license: apache-2.0
Qwen 2.1 AIO — ComfyUI
An all-in-one Qwen 2.1 checkpoint collection for ComfyUI, provided in two variants: INT8 and BF16.
The goal of these releases is to provide convenient, ready-to-use Qwen 2.1 models without requiring users to separately manage the individual model components.
Available Versions
| Version | Qwen 2.1 | Text Encoder | Format |
|---|---|---|---|
| INT8 | INT8 | W4A8 | Quantized |
| BF16 | BF16 | BF16 | Full BF16 |
INT8
The INT8 AIO version contains:
- Qwen 2.1 — INT8
- Text Encoder — W4A8
- Optimized for reduced VRAM usage compared with the full BF16 version
Filename:
Qwen2.1_AIO_INT8.safetensors
BF16
The BF16 AIO version contains:
- Qwen 2.1 — BF16
- Text Encoder — BF16
- No quantization; all components are stored in BF16
Filename:
Qwen2.1_AIO_BF16.safetensors
ComfyUI
These checkpoints are intended for use with ComfyUI.
Place the downloaded checkpoint in:
ComfyUI/models/checkpoints/
Then select the appropriate AIO checkpoint using your normal ComfyUI checkpoint/model loader workflow.
Which Version Should I Use?
INT8 is intended for users who want to reduce VRAM requirements while retaining a high-quality Qwen 2.1 workflow.
BF16 provides the full BF16 version and is intended for systems with sufficient VRAM.
In short:
INT8
Qwen 2.1 → INT8
Text Encoder → W4A8
↓
Lower VRAM requirements
BF16
Qwen 2.1 → BF16
Text Encoder → BF16
↓
Full BF16 precision
Files
Qwen2.1_AIO_INT8.safetensors
Qwen2.1_AIO_BF16.safetensors
Both files are packaged as all-in-one checkpoints for convenient ComfyUI use.
Quantization
The INT8 release uses:
- Qwen 2.1: INT8
- Text Encoder: W4A8
The BF16 release uses:
- Qwen 2.1: BF16
- Text Encoder: BF16
Credits
Qwen 2.1 model components are based on the original Qwen model release.
This repository provides the AIO packaging and quantized variants for convenient use in ComfyUI.
License
This repository is released under the Apache 2.0 license, subject to the licensing terms of the underlying Qwen model and its components.