bitsandbytes

bitsandbytes is the easiest option for quantizing a model to 8 and 4-bit. 8-bit quantization multiplies outliers in fp16 with non-outliers in int8, converts the non-outlier values back to fp16, and then adds them together to return the weights in fp16. This reduces the degradative effect outlier values have on a model’s performance.

4-bit quantization compresses a model even further, and it is commonly used with QLoRA to finetune quantized LLMs.

To use bitsandbytes, make sure you have the following libraries installed:

pip install diffusers transformers accelerate bitsandbytes -U

Now you can quantize a model by passing a BitsAndBytesConfig to from_pretrained(). This works for any model in any modality, as long as it supports loading with Accelerate and contains torch.nn.Linear layers.

from diffusers import FluxTransformer2DModel, BitsAndBytesConfig

quantization_config = BitsAndBytesConfig(load_in_8bit=True)

model_8bit = FluxTransformer2DModel.from_pretrained(
    "black-forest-labs/FLUX.1-dev", 
    subfolder="transformer",
    quantization_config=quantization_config
)

from diffusers import FluxTransformer2DModel, BitsAndBytesConfig

quantization_config = BitsAndBytesConfig(load_in_8bit=True)

model_8bit = FluxTransformer2DModel.from_pretrained(
    "black-forest-labs/FLUX.1-dev", 
    subfolder="transformer",
    quantization_config=quantization_config,
    torch_dtype=torch.float32
)
model_8bit.transformer_blocks.layers[-1].norm2.weight.dtype

Check your memory footprint with the get_memory_footprint method:

print(model.get_memory_footprint())

Quantized models can be loaded from the from_pretrained() method without needing to specify the quantization_config parameters:

from diffusers import FluxTransformer2DModel, BitsAndBytesConfig

quantization_config = BitsAndBytesConfig(load_in_4bit=True)

model_4bit = FluxTransformer2DModel.from_pretrained(
    "hf-internal-testing/flux.1-dev-nf4-pkg", subfolder="transformer"
)

8-bit (LLM.int8() algorithm)

This section explores some of the specific features of 8-bit models, such as outlier thresholds and skipping module conversion.

Outlier threshold

An “outlier” is a hidden state value greater than a certain threshold, and these values are computed in fp16. While the values are usually normally distributed ([-3.5, 3.5]), this distribution can be very different for large models ([-60, 6] or [6, 60]). 8-bit quantization works well for values ~5, but beyond that, there is a significant performance penalty. A good default threshold value is 6, but a lower threshold may be needed for more unstable models (small models or finetuning).

To find the best threshold for your model, we recommend experimenting with the llm_int8_threshold parameter in BitsAndBytesConfig:

from diffusers import FluxTransformer2DModel, BitsAndBytesConfig

quantization_config = BitsAndBytesConfig(
    load_in_8bit=True, llm_int8_threshold=10,
)

model_8bit = FluxTransformer2DModel.from_pretrained(
    "black-forest-labs/FLUX.1-dev",
    subfolder="transformer",
    quantization_config=quantization_config,
)

Skip module conversion

For some models, you don’t need to quantize every module to 8-bit which can actually cause instability. For example, for diffusion models like Stable Diffusion 3, the proj_out module can be skipped using the llm_int8_skip_modules parameter in BitsAndBytesConfig:

from diffusers import SD3Transformer2DModel, BitsAndBytesConfig

quantization_config = BitsAndBytesConfig(
    load_in_8bit=True, llm_int8_skip_modules=["proj_out"],
)

model_8bit = SD3Transformer2DModel.from_pretrained(
    "stabilityai/stable-diffusion-3-medium-diffusers",
    subfolder="transformer",
    quantization_config=quantization_config,
)

4-bit (QLoRA algorithm)

This section explores some of the specific features of 4-bit models, such as changing the compute data type, using the Normal Float 4 (NF4) data type, and using nested quantization.

Compute data type

To speedup computation, you can change the data type from float32 (the default value) to bf16 using the bnb_4bit_compute_dtype parameter in BitsAndBytesConfig:

import torch
from diffusers import BitsAndBytesConfig

quantization_config = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_compute_dtype=torch.bfloat16)

Normal Float 4 (NF4)

NF4 is a 4-bit data type from the QLoRA paper, adapted for weights initialized from a normal distribution. You should use NF4 for training 4-bit base models. This can be configured with the bnb_4bit_quant_type parameter in the BitsAndBytesConfig:

from diffusers import BitsAndBytesConfig

nf4_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
)

model_nf4 = SD3Transformer2DModel.from_pretrained(
    "stabilityai/stable-diffusion-3-medium-diffusers",
    subfolder="transformer",
    quantization_config=nf4_config,
)

For inference, the bnb_4bit_quant_type does not have a huge impact on performance. However, to remain consistent with the model weights, you should use the bnb_4bit_compute_dtype and torch_dtype values.

Nested quantization

Nested quantization is a technique that can save additional memory at no additional performance cost. This feature performs a second quantization of the already quantized weights to save an additional 0.4 bits/parameter.

from diffusers import BitsAndBytesConfig

double_quant_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
)

double_quant_model = SD3Transformer2DModel.from_pretrained(
    "stabilityai/stable-diffusion-3-medium-diffusers",
    subfolder="transformer",
    quantization_config=double_quant_config,
)

Dequantizing bitsandbytes models

Once quantized, you can dequantize the model to the original precision but this might result in a small quality loss of the model. Make sure you have enough GPU RAM to fit the dequantized model.

from diffusers import BitsAndBytesConfig

double_quant_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
)

double_quant_model = SD3Transformer2DModel.from_pretrained(
    "stabilityai/stable-diffusion-3-medium-diffusers",
    subfolder="transformer",
    quantization_config=double_quant_config,
)
model.dequantize()

Resources

• End-to-end notebook showing Flux.1 Dev inference in a free-tier Colab
• Training