nace-ai
/

kapitsa-4b

Text Generation

local-inference

Model card Files Files and versions

Kapitsa 4B

Qwen3.5 4B (4-bit, 4.6 GB) running locally on Apple Silicon through Kapitsa.

kapitsa serve --model 4b

Speed

Mac	Prefill (tok/s)	Decode (tok/s)
M4 Pro 48 GB	1,930	201
M3 Pro 36 GB	703	120
M1 Pro 16 GB	401	97

Architecture

Dense transformer, 4B parameters, 4-bit quantization.

Weights

lmstudio-community/Qwen3.5-4B-MLX-4bit

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for nace-ai/kapitsa-4b

Base model

Qwen/Qwen3.5-4B-Base

Finetuned

Qwen/Qwen3.5-4B

Quantized

lmstudio-community/Qwen3.5-4B-MLX-4bit

Finetuned

(1)

this model