Bielik-11B-v3.0-Instruct MLX (MXFP4)

MLX MXFP4 (mixed-precision 4-bit) quantized version of speakleash/Bielik-11B-v3.0-Instruct for Apple Silicon.

Model Details

Property Value
Original Model speakleash/Bielik-11B-v3.0-Instruct
Format MLX MXFP4 (4.25 bits/weight)
Size ~5.5 GB
Peak Memory ~6.0 GB
Generation Speed ~28 tok/s (M3 Ultra)

Smallest variant - ideal for Macs with limited memory (8-16GB).

Other Quantizations

Variant Size Memory Link
bf16 22 GB 22.4 GB LibraxisAI/Bielik-11B-v3.0-mlx-bf16
q8 11 GB 11.9 GB LibraxisAI/Bielik-11B-v3.0-mlx-q8
q5 7.2 GB 7.8 GB LibraxisAI/Bielik-11B-v3.0-mlx-q5
mxfp4 (this) 5.5 GB 6.0 GB -
q4 5.9 GB 6.4 GB LibraxisAI/Bielik-11B-v3.0-mlx-q4

Usage

pip install mlx-lm

mlx_lm.generate --model LibraxisAI/Bielik-11B-v3.0-mlx-mxfp4 --prompt "Cze艣膰, jak si臋 masz?"
mlx_lm.chat --model LibraxisAI/Bielik-11B-v3.0-mlx-mxfp4
from mlx_lm import load, generate

model, tokenizer = load("LibraxisAI/Bielik-11B-v3.0-mlx-mxfp4")
response = generate(model, tokenizer, prompt="Wyja艣nij czym jest sztuczna inteligencja.", max_tokens=256)
print(response)

About Bielik

Bielik is a Polish language model developed by SpeakLeash. This conversion enables native execution on Apple Silicon Macs using the MLX framework.

License

Apache 2.0 - see original model for terms.


Converted by LibraxisAI using mlx-lm

Downloads last month
16
Safetensors
Model size
11B params
Tensor type
U8
U32
BF16
MLX
Hardware compatibility
Log In to add your hardware

4-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. 馃檵 Ask for provider support

Model tree for LibraxisAI/Bielik-11B-v3.0-mlx-mxfp4