IMF2000/zytturm-8b

Zytturm

Zytturm is an 8B language model fine-tuned for digital asset security. It covers custody, smart contract security, post-quantum migration, third-party supplier evaluation, and incident analysis. It runs locally on consumer hardware with no data egress.

The model is intended for security work where sending prompts or internal data to a third-party API is not appropriate. That includes regulated, data-residency-sensitive and air-gapped environments. Zytturm runs entirely on local hardware, with no network dependency, telemetry or vendor relationship.

Frontier models still have stronger general capability. The trade-off here is straightforward: a smaller specialist model that can stay inside your own environment.

The name

The Zytturm is the medieval clock tower in Zug, Switzerland, at the centre of Crypto Valley and home to the Ethereum Foundation. The tower dates to the thirteenth century, and its astronomical clock has kept time since 1574.

---

Lineage

Zytturm is a supervised fine-tune of NVIDIA's Nemotron-Cascade-8B-Thinking, itself a post-trained derivative of Qwen3-8B.

Layer	Contribution
Architecture	Qwen3-8B
Post-training	Nemotron-Cascade RL pipeline (NVIDIA)
Domain fine-tune	Zytturm (this release)

The base model supplies the general reasoning capability. The Zytturm fine-tune adds domain vocabulary, control frameworks and answer patterns specific to digital asset security.

---

Capabilities

Zytturm was tested on a held-out set of 93 domain questions covering 42 requirements. Three independent LLM judges scored the answers: DeepSeek V4, Gemini 3.1 Flash-Lite and Gemini 3.8 Flash.

Each answer was scored from 1–5 across four dimensions:

• technical accuracy
• coverage of expected points
• structure
• safety

Overall performance

Model	Size	Deployment	Avg Score
DeepSeek V4	~600B	Cloud API	3.73
Zytturm	8B	Local	3.51
Nemotron base (un-tuned)	8B	Local	2.68
Qwen2.5-7B (general peer)	7B	Local	2.37

The ordering was the same across all three judges. Absolute scores moved by roughly 0.5 points depending on how strictly each judge scored, but Zytturm remained above the un-tuned base and same-size general model, and below the frontier reference.

Performance gap

Zytturm reaches approximately 94% of DeepSeek V4's overall score on this test set. The difference becomes clearer when the score is broken down by dimension:

Dimension	DeepSeek	Zytturm	Note
Technical accuracy	3.97	2.91	Frontier knowledge advantage
Coverage of expected points	3.00	2.69
Structure & clarity	3.06	4.14	Fine-tune advantage
Safety (avoiding common mistakes)	4.90	4.29

The fine-tune has its biggest effect on how answers are structured and presented. Zytturm tends to produce concise, scannable and action-oriented security responses. Technical accuracy is still constrained by the underlying 8B model.

Domain results

Scores below are the average of accuracy, coverage, structure and safety.

Domain	Requirements	Zytturm
Off-chain infrastructure & bridges	LL-09, LL-10	4.17–4.75
Smart contract vulnerability detection	SC-01, SC-02	3.75–4.42
Vendor contract clauses	TP-06	4.50
Incident reasoning & controls	LL-04, LL-06	4.00–4.25
Post-quantum fundamentals	PQ-02	4.50
Compliance (DORA)	CR-06	4.12

The strongest results line up with areas that received the most attention in the training corpus, particularly off-chain infrastructure security, incident reasoning and third-party supplier evaluation.

---

Training

Zytturm was trained using supervised fine-tuning with QLoRA (rank 32) on a curated domain corpus of approximately 15,000 examples.

Corpus composition

The training data came from three main sources.

1. Public security datasets

Six openly licensed datasets were used, covering smart contract vulnerabilities, on-chain attack analysis, post-quantum cryptography, SOC analyst training and general cyber defence.

The sources included community datasets from smart contract auditing, blockchain forensics and threat intelligence. All public sources were released under permissive licences including MIT, Apache-2.0 and CC-BY-4.0.

2. Synthetically generated examples

A substantial part of the corpus was generated through the DeepSeek API. The requirement specification was used as the source of truth for coverage.

Synthetic generation concentrated on areas where suitable public datasets were limited or unavailable, including compliance mapping, incident reasoning, third-party supplier evaluation and post-quantum migration strategy.

Smart contract vulnerability examples went through an additional programmatic check. Generated code was rejected unless it actually contained the claimed vulnerability and the accompanying answer identified the relevant vulnerable line.

3. Human curation and review

All examples were normalised into a common ShareGPT format.

A stratified random sample was reviewed manually and marked as good, bad or requiring further review. Examples marked bad were removed.

The resulting corpus was then rebalanced across requirement categories. Thin categories were upsampled and disproportionately large sources were downsampled so that no single dataset or generation batch dominated training.

The final corpus combines human-curated public data, validated synthetic examples and manual review.

Configuration

Parameter	Value
Base model	nvidia/Nemotron-Cascade-8B-Thinking
Method	QLoRA (4-bit)
LoRA rank	32
LoRA alpha	64
Target modules	All linear layers
Sequence length	4,096
Effective batch size	16
Learning rate	2e-4 (cosine decay)
Epochs	2

Final loss was 0.84 train / 0.87 eval. Eval loss was still falling at the end of training, with no observed overfitting.

---

Formats

Format	Size	Target hardware
GGUF Q4_K_M	~4.7 GB	GPU with 8 GB+ VRAM, or CPU

The Q4_K_M quantisation keeps full quality on the attention and feed-forward tensors that are most sensitive to quantisation error.

---

Quick start

Requires Ollama.

# Download the GGUF, then:
ollama create zytturm -f Modelfile
ollama run zytturm

Sample Modelfile:

FROM ./zytturm-8b.Q4_K_M.gguf

SYSTEM "You are Zytturm, a digital asset security professional assistant. You provide expert guidance on custody, smart contract security, post-quantum migration, third-party supplier evaluation, and lessons from major incidents."
PARAMETER temperature 0.3
PARAMETER num_ctx 8192
PARAMETER num_predict 1024

---

Intended use

Zytturm is intended for:

• design review support for digital asset platforms
• threat modelling and attack surface analysis
• smart contract vulnerability pattern recognition
• post-quantum migration planning
• third-party supplier and vendor risk assessment
• incident root-cause analysis and control mapping

Out of scope: real-time price data, transaction execution, legal advice, formal compliance certification and penetration testing against live systems.

It is not a replacement for human audit, legal counsel or vendor due diligence.

---

Limitations

Code generation. Solidity and Rust generation is unreliable. Solana/Anchor program generation scored 1.25/5 on the evaluation set. Zytturm should not be used as a code generator. It is better suited to code review with a human in the loop.

Technical accuracy. Zytturm scored 2.91 for technical accuracy against DeepSeek V4's 3.97 on the same test set. The gap is real and material. Where an answer conflicts with a trusted reference, use the reference.

Regressions. Domain fine-tuning reduced performance in some areas where the base model performed better, particularly insurance coverage analysis (TP-07) and blockchain address format migration (PQ-06).

Reasoning behaviour. The un-tuned base model produces a long internal thinking block before returning an answer. With enough token budget for that phase it can produce useful responses; at lower budgets it may return empty content. Zytturm was trained on direct Q&A and bypasses that deliberation phase, so comparisons need to account for the difference.

---

Evaluation methodology

The held-out test set was created independently from the training corpus. Questions used different wording and framing from the training data to test generalisation rather than simple memorisation.

The test set covers 42 requirements across digital asset security, smart contract security, post-quantum, third-party supplier evaluation, compliance and incident analysis.

Each answer was scored from 1–5 for technical accuracy, coverage of expected points, structure and safety. Three independent LLM judges performed the scoring. Each judge received a reference answer and question-specific rubric but was not told which model had produced the response.

The model ordering was consistent across all three judges.

This evaluation is comparative rather than absolute. It measures how closely model answers match expert-written references on the selected domain questions. It does not establish how useful the model will be in a live security workflow. Real-world efficacy testing was not performed.

---

License

Released under the NVIDIA Open Model License, inherited from the base model. Commercial use, modification and distribution are permitted.

See NOTICE for attribution requirements.

---

Citation

@misc{zytturm_2026,
  title  = {Zytturm: An 8B Digital Asset Security Assistant},
  author = {Alistair C},
  year   = {2026},
  note   = {Fine-tune of nvidia/Nemotron-Cascade-8B-Thinking}
}

---

Header image: "Zytturm Ostseite Zug" by Schulerst, licensed under CC BY-SA 3.0.