Text-to-Image
Diffusers
Safetensors
MiniMax H3
English
orbitquant
w4a8
4-bit precision
gptq
quantized
image-editing
Instructions to use WaveCut/MiniMax-H3-Image-OrbitQuant-W4A8 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use WaveCut/MiniMax-H3-Image-OrbitQuant-W4A8 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("WaveCut/MiniMax-H3-Image-OrbitQuant-W4A8", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
- DiffusionBee
Remove the pruned release: drop its link and restore the original chart
Browse files- README.md +1 -4
- SHA256SUMS +2 -2
- assets/latency_and_memory.png +2 -2
README.md
CHANGED
|
@@ -25,13 +25,10 @@ tags:
|
|
| 25 |
The diffusion transformer and the MiniMax-H3 Qwen3-VL text encoder both run 8-bit activations over 4-bit weights. The 4-bit codes were chosen with GPTQ rounding against calibration prompts that are not in the evaluation set.
|
| 26 |
The runtime is the official MiniMax-H3-Image modular pipeline of [Diffusers](https://github.com/huggingface/diffusers/pull/15230) with fused OrbitQuant kernels, and the [ComfyUI-OrbitQuant](https://github.com/iamwavecut/ComfyUI-OrbitQuant) nodes.
|
| 27 |
The download is 27.5 GB. The original needs 135.4 GB: the 66.3 GB transformer and the 66.7 GB text encoder of MiniMax-H3, which it downloads separately.
|
| 28 |
-
The [pruned release](https://huggingface.co/WaveCut/MiniMax-H3-Image-Pruned-OrbitQuant-W4A8) applies the same recipe to [MiniMax-H3-Image-Pruned](https://huggingface.co/WaveCut/MiniMax-H3-Image-Pruned), the original without the weights its inference never reads. On the same machine both releases have the same speed, memory and download size (the chart).
|
| 29 |
|
| 30 |
**On one RTX PRO 6000 a 4096×4096 image takes 13.5 s with a 29.8 GiB device peak, and a 2048×2048 image takes 2.4 s. The BF16 original needs 22.4 s and 4.4 s for the same work even when each of its models is resident alone. With model offload the release also runs in 23.4 GiB of free device memory (a simulated 24 GB card).**
|
| 31 |
|
| 32 |
-
 describes the run). The original does not fit that card with all models on the GPU. The numbers in the text and the tables of this card come from its first measurement.
|
| 35 |
|
| 36 |
Left: the BF16 original. Right: this release. Same prompt, seed and size; the panels are downscaled to 1024 pixels wide. LPIPS against the BF16 image is in parentheses (lower is closer, see [Quality](#quality)).
|
| 37 |
|
|
|
|
| 25 |
The diffusion transformer and the MiniMax-H3 Qwen3-VL text encoder both run 8-bit activations over 4-bit weights. The 4-bit codes were chosen with GPTQ rounding against calibration prompts that are not in the evaluation set.
|
| 26 |
The runtime is the official MiniMax-H3-Image modular pipeline of [Diffusers](https://github.com/huggingface/diffusers/pull/15230) with fused OrbitQuant kernels, and the [ComfyUI-OrbitQuant](https://github.com/iamwavecut/ComfyUI-OrbitQuant) nodes.
|
| 27 |
The download is 27.5 GB. The original needs 135.4 GB: the 66.3 GB transformer and the 66.7 GB text encoder of MiniMax-H3, which it downloads separately.
|
|
|
|
| 28 |
|
| 29 |
**On one RTX PRO 6000 a 4096×4096 image takes 13.5 s with a 29.8 GiB device peak, and a 2048×2048 image takes 2.4 s. The BF16 original needs 22.4 s and 4.4 s for the same work even when each of its models is resident alone. With model offload the release also runs in 23.4 GiB of free device memory (a simulated 24 GB card).**
|
| 30 |
|
| 31 |
+

|
|
|
|
|
|
|
| 32 |
|
| 33 |
Left: the BF16 original. Right: this release. Same prompt, seed and size; the panels are downscaled to 1024 pixels wide. LPIPS against the BF16 image is in parentheses (lower is closer, see [Quality](#quality)).
|
| 34 |
|
SHA256SUMS
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
edfab8ece9d0ccd39106c4fe44105ce599b27ca7dd22601b736a8e9219cb4d26 LICENSE
|
| 2 |
59b99642b95ea21630e311198ddbfffbfe05aadba0c2f5d884cbdf4efcc90f44 LICENSE-MiniMax-H3
|
| 3 |
1e465a4dbc343d524371e1210a3631f1d2a34aeaa5e85022b88f1e0b232be7ed NOTICE
|
| 4 |
-
|
| 5 |
c0951934d193eeb212cc962133b828fd7d435b15dc53199c77c6231ed3795610 assets/e01.jpg
|
| 6 |
5968066c6f85cd3490f56ff9f58b2eb9a057af13b04ac1e7a1e705f579f87fa6 assets/e03.jpg
|
| 7 |
0a09c1573b126004faad2608dab7f60ffdfe1f344f1f83194a86aaafbf688edd assets/e06.jpg
|
|
@@ -10,7 +10,7 @@ c0951934d193eeb212cc962133b828fd7d435b15dc53199c77c6231ed3795610 assets/e01.jpg
|
|
| 10 |
59c78dd40cd42d2551a4062306af53fd938e69c860472937fb0d0802ec885ba0 assets/e12.jpg
|
| 11 |
2be488367911cd77ebe80943214ce61066270abb732e0905f78a7c831e8d63fc assets/e13.jpg
|
| 12 |
6feefe80606b5755ffabb3f2df756c6800a2afb0b6bf3c38df4fc0f421bf36bf assets/e16.jpg
|
| 13 |
-
|
| 14 |
01df62652f6d23d6fc2ad864e380949af70e521f8950e579e4c5c0c114720c1c assets/s01-crop.jpg
|
| 15 |
168a56c303e217cdd5b48b9cb4759175d09db3d7ab98bad7a3d55ca8c3c0860b assets/s01.jpg
|
| 16 |
59abe1cbcec4b0c2b138bf76e8027245acc84e54be5786c4a64e2e77e6a49b55 assets/s02.jpg
|
|
|
|
| 1 |
edfab8ece9d0ccd39106c4fe44105ce599b27ca7dd22601b736a8e9219cb4d26 LICENSE
|
| 2 |
59b99642b95ea21630e311198ddbfffbfe05aadba0c2f5d884cbdf4efcc90f44 LICENSE-MiniMax-H3
|
| 3 |
1e465a4dbc343d524371e1210a3631f1d2a34aeaa5e85022b88f1e0b232be7ed NOTICE
|
| 4 |
+
085f9f5f7ea7a68607790bd4771f6e9c723ce59e52acb615fadec8bbbe451a05 README.md
|
| 5 |
c0951934d193eeb212cc962133b828fd7d435b15dc53199c77c6231ed3795610 assets/e01.jpg
|
| 6 |
5968066c6f85cd3490f56ff9f58b2eb9a057af13b04ac1e7a1e705f579f87fa6 assets/e03.jpg
|
| 7 |
0a09c1573b126004faad2608dab7f60ffdfe1f344f1f83194a86aaafbf688edd assets/e06.jpg
|
|
|
|
| 10 |
59c78dd40cd42d2551a4062306af53fd938e69c860472937fb0d0802ec885ba0 assets/e12.jpg
|
| 11 |
2be488367911cd77ebe80943214ce61066270abb732e0905f78a7c831e8d63fc assets/e13.jpg
|
| 12 |
6feefe80606b5755ffabb3f2df756c6800a2afb0b6bf3c38df4fc0f421bf36bf assets/e16.jpg
|
| 13 |
+
c9ff1719a7dc108f8a1297e04076c8e791bf975762502db345e65df01a78027a assets/latency_and_memory.png
|
| 14 |
01df62652f6d23d6fc2ad864e380949af70e521f8950e579e4c5c0c114720c1c assets/s01-crop.jpg
|
| 15 |
168a56c303e217cdd5b48b9cb4759175d09db3d7ab98bad7a3d55ca8c3c0860b assets/s01.jpg
|
| 16 |
59abe1cbcec4b0c2b138bf76e8027245acc84e54be5786c4a64e2e77e6a49b55 assets/s02.jpg
|
assets/latency_and_memory.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|