NIST-focused local model trained on 596 publications and 530K examples. Good for brainstorming, drafts, and finding likely docs. v1.1.1 adds eval caveats. Verify exact controls, RMF steps, and SP 800-207 components against source docs.

HackIDLE-NIST-Coder v1.1.1

HackIDLE-NIST-Coder v1.1.1 is a prompt and metadata patch over the v1.1 model built from Qwen2.5-Coder-7B-Instruct and trained on 530,912 examples from 596 NIST publications.

Use it for NIST-grounded brainstorming, draft explanations, and finding likely publications or control families to read next.

Do not use it as the source of truth for exact control names, RMF step ordering, SP 800-207 reference-architecture components, or compliance language that will be reused without checking the original publication.

Training data

The underlying v1.1 corpus includes:

FIPS cryptographic standards
SP 800 security guidance and controls
SP 1800 practice guides
NIST IR technical reports
CSWP material, including CSF 2.0, Zero Trust, Post-Quantum Cryptography, IoT security, and privacy engineering

Dataset:

https://huggingface.co/datasets/ethanolivertroy/nist-cybersecurity-training

Current eval status

I ran a small local smoke eval on April 22, 2026 against etgohome/hackidle-nist-coder:latest. In that local Ollama install, latest matched the v1.1 line.

Result: 1/5 cases passed.

The model stayed in-domain and handled a rough FIPS 140-2 vs. FIPS 140-3 comparison. It still missed exact grounding on:

SP 800-207 reference-architecture component names
the full SP 800-37 Rev. 2 RMF sequence
the exact CM-6 control name and description
stronger publication selection and logging/audit grounding for a contractor remote-access planning prompt

That is the important limitation. The model can sound close while still being wrong on exact NIST structure.

Good uses

This model is useful for:

brainstorming where to start in NIST
drafting first-pass explanations
surfacing likely document families
turning NIST-flavored questions into something a human can verify
local experimentation with domain fine-tuning on Apple Silicon

It is not reliable enough yet for:

exact control names
exact framework step ordering
exact reference-architecture component naming
answers that need source-level correctness on the first pass

Training method

Base model: Qwen2.5-Coder-7B-Instruct-4bit
Fine-tuning method: LoRA with MLX
Version note: v1.1.1 did not retrain the model. It updates the public prompt, model-card language, and eval caveat.
Hardware: Apple M4 Max, 128 GB RAM
Training iterations: 1,000, plus checkpoint recovery work
Final training loss: 1.420
Best validation loss: 1.512
Trainable parameters: 11.5M

Other formats

License

The base model is Qwen2.5-Coder-7B-Instruct, released under Apache 2.0. The NIST source publications used for the dataset are public domain U.S. government works.