Total runs:
60.8K
Run Growth:
20.4K
Growth Rate:
32.25%
Sponsored by
PostMcpAi
gpustack / FLUX.1-mini-GGUF
Model's Last Updated: December 02 2024
text-to-image
Introduction of FLUX.1-mini-GGUF
Model Details of FLUX.1-mini-GGUF
FLUX.1-mini-GGUF
!!! Experimental supported by gpustack/llama-box v0.0.86+ only !!!
Model creator
:
ARC Lab, Tencent PCG
Original model
:
flux-mini
GGUF quantization
: based on stable-diffusion.cpp
ac54e
that patched by llama-box.
| Quantization | OpenAI CLIP ViT-L/14 Quantization | Google T5-xxl Quantization | VAE Quantization |
|---|---|---|---|
| FP16 | FP16 | FP16 | FP16 |
| Q8_0 | FP16 | Q8_0 | FP16 |
| (pure) Q8_0 | Q8_0 | Q8_0 | FP16 |
| Q4_1 | FP16 | Q8_0 | FP16 |
| Q4_0 | FP16 | Q8_0 | FP16 |
| (pure) Q4_0 | Q4_0 | Q4_0 | FP16 |
🔥 Flux-Mini
A 3.2B MMDiT distilled from Flux-dev for efficient text-to-image generation
Nowadays, text-to-image (T2I) models are growing stronger but larger, which limits their pratical applicability, especially on consumer-level devices. To bridge this gap, we distilled the
12B
Flux-dev
model into a
3.2B
Flux-mini
model, and trying to preserve its strong image generation capabilities. Specifically, we prune the original
Flux-dev
by reducing its depth from
19 + 38
(number of double blocks and single blocks) to
5 + 10
. The pruned model is further tuned with denosing and feature alignment objectives on a curated image-text dataset.
🔥🔥 Nonetheless, with limited computing and data resources, the capability of our Flux-mini is still limited in certain domains. To facilitate the development of flux-based models, we open-sourced the codes to distill Flux in this folder . We appeal people interested in this project to collaborate together to build a more applicable and powerful text-to-image model!
⏰ Timeline
[2024.11.26] We are delighted to release the first version of Flux-Mini!
⚡️ Efficiency Comparison
We compared our Flux-Mini with Flux-Dev on
a single H20 GPU
with
BF16
precision, with
batch-size=1
,
deepspeed stage=2
,
gradient_checkpoint=True
. For inference, we adopt
num_steps=50
. The costs of T5, CLIP and VAE are included.
OOM
means out-of-memory.
| Resolution | Training Strategy | Model | Training Speed (s/img) | Training Memory (GB) | Inference Speed (s/img) | Inference Memory (GB) |
|---|---|---|---|---|---|---|
| 512 | LoRA(r=16) | Flux-dev | 1.10 | 35.91 | 11.11 | 35.11 |
| 512 | LoRA(r=16) | Flux-Mini | 0.33 | 19.06 | 3.07 | 18.49 |
| 512 | Fully Finetune | Flux-dev | OOM | OOM | 11.11 | 35.11 |
| 512 | Fully Finetune | Flux-Mini | 0.57 | 83.7 | 3.07 | 18.49 |
| 1024 | LoRA(r=16) | Flux-dev | 2.93 | 38.03 | 38.26 | 42.24 |
| 1024 | LoRA(r=16) | Flux-Mini | 1.05 | 22.21 | 10.31 | 25.61 |
| 1024 | Fully Finetune | Flux-dev | OOM | OOM | 38.26 | 42.24 |
| 1024 | Fully Finetune | Flux-Mini | 1.30 | 83.71 | 10.31 | 25.61 |
⛅ Limitations
Compared with advanced text-to-image models, our model was trained with limited computing resources and synthetic data with mediocre quality. Thus, the generation capability of our model is still limited in certain categories.
The current model is ok with generating common images such as human/animal faces, landscapes, fantasy and abstract scenes.
Unfortunately, it is still incompetent in many scenarios. Including but not limited to:
- Fine-grained details, such as human and animal structures
- Typography
- Perspective and Geometric Structure
- Dynamics and Motion
- Commonsense knowledge, e.g., brand logo
- Physical Plausibility
- Cultural Diversity
Since our model is trained with prompts in JourneyDB, we encourage users to apply our model with similar prompt formats (compositions of nouns and adjectives) to achieve the best quality. For example: "profile of sad Socrates, full body, high detail, dramatic scene, Epic dynamic action, wide angle, cinematic, hyper-realistic, concept art, warm muted tones as painted by Bernie Wrightson, Frank Frazetta."
We welcome everyone in the community of collaborate and PR for this model.
💻 Flux-NPU
The widespread development of NPUs has provided extra device options for model training and inference. To facilitate the usage of flux, We provide a codebase that could run the training and inference code of FLUX on NPUs.
Please find out more details in this folder .
⚡️ Efficiency Comparison on NPU.
We compared our Flux-Mini with Flux-Dev on a single
Ascend 910B NPU
with
BF16
precision, with
batch-size=1
,
deepspeed stage=2
,
gradient_checkpoint=True
. For inference, we adopt
num_steps=50
. The costs of T5, CLIP and VAE are included.
OOM
means out-of-memory.
| Resolution | Training Strategy | Model | Training Speed (s/img) | Training Memory (GB) | Inference Speed (s/img) | Inference Memory (GB) |
|---|---|---|---|---|---|---|
| 512 | LoRA(r=16) | Flux-dev | 1.07 | 38.45 | 11.00 | 58.62 |
| 512 | LoRA(r=16) | Flux-Mini | 0.37 | 20.64 | 3.26 | 19.48 |
| 512 | Fully Finetune | Flux-dev | OOM | OOM | 11.00 | 58.62 |
| 512 | Fully Finetune | Flux-Mini | OOM | OOM | 3.26 | 19.48 |
| 1024 | LoRA(r=16) | Flux-dev | 3.01 | 44.69 | OOM | OOM |
| 1024 | LoRA(r=16) | Flux-Mini | 1.06 | 25.84 | 10.60 | 27.76 |
| 1024 | Fully Finetune | Flux-dev | OOM | OOM | OOM | OOM |
| 1024 | Fully Finetune | Flux-Mini | OOM | OOM | 10.60 | 27.76 |
🐾 Disclaimer
Users are granted the freedom to create images using our model and tools, but they are expected to comply with local laws and utilize it responsibly. The developers do not assume any responsibility for potential misuse by users.
👍 Acknowledgements
We thank the authors of the following repos for their excellent contribution!
🔎 License
Our Flux-mini model weights follows the liscence of Flux-Dev non-commercial License .
The other codes follow the Apache-2.0 License.
Runs of gpustack FLUX.1-mini-GGUF on huggingface.co
149
Total runs
0
24-hour runs
-4
3-day runs
1
7-day runs
47
30-day runs
More Information About FLUX.1-mini-GGUF huggingface.co Model
Url of FLUX.1-mini-GGUF
FLUX.1-mini-GGUF huggingface.co Url
Provider of FLUX.1-mini-GGUF huggingface.co
Other API from gpustack
Updated:October 31 2024
Total runs:
20.8K
Run Growth:
-5.4K
Growth Rate:
-25.82%
Total runs:
6.0K
Run Growth:
2.1K
Growth Rate:
35.79%
Total runs:
4.5K
Run Growth:
252
Growth Rate:
5.58%
Total runs:
2.8K
Run Growth:
2.0K
Growth Rate:
70.71%
Total runs:
2.7K
Run Growth:
647
Growth Rate:
23.94%
Total runs:
2.6K
Run Growth:
651
Growth Rate:
25.21%
Total runs:
2.0K
Run Growth:
-387
Growth Rate:
-19.08%
Total runs:
1.9K
Run Growth:
804
Growth Rate:
41.85%
Total runs:
1.8K
Run Growth:
1.3K
Growth Rate:
73.09%
Total runs:
1.6K
Run Growth:
809
Growth Rate:
51.23%
Total runs:
1.5K
Run Growth:
199
Growth Rate:
13.35%
Total runs:
806
Run Growth:
-54
Growth Rate:
-6.70%
Total runs:
645
Run Growth:
219
Growth Rate:
33.95%
Total runs:
468
Run Growth:
-45
Growth Rate:
-9.20%
Total runs:
331
Run Growth:
-19
Growth Rate:
-5.74%
Total runs:
328
Run Growth:
97
Growth Rate:
29.57%
Total runs:
326
Run Growth:
13
Growth Rate:
3.99%
Total runs:
323
Run Growth:
166
Growth Rate:
51.39%
Total runs:
311
Run Growth:
74
Growth Rate:
23.79%
Total runs:
296
Run Growth:
156
Growth Rate:
52.70%
Total runs:
272
Run Growth:
112
Growth Rate:
41.18%
Total runs:
271
Run Growth:
1
Growth Rate:
0.37%
Total runs:
269
Run Growth:
12
Growth Rate:
4.46%
Total runs:
237
Run Growth:
83
Growth Rate:
35.02%
Total runs:
205
Run Growth:
116
Growth Rate:
56.59%
Total runs:
149
Run Growth:
67
Growth Rate:
44.97%
Total runs:
139
Run Growth:
51
Growth Rate:
36.69%
Total runs:
106
Run Growth:
15
Growth Rate:
14.15%
Total runs:
80
Run Growth:
-19
Growth Rate:
-23.75%
Total runs:
0
Run Growth:
0
Growth Rate:
0.00%