|
--- |
|
base_model: black-forest-labs/FLUX.1-dev |
|
--- |
|
|
|
*Note that all these models are derivatives of black-forest-labs/FLUX.1-dev and therefore covered by the |
|
[FLUX.1 [dev] Non-Commercial License](https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md) license.* |
|
|
|
*Some models are derivatives of finetunes, and are included with the permission of the finetuner* |
|
|
|
# Optimised Flux GGUF models |
|
|
|
A collection of GGUF models using mixed quantization (different layers quantized to different precision to optimise fidelity v. memory). |
|
|
|
They were created using the [convert.py script](https://github.com/chrisgoringe/mixed-gguf-converter). |
|
|
|
They can be loaded in ComfyUI using the [ComfyUI GGUF Nodes](https://github.com/city96/ComfyUI-GGUF). Just put the gguf files in your |
|
models/unet directory. |
|
|
|
## Naming convention (mx for 'mixed') |
|
|
|
[original_model_name]_mxN_N.gguf |
|
|
|
where N_N is the average number of bits per parameter. |
|
|
|
## Good choices to start with |
|
``` |
|
- 9_2 is a good choice for 16 GB cards |
|
- 6_9 just fits on a 12 GB card |
|
- 5_9 is comfortable on 12 GB cards |
|
``` |
|
|
|
## Speed? |
|
|
|
On an A40 (plenty of VRAM), everything except the model identical, the time taken to generate an image (30 steps, deis sampler) was: |
|
|
|
- 5_1 => 40.1s |
|
- 5_9 => 55.4s |
|
- 6_9 => 52.1s |
|
- 7_4 => 49.7s |
|
- 7_6 => 43.6s |
|
- 8_4 => 46.8s |
|
- 9_2 => 42.8s |
|
- 9_6 => 48.2s |
|
|
|
for comparison: |
|
- bfloat16 (default) => |
|
- fp8_e4m3fn => |
|
- fp8_e5m2 => |
|
|
|
|
|
|
|
## How is this optimised? |
|
|
|
The process for optimisation is as follows: |
|
|
|
- 240 prompts used for flux images popular at civit.ai were run through the full Flux.1-dev model with randomised resolution and step count. |
|
- For a randomly selected step in the inference, the hidden states before and after the layer stack were captured. |
|
- For each layer in turn, and for each quantization: |
|
- A single layer was quantized |
|
- The initial hidden states were processed by the modified layer stack |
|
- The error (MSE) in the final hidden state was calculated |
|
- This gives a 'cost' for each possible layer quantization - how much different it is to the full model |
|
- An optimised quantization is one that gives the desired reduction in size for the smallest total cost |
|
- A series of recipies for optimization have been created from the calculated costs |
|
- the various 'in' blocks, the final layer blocks, and all normalization scale parameters are stored in float32 |
|
|
|
## Also note |
|
|
|
- Tests on using bitsandbytes quantizations showed they did not perform as well as the equivalent sized GGUF quants |
|
- Different quantizations of different parts of a layer gave significantly worse results |
|
- Leaving bias in 16 bit made no relevant difference |
|
- Costs were evaluated for the original Flux.1-dev model. They are assumed to be essentially the same for finetunes |
|
|
