Downloads · 30 days
71
24% of all-time downloads
LFM2.5-350M-MLX-8bit
ipetrukha/LFM2.5-350M-MLX-8bit
LFM2.5-350M-MLX-8bit is a text generation model from ipetrukha. Use it when you need the model to write or continue text. It is set up for mlx.
Re-quantization of LiquidAI/LFM2.5-350M (revision 9e6c6ccf47cd318696e137d381a7ded8fe4df09f) with mlx-lm 0.31.3: affine, 8 bit, group size 64, no per-layer overrides — the same recipe as LiquidAI/LFM2.5-350M-MLX-8bit.
All-time downloads
291
Public
Parameters
354M
377 MB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors377 MB · 99%
Parameter types
How the weights are stored.
U32354M · 100%
Try a prompt
Base models
Model card
From the Hugging Face model README
LFM2.5-350M-MLX-8bit
Re-quantization of LiquidAI/LFM2.5-350M (revision 9e6c6ccf47cd318696e137d381a7ded8fe4df09f) with mlx-lm 0.31.3: affine, 8 bit, group size 64, no per-layer overrides — the same recipe as LiquidAI/LFM2.5-350M-MLX-8bit.
The one difference: this checkpoint stores quantization scales and biases in bfloat16, while the official MLX quant stores them in float32. Published so cross-engine benchmarks compare engines under one activation dtype.
Reproduce:
mlx_lm.convert --hf-path LiquidAI/LFM2.5-350M -q --q-bits 8 --q-group-size 64
- Task
- Text Generation
- Library
- mlx
- Type
- lfm2
- Languages
- en
- Created
- Sep 8, 2026
- Updated
- Sep 8, 2026