Instructions to use IMF2000/zytturm-8b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use IMF2000/zytturm-8b with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf IMF2000/zytturm-8b:Q4_K_M # Run inference directly in the terminal: llama cli -hf IMF2000/zytturm-8b:Q4_K_M
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf IMF2000/zytturm-8b:Q4_K_M # Run inference directly in the terminal: llama cli -hf IMF2000/zytturm-8b:Q4_K_M
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-app # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama serve -hf IMF2000/zytturm-8b:Q4_K_M # Run inference directly in the terminal: ./build/bin/llama cli -hf IMF2000/zytturm-8b:Q4_K_M
Use Docker
docker model run hf.co/IMF2000/zytturm-8b:Q4_K_M
- Llama new
- LM Studio
- Jan
- Ollama
How to use IMF2000/zytturm-8b with Ollama:
ollama run hf.co/IMF2000/zytturm-8b:Q4_K_M
- Unsloth Desktop
- Pi
How to use IMF2000/zytturm-8b with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf IMF2000/zytturm-8b:Q4_K_M
Configure the model in Pi
# Install Pi: npm install -g @earendil-works/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "llama-cpp": { "baseUrl": "http://localhost:9931/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "IMF2000/zytturm-8b:Q4_K_M" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use IMF2000/zytturm-8b with Docker Model Runner:
docker model run hf.co/IMF2000/zytturm-8b:Q4_K_M
- Lemonade
How to use IMF2000/zytturm-8b with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull IMF2000/zytturm-8b:Q4_K_M
Run and chat with the model
lemonade run user.zytturm-8b-Q4_K_M
List all available models
lemonade list
- Hermes Agent
How to use IMF2000/zytturm-8b with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf IMF2000/zytturm-8b:Q4_K_M
Configure Hermes
# Install Hermes: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash hermes setup # Point Hermes at the local server: hermes config set model.provider custom hermes config set model.base_url http://127.0.0.1:9931/v1 hermes config set model.default IMF2000/zytturm-8b:Q4_K_M
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use IMF2000/zytturm-8b with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf IMF2000/zytturm-8b:Q4_K_M
Configure OpenClaw
# Install OpenClaw: npm install -g openclaw@latest # Register the local server and set it as the default model: openclaw onboard --non-interactive --mode local \ --auth-choice custom-api-key \ --custom-base-url http://127.0.0.1:9931/v1 \ --custom-model-id "IMF2000/zytturm-8b:Q4_K_M" \ --custom-provider-id llama-cpp \ --custom-compatibility openai \ --custom-text-input \ --accept-risk \ --skip-health
Run OpenClaw
openclaw agent --local --agent main --message "Hello from Hugging Face"
Update README.md
Browse files
README.md
CHANGED
|
@@ -15,25 +15,24 @@ language:
|
|
| 15 |
---
|
| 16 |
|
| 17 |
# Zytturm
|
|
|
|
| 18 |
<img src="https://upload.wikimedia.org/wikipedia/commons/c/c9/Zytturm_Ostseite_Zug.jpg" alt="Zytturm, Zug" width="300">
|
| 19 |
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
no data egress.
|
| 24 |
|
| 25 |
-
|
| 26 |
|
| 27 |
## The name
|
| 28 |
|
| 29 |
-
The Zytturm is the medieval clock tower in Zug, Switzerland,
|
| 30 |
|
| 31 |
---
|
| 32 |
|
| 33 |
## Lineage
|
| 34 |
|
| 35 |
-
Zytturm is a supervised fine-tune of **NVIDIA's Nemotron-Cascade-8B-Thinking**,
|
| 36 |
-
which is itself a post-trained derivative of **Qwen3-8B**.
|
| 37 |
|
| 38 |
| Layer | Contribution |
|
| 39 |
|:---|:---|
|
|
@@ -41,20 +40,20 @@ which is itself a post-trained derivative of **Qwen3-8B**.
|
|
| 41 |
| Post-training | Nemotron-Cascade RL pipeline (NVIDIA) |
|
| 42 |
| Domain fine-tune | Zytturm (this release) |
|
| 43 |
|
| 44 |
-
The base model
|
| 45 |
-
specialises it for the digital asset security domain — teaching it the
|
| 46 |
-
specific vocabulary, control frameworks, and reasoning patterns of the
|
| 47 |
-
field.
|
| 48 |
|
| 49 |
---
|
| 50 |
|
| 51 |
## Capabilities
|
| 52 |
|
| 53 |
-
Zytturm was
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
|
|
|
|
|
|
|
|
|
| 58 |
|
| 59 |
### Overall performance
|
| 60 |
|
|
@@ -65,15 +64,11 @@ of expected points, structure, and safety.
|
|
| 65 |
| Nemotron base (un-tuned) | 8B | Local | 2.68 |
|
| 66 |
| Qwen2.5-7B (general peer) | 7B | Local | 2.37 |
|
| 67 |
|
| 68 |
-
|
| 69 |
-
score differs by approximately 0.5 points depending on judge strictness,
|
| 70 |
-
but Zytturm is consistently above the un-tuned base and the same-size
|
| 71 |
-
general peer, and consistently below the frontier reference.
|
| 72 |
|
| 73 |
-
###
|
| 74 |
|
| 75 |
-
Zytturm reaches approximately 94% of DeepSeek V4's overall score on this
|
| 76 |
-
test set — but the two models win on different dimensions:
|
| 77 |
|
| 78 |
| Dimension | DeepSeek | Zytturm | Note |
|
| 79 |
|:---|---:|---:|:---|
|
|
@@ -82,14 +77,11 @@ test set — but the two models win on different dimensions:
|
|
| 82 |
| Structure & clarity | 3.06 | **4.14** | Fine-tune advantage |
|
| 83 |
| Safety (avoiding common mistakes) | 4.90 | 4.29 | |
|
| 84 |
|
| 85 |
-
The
|
| 86 |
-
not knowledge transfer. Zytturm answers in the register of a security
|
| 87 |
-
professional — structured, scannable, action-oriented — but its raw
|
| 88 |
-
technical accuracy remains bounded by the 8B base model.
|
| 89 |
|
| 90 |
-
###
|
| 91 |
|
| 92 |
-
Scores are the average of accuracy, coverage, structure
|
| 93 |
|
| 94 |
| Domain | Requirements | Zytturm |
|
| 95 |
|:---|---:|---:|
|
|
@@ -100,48 +92,41 @@ Scores are the average of accuracy, coverage, structure, and safety.
|
|
| 100 |
| Post-quantum fundamentals | PQ-02 | **4.50** |
|
| 101 |
| Compliance (DORA) | CR-06 | **4.12** |
|
| 102 |
|
| 103 |
-
The strongest
|
| 104 |
-
fine-tuning corpus: off-chain infrastructure security, incident reasoning,
|
| 105 |
-
and third-party supplier evaluation.
|
| 106 |
|
| 107 |
---
|
| 108 |
|
| 109 |
## Training
|
| 110 |
|
| 111 |
-
|
| 112 |
-
approximately 15,000 training examples.
|
| 113 |
|
| 114 |
### Corpus composition
|
| 115 |
|
| 116 |
-
The training
|
| 117 |
-
|
| 118 |
-
**1. Public security datasets
|
| 119 |
-
|
| 120 |
-
cryptography, SOC analyst training
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
|
| 124 |
-
|
| 125 |
-
|
| 126 |
-
corpus was generated
|
| 127 |
-
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
generation batch from dominating the training signal.
|
| 142 |
-
|
| 143 |
-
The final corpus is therefore a **blend of human-curated public data,
|
| 144 |
-
validated synthetic generation, and manual review**.
|
| 145 |
|
| 146 |
### Configuration
|
| 147 |
|
|
@@ -157,8 +142,7 @@ validated synthetic generation, and manual review**.
|
|
| 157 |
| Learning rate | 2e-4 (cosine decay) |
|
| 158 |
| Epochs | 2 |
|
| 159 |
|
| 160 |
-
Final loss
|
| 161 |
-
continued to fall through the end of training.
|
| 162 |
|
| 163 |
---
|
| 164 |
|
|
@@ -168,8 +152,7 @@ continued to fall through the end of training.
|
|
| 168 |
|:---|---:|:---|
|
| 169 |
| GGUF Q4_K_M | ~4.7 GB | GPU with 8 GB+ VRAM, or CPU |
|
| 170 |
|
| 171 |
-
The Q4_K_M quantisation
|
| 172 |
-
feed-forward tensors most sensitive to quantisation error.
|
| 173 |
|
| 174 |
---
|
| 175 |
|
|
@@ -185,11 +168,10 @@ ollama run zytturm
|
|
| 185 |
|
| 186 |
Sample `Modelfile`:
|
| 187 |
|
| 188 |
-
```
|
| 189 |
FROM ./zytturm-8b.Q4_K_M.gguf
|
| 190 |
|
| 191 |
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."
|
| 192 |
-
|
| 193 |
PARAMETER temperature 0.3
|
| 194 |
PARAMETER num_ctx 8192
|
| 195 |
PARAMETER num_predict 1024
|
|
@@ -199,71 +181,50 @@ PARAMETER num_predict 1024
|
|
| 199 |
|
| 200 |
## Intended use
|
| 201 |
|
| 202 |
-
|
| 203 |
-
- Threat modelling and attack surface analysis
|
| 204 |
-
- Smart contract vulnerability pattern recognition
|
| 205 |
-
- Post-quantum migration planning
|
| 206 |
-
- Third-party supplier and vendor risk assessment
|
| 207 |
-
- Incident root-cause analysis and control mapping
|
| 208 |
|
| 209 |
-
|
| 210 |
-
|
| 211 |
-
|
| 212 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 213 |
|
| 214 |
---
|
| 215 |
|
| 216 |
## Limitations
|
| 217 |
|
| 218 |
-
**Code generation.**
|
| 219 |
-
On the evaluation set, Solana/Anchor program generation scored 1.25/5.
|
| 220 |
-
Do not use it as a code generator. Use it as a code reviewer with a
|
| 221 |
-
human in the loop.
|
| 222 |
|
| 223 |
-
**Technical accuracy
|
| 224 |
-
is meaningfully below DeepSeek V4's (3.97) on the same test set. This is
|
| 225 |
-
a genuine capability gap, not an artefact of the evaluation. Where the
|
| 226 |
-
model's answer diverges from the reference, prefer the reference.
|
| 227 |
|
| 228 |
-
**Regressions.**
|
| 229 |
-
categories that the base model handled more capably, notably insurance
|
| 230 |
-
coverage analysis (TP-07) and blockchain address format migration
|
| 231 |
-
(PQ-06).
|
| 232 |
|
| 233 |
-
**Reasoning
|
| 234 |
-
internal thinking block before answering. At a token budget that
|
| 235 |
-
accommodates the thinking phase, it produces useful answers; at lower
|
| 236 |
-
budgets it returns empty content. Zytturm was trained on direct Q&A and
|
| 237 |
-
bypasses the deliberation phase. Comparisons between the two must account
|
| 238 |
-
for this.
|
| 239 |
|
| 240 |
---
|
| 241 |
|
| 242 |
## Evaluation methodology
|
| 243 |
|
| 244 |
-
The held-out test set was
|
| 245 |
-
|
| 246 |
-
|
| 247 |
-
security, smart contract security, post-quantum, third-party supplier
|
| 248 |
-
evaluation, compliance, and incident analysis.
|
| 249 |
|
| 250 |
-
Each
|
| 251 |
-
points, structure, and safety by three independent LLM judges. The judges
|
| 252 |
-
had access to a reference answer and a rubric per question, and no
|
| 253 |
-
knowledge of which model produced which response. Rankings were
|
| 254 |
-
consistent across all three judges.
|
| 255 |
|
| 256 |
-
The
|
| 257 |
-
|
| 258 |
-
on domain questions
|
| 259 |
-
security workflow. Real-world efficacy testing was not performed.
|
| 260 |
|
| 261 |
---
|
| 262 |
|
| 263 |
## License
|
| 264 |
|
| 265 |
-
Released under the **NVIDIA Open Model License**, inherited from the base
|
| 266 |
-
model. Commercial use, modification, and distribution permitted.
|
| 267 |
|
| 268 |
See `NOTICE` for attribution requirements.
|
| 269 |
|
|
@@ -271,7 +232,7 @@ See `NOTICE` for attribution requirements.
|
|
| 271 |
|
| 272 |
## Citation
|
| 273 |
|
| 274 |
-
```
|
| 275 |
@misc{zytturm_2026,
|
| 276 |
title = {Zytturm: An 8B Digital Asset Security Assistant},
|
| 277 |
author = {Alistair C},
|
|
|
|
| 15 |
---
|
| 16 |
|
| 17 |
# Zytturm
|
| 18 |
+
|
| 19 |
<img src="https://upload.wikimedia.org/wikipedia/commons/c/c9/Zytturm_Ostseite_Zug.jpg" alt="Zytturm, Zug" width="300">
|
| 20 |
|
| 21 |
+
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.
|
| 22 |
+
|
| 23 |
+
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.
|
|
|
|
| 24 |
|
| 25 |
+
Frontier models still have stronger general capability. The trade-off here is straightforward: a smaller specialist model that can stay inside your own environment.
|
| 26 |
|
| 27 |
## The name
|
| 28 |
|
| 29 |
+
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.
|
| 30 |
|
| 31 |
---
|
| 32 |
|
| 33 |
## Lineage
|
| 34 |
|
| 35 |
+
Zytturm is a supervised fine-tune of **NVIDIA's Nemotron-Cascade-8B-Thinking**, itself a post-trained derivative of **Qwen3-8B**.
|
|
|
|
| 36 |
|
| 37 |
| Layer | Contribution |
|
| 38 |
|:---|:---|
|
|
|
|
| 40 |
| Post-training | Nemotron-Cascade RL pipeline (NVIDIA) |
|
| 41 |
| Domain fine-tune | Zytturm (this release) |
|
| 42 |
|
| 43 |
+
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.
|
|
|
|
|
|
|
|
|
|
| 44 |
|
| 45 |
---
|
| 46 |
|
| 47 |
## Capabilities
|
| 48 |
|
| 49 |
+
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.
|
| 50 |
+
|
| 51 |
+
Each answer was scored from 1–5 across four dimensions:
|
| 52 |
+
|
| 53 |
+
- technical accuracy
|
| 54 |
+
- coverage of expected points
|
| 55 |
+
- structure
|
| 56 |
+
- safety
|
| 57 |
|
| 58 |
### Overall performance
|
| 59 |
|
|
|
|
| 64 |
| Nemotron base (un-tuned) | 8B | Local | 2.68 |
|
| 65 |
| Qwen2.5-7B (general peer) | 7B | Local | 2.37 |
|
| 66 |
|
| 67 |
+
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.
|
|
|
|
|
|
|
|
|
|
| 68 |
|
| 69 |
+
### Performance gap
|
| 70 |
|
| 71 |
+
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:
|
|
|
|
| 72 |
|
| 73 |
| Dimension | DeepSeek | Zytturm | Note |
|
| 74 |
|:---|---:|---:|:---|
|
|
|
|
| 77 |
| Structure & clarity | 3.06 | **4.14** | Fine-tune advantage |
|
| 78 |
| Safety (avoiding common mistakes) | 4.90 | 4.29 | |
|
| 79 |
|
| 80 |
+
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.
|
|
|
|
|
|
|
|
|
|
| 81 |
|
| 82 |
+
### Domain results
|
| 83 |
|
| 84 |
+
Scores below are the average of accuracy, coverage, structure and safety.
|
| 85 |
|
| 86 |
| Domain | Requirements | Zytturm |
|
| 87 |
|:---|---:|---:|
|
|
|
|
| 92 |
| Post-quantum fundamentals | PQ-02 | **4.50** |
|
| 93 |
| Compliance (DORA) | CR-06 | **4.12** |
|
| 94 |
|
| 95 |
+
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.
|
|
|
|
|
|
|
| 96 |
|
| 97 |
---
|
| 98 |
|
| 99 |
## Training
|
| 100 |
|
| 101 |
+
Zytturm was trained using supervised fine-tuning with QLoRA (rank 32) on a curated domain corpus of approximately 15,000 examples.
|
|
|
|
| 102 |
|
| 103 |
### Corpus composition
|
| 104 |
|
| 105 |
+
The training data came from three main sources.
|
| 106 |
+
|
| 107 |
+
**1. Public security datasets**
|
| 108 |
+
|
| 109 |
+
Six openly licensed datasets were used, covering smart contract vulnerabilities, on-chain attack analysis, post-quantum cryptography, SOC analyst training and general cyber defence.
|
| 110 |
+
|
| 111 |
+
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.
|
| 112 |
+
|
| 113 |
+
**2. Synthetically generated examples**
|
| 114 |
+
|
| 115 |
+
A substantial part of the corpus was generated through the DeepSeek API. The requirement specification was used as the source of truth for coverage.
|
| 116 |
+
|
| 117 |
+
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.
|
| 118 |
+
|
| 119 |
+
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.
|
| 120 |
+
|
| 121 |
+
**3. Human curation and review**
|
| 122 |
+
|
| 123 |
+
All examples were normalised into a common ShareGPT format.
|
| 124 |
+
|
| 125 |
+
A stratified random sample was reviewed manually and marked as good, bad or requiring further review. Examples marked bad were removed.
|
| 126 |
+
|
| 127 |
+
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.
|
| 128 |
+
|
| 129 |
+
The final corpus combines **human-curated public data, validated synthetic examples and manual review**.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 130 |
|
| 131 |
### Configuration
|
| 132 |
|
|
|
|
| 142 |
| Learning rate | 2e-4 (cosine decay) |
|
| 143 |
| Epochs | 2 |
|
| 144 |
|
| 145 |
+
Final loss was 0.84 train / 0.87 eval. Eval loss was still falling at the end of training, with no observed overfitting.
|
|
|
|
| 146 |
|
| 147 |
---
|
| 148 |
|
|
|
|
| 152 |
|:---|---:|:---|
|
| 153 |
| GGUF Q4_K_M | ~4.7 GB | GPU with 8 GB+ VRAM, or CPU |
|
| 154 |
|
| 155 |
+
The Q4_K_M quantisation keeps full quality on the attention and feed-forward tensors that are most sensitive to quantisation error.
|
|
|
|
| 156 |
|
| 157 |
---
|
| 158 |
|
|
|
|
| 168 |
|
| 169 |
Sample `Modelfile`:
|
| 170 |
|
| 171 |
+
```text
|
| 172 |
FROM ./zytturm-8b.Q4_K_M.gguf
|
| 173 |
|
| 174 |
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."
|
|
|
|
| 175 |
PARAMETER temperature 0.3
|
| 176 |
PARAMETER num_ctx 8192
|
| 177 |
PARAMETER num_predict 1024
|
|
|
|
| 181 |
|
| 182 |
## Intended use
|
| 183 |
|
| 184 |
+
Zytturm is intended for:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 185 |
|
| 186 |
+
- design review support for digital asset platforms
|
| 187 |
+
- threat modelling and attack surface analysis
|
| 188 |
+
- smart contract vulnerability pattern recognition
|
| 189 |
+
- post-quantum migration planning
|
| 190 |
+
- third-party supplier and vendor risk assessment
|
| 191 |
+
- incident root-cause analysis and control mapping
|
| 192 |
+
|
| 193 |
+
**Out of scope:** real-time price data, transaction execution, legal advice, formal compliance certification and penetration testing against live systems.
|
| 194 |
+
|
| 195 |
+
It is not a replacement for human audit, legal counsel or vendor due diligence.
|
| 196 |
|
| 197 |
---
|
| 198 |
|
| 199 |
## Limitations
|
| 200 |
|
| 201 |
+
**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.
|
|
|
|
|
|
|
|
|
|
| 202 |
|
| 203 |
+
**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.
|
|
|
|
|
|
|
|
|
|
| 204 |
|
| 205 |
+
**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).
|
|
|
|
|
|
|
|
|
|
| 206 |
|
| 207 |
+
**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.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 208 |
|
| 209 |
---
|
| 210 |
|
| 211 |
## Evaluation methodology
|
| 212 |
|
| 213 |
+
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.
|
| 214 |
+
|
| 215 |
+
The test set covers 42 requirements across digital asset security, smart contract security, post-quantum, third-party supplier evaluation, compliance and incident analysis.
|
|
|
|
|
|
|
| 216 |
|
| 217 |
+
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.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 218 |
|
| 219 |
+
The model ordering was consistent across all three judges.
|
| 220 |
+
|
| 221 |
+
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.
|
|
|
|
| 222 |
|
| 223 |
---
|
| 224 |
|
| 225 |
## License
|
| 226 |
|
| 227 |
+
Released under the **NVIDIA Open Model License**, inherited from the base model. Commercial use, modification and distribution are permitted.
|
|
|
|
| 228 |
|
| 229 |
See `NOTICE` for attribution requirements.
|
| 230 |
|
|
|
|
| 232 |
|
| 233 |
## Citation
|
| 234 |
|
| 235 |
+
```bibtex
|
| 236 |
@misc{zytturm_2026,
|
| 237 |
title = {Zytturm: An 8B Digital Asset Security Assistant},
|
| 238 |
author = {Alistair C},
|