---

license: cc-by-nc-4.0

base_model: FrontisAI/Frontis-MA1-30B

base_model_relation: quantized

library_name: llama.cpp

pipeline_tag: text-generation

tags:

- openmle

- frontis-ma1

- gguf

- q4-k-m

- local-inference

- moe

- coding

---

Frontis-MA1-30B-GGUF

<p align="center">

<a href="https://arxiv.org/abs/2607.28568">📄 Paper</a>

&nbsp;•&nbsp;

<a href="https://frontisai.github.io/OpenRSI/">🌐 Project</a>

&nbsp;•&nbsp;

<a href="https://github.com/FrontisAI/OpenRSI">💻 Code</a>

&nbsp;•&nbsp;

<a href="https://huggingface.co/collections/FrontisAI/frontis-ma1">🤗 Models</a>

&nbsp;•&nbsp;

<a href="https://huggingface.co/datasets/FrontisAI/OpenMLE-Tasks">🧩 Tasks</a>

&nbsp;•&nbsp;

<a href="https://huggingface.co/datasets/FrontisAI/OpenMLE-SFT-Traces">📚 SFT Traces</a>

</p>

This repository is the official Q4_K_M GGUF derivative of Frontis-MA1-30B. It is provided for practical local inference with llama.cpp.

It accompanies the paper Frontis-MA1: Training an AI4AI Model towards Recursive Self-Improvement in Machine Learning Engineering and the OpenRSI code release.

Files

| File | Size | Purpose |

| --- | ---: | --- |

| Frontis-MA1-30B-Q4_K_M.gguf | 17.28 GiB | Q4_K_M language model |

| checksums.txt | — | SHA-256 integrity manifest |

Only one quantization is published intentionally. Q4_K_M is the default local-deployment format for this release; the canonical BF16 Transformers weights remain in the base repository.

Quickstart

Tested conversion and inference tool: llama.cpp b9637, commit aedb2a5e9ca3d4064148bbb919e0ddc0c1b70ab3.

llama-cli \
  -m ./Frontis-MA1-30B-Q4_K_M.gguf \
  -ngl all \
  -c 32768 \
  -n 1024 \
  -cnv -st --simple-io \
  -p "Build a strong tabular classification baseline and explain the validation design."

Reduce -c when memory is limited. On systems that cannot offload all layers, set -ngl to a smaller value or let llama.cpp choose automatically.

The bundled chat template always begins an explicit thinking block. Allow sufficient -n budget for the model to finish thinking and produce its final answer.

Release validation

The final file passed SHA-256 verification, a full 579-tensor GGUF structure read, and a real llama-cli load-and-generate smoke with full GPU offload on one NVIDIA H200. This validates the release artifact and command path; it is not a consumer-hardware speed benchmark.

Scope and quality

• This is a lossy 4-bit quantization. Use the BF16 repository when maximum fidelity or paper-result reproduction is required.
• The paper's reported scores were obtained with the canonical model and the OpenMLE-Evo harness; they are not GGUF one-shot benchmark results.
• Generated code may be incorrect or unsafe. Execute it only in an isolated environment with explicit resource limits.

Paper result

The canonical BF16 model reaches 53.03% Medal Average and 0.7055 Human Rank with OpenMLE-Evo on the official 22-task MLE-Bench Lite split, compared with 34.85% and 0.5573 for its base model under the same harness. These are BF16 model–harness results, not GGUF one-shot scores.

License

Original Frontis-MA1 material is released under CC BY-NC 4.0 for attribution-required, non-commercial use. Commercial use is not granted. The upstream Qwen Apache License 2.0 notice is preserved in LICENSE-UPSTREAM-APACHE-2.0 and NOTICE.