下 dir 目录为例)
下载单个文件到指定本地文件夹(以下载 README.md 到当前路径下 dir 目录为例)
modelscope download --model calcuis/wan-1.3b-gguf README.md --local_dir ./dir
SDK 下载
SDK 下载
# 模型下载
from modelscope import snapshot_download
model_dir = snapshot_download('calcuis/wan-1.3b-gguf')
Git 下载
请确保 lfs 已经被正确安装
Git 下载
git lfs install
git clone https://www.modelscope.cn/calcuis/wan-1.3b-gguf.git
如果您希望跳过 lfs 大文件下载,可以使用如下命令
跳过 LFS
GIT_LFS_SKIP_SMUDGE=1 git clone https://www.modelscope.cn/calcuis/wan-1.3b-gguf.git
ModelScope 模型页直接下载模型文件;无需将模型文件放在本站服务器。
Notebook 快速开发
下载并安装 ModelScope library
下载并安装 ModelScope library
pip install "modelscope[audio,cv,nlp,multi-modal,science]" -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html
模型加载和推理
模型加载和推理
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks
p = pipeline('text-generation', 'calcuis/wan-1.3b-gguf')
完整文档
来源: HuggingFace
---
license: apache-2.0
language:
- en
base_model:
- Wan-AI/Wan2.1-VACE-1.3B
- Wan-AI/Wan2.1-T2V-1.3B
pipeline_tag: text-to-video
tags:
- gguf-node
- gguf-connector
widget:
- text: >-
a pig moving quickly in a beautiful winter scenery nature trees sunset
tracking camera
parameters:
negative_prompt: blurry ugly bad
output:
url: samples\ComfyUI_00003_.webp
- text: >-
a pig moving quickly in a beautiful winter scenery nature trees sunset
tracking camera
parameters:
negative_prompt: blurry ugly bad
output:
url: samples\ComfyUI_00002_.webp
- text: >-
a pig moving quickly in a beautiful winter scenery nature trees sunset
tracking camera
parameters:
negative_prompt: blurry ugly bad
output:
url: samples\ComfyUI_00001_.webp
- text: >-
a pig moving quickly in a beautiful winter scenery nature trees sunset
tracking camera
parameters:
negative_prompt: blurry ugly bad
output:
url: samples\ComfyUI_00004_.webp
- text: >-
a pig moving quickly in a beautiful winter scenery nature trees sunset
tracking camera
parameters:
negative_prompt: blurry ugly bad
output:
url: samples\ComfyUI_00005_.webp
- text: >-
a pig moving quickly in a beautiful winter scenery nature trees sunset
tracking camera
parameters:
negative_prompt: blurry ugly bad
output:
url: samples\ComfyUI_00006_.webp
---
gguf quantized version of wan 1.3b models
- run it straight with
gguf-connector
- opt a
gguffile in the current directory to interact with by:
code
ggc w2>
>GGUF file(s) available. Select which one to use:
>
>1. wan2.1-t2v-1.3b-q4_0.gguf
>2. wan2.1-t2v-1.3b-q8_0.gguf
>3. wan2.1-vace-1.3b-q4_0.gguf
>4. wan2.1-vace-1.3b-q8_0.gguf
>
>Enter your choice (1 to 4): _
>>>
run it with gguf-node via comfyui
- drag wan to >
./ComfyUI/models/diffusion_models
- drag umt5 to >
./ComfyUI/models/text_encoders
- drag pig to >
./ComfyUI/models/vae
<Gallery />
review
wanarchitecture; should work on both comfyui-gguf and gguf nodes
- full set gguf works right away (model + encoder + vae); gguf node is recommended for full gguf
- vace model is recommended, since it doesn't need vision clip to work (for i2v and v2v) and run faster than fun model a lot, according to the initial test results
- upgrade your node for umt5 gguf encoder support
- note: for umt5 gguf, you might encounter oom after rebuilding your tokenizer in the first prompt (once built the tokenizer alive during the session; before you kill it); don't panic, prompt again it should work
!screenshot
reference
- base model from wan-ai
- comfyui from comfyanonymous
- comfyui-gguf city96 (special thanks; fully compatible)
- pig architecture from connector
- gguf-connector (pypi)