lfm2-5-8b-a1b-mxfp4-mlx

MLX quantization of LiquidAI/LFM2.5-8B-A1B for Apple Silicon.

Variant: Block float MX FP4
Disk size: 4308 MB
Quantized by: sahilchachra

Benchmark results

Evaluated on Apple M5 Pro with MLX. Model loaded once; performance and quality measured in a single pass.

Performance

	This model	FP16 baseline
Decode tok/s (steady-state)	205.03	77.32
Prefill tok/s (steady-state)	450.86	227.43
Decode tok/s (avg, long traces)	196.26	76.34
Peak memory (GB)	4.893	17.142
Disk size (MB)	4308	16169

Warmed, short-prompt, chat-templated, thinking disabled. Represents steady-state decode for typical chat use; long thinking traces will be slower due to KV-cache growth.

Quality

Benchmark	This model	FP16 baseline	n
MATH-500 (math reasoning)	56.7% (answered 23/30)	70.0% (answered 25/30)	30
IFEval (instruction following)	77.3%	79.5%	44
HumanEval (code, pass@1)	73.3%	73.3%	30

MATH-500 per-level accuracy

Level	This model	FP16 baseline
level 1	83.3%	83.3%
level 2	83.3%	100.0%
level 3	50.0%	50.0%
level 4	66.7%	66.7%
level 5	0.0%	50.0%

Context scaling (decode tok/s)

Context length	Decode tok/s
~128 tokens	206.1
~256 tokens	205.4
~512 tokens	205.6
~1024 tokens	204.5

Usage

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("sahilchachra/lfm2-5-8b-a1b-mxfp4-mlx")
response = generate(model, tokenizer, prompt="Your prompt here", max_tokens=256, verbose=True)

All variants in this collection

Model	Variant
sahilchachra/lfm2-5-8b-a1b-mxfp4-mlx	Block float MX FP4 ← this model
sahilchachra/lfm2-5-8b-a1b-optiq-5bpw-mlx	OptiQ mixed-precision (target 5.0 bpw)

Notes

• Requires Apple Silicon (M1 or later) with MLX
• Benchmarks run on Apple M5 Pro, 24 GB unified memory
• License: see LiquidAI/LFM2.5-8B-A1B for the original model's license

Original model

See LiquidAI/LFM2.5-8B-A1B for full model details and intended use.