AxionML / Laguna-XS-2.1-NVFP4

AxionML Laguna-XS-2.1-NVFP4

Mirrored by AxionML for open-source serving and deployment use cases. Part of AxionML's effort to provide ready-to-serve quantized models for the community.

Quantized by Poolside. The weights in this repository are an unmodified copy of poolside/Laguna-XS-2.1-NVFP4 (revision d32afde8b09af1539b49ff96ff5551c674485f8e ). All credit for the quantization belongs to Poolside.

This is an NVFP4-quantized version of poolside/Laguna-XS-2.1 (33B total parameters, 3B activated), quantized with NVIDIA Model Optimizer .

About NVFP4 quantization: NVFP4 on Blackwell couples a compact E2M1 FP4 codebook with blockwise FP8 (E4M3) scaling over 16-element micro-blocks, so that 4-bit stored values remain numerically useful for neural-network computation. The E2M1 codebook provides a small, nonuniform set of representable magnitudes up to ±6 and relies on saturating behavior rather than IEEE NaN/Inf encodings to maximize usable range per bit. Using an FP8 block scale (rather than power-of-two-only E8M0) enables fractional scales and error-minimizing scale selection. On Blackwell Tensor Cores, native FP4 multipliers exploit E2M1 simplicity while higher-precision FP32 accumulation protects dot-product accuracy.

Ready for commercial and non-commercial use under OpenMDW-1.1 (included as LICENSE.md ), consistent with Poolside's Acceptable Use Policy .

Model Summary

Evaluation Results

Scores reported by Poolside for this checkpoint (mean pass@1, Harbor framework).

Quantization Details

• Quantization format: NVFP4 (compressed-tensors nvfp4-pack-quantized , group size 16) on routed and shared expert MLPs; Poolside's official NVFP4 release
• KV cache: FP8 (static per-tensor)

Usage

Deploy with SGLang

python3 -m sglang.launch_server \
    --model-path AxionML/Laguna-XS-2.1-NVFP4 \
    --trust-remote-code

Deploy with vLLM

vllm serve AxionML/Laguna-XS-2.1-NVFP4 \
    --enable-auto-tool-choice \
    --tool-call-parser poolside_v1 \
    --reasoning-parser poolside_v1

Quantization is detected automatically in vLLM, SGLang ( sgl-project/sglang#24204 ) and Transformers. The FP8 KV cache needs vLLM ≥ 0.22.0; tool calling needs a vLLM build with vllm#47311 (otherwise use --tool-call-parser glm47 ).

Limitations

The base model was trained on data that may contain toxic language and societal biases. The quantized model inherits these limitations. It may generate inaccurate, biased, or offensive content. Please refer to the original model card and the upstream quantized model card for full details.

Credits

• Base model: poolside/Laguna-XS-2.1
• Quantization: poolside/Laguna-XS-2.1-NVFP4 by Poolside
• Mirror: AxionML