Laguna-XS-2.1-NVFP4-mlx

[!WARNING]

FOR TESTING ONLY

This is an experimental NVFP4 MLX build of Laguna XS 2.1, published for testing purposes only. It has not been validated for quality or correctness and is not an official release. Do not use it in production. For supported checkpoints, see the Laguna XS 2.1 model card and its variants.

<p align="center"> <img alt="poolside-banner" src="https://poolside.ai/assets/laguna/laguna-xs2-1-banner.svg" width="800px"> </p>

Laguna XS 2.1-NVFP4-mlx

Laguna XS 2.1-NVFP4-mlx is an NVFP4 (4-bit) MLX build of Laguna XS 2.1, a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token designed for agentic coding and long-horizon work on a local machine. It uses Sliding Window Attention with per-head gating in 30 out of 40 layers for fast inference and low KV cache requirements.

Usage

from pathlib import Path
from mlx_lm import load, generate

MODEL_PATH = "poolside/Laguna-XS-2.1-NVFP4-mlx"

model, tokenizer, config = load(
    MODEL_PATH,
    tokenizer_config={"trust_remote_code": True},
    return_config=True,
)


prompt = "write go code to print the pascal triangle"
response = generate(model, tokenizer, prompt=prompt, verbose=True, max_tokens=256)

License

This model is licensed under the OpenMDW-1.1 License.

Intended and Responsible Use

Laguna XS 2.1-NVFP4-mlx is designed for software engineering and agentic coding use cases, and you are responsible for confirming that it is appropriate for your intended application. Laguna XS 2.1-NVFP4-mlx is subject to the OpenMDW-1.1 License, and should be used consistently with Poolside's Acceptable Use Policy. We advise against circumventing Laguna XS 2.1-NVFP4-mlx safety guardrails without implementing substantially equivalent mitigations appropriate for your use case.

Please report security vulnerabilities or safety concerns to [email protected].