Instructions to use nickprock/archai-jev-zagreus-0.4b-ita with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries
PEFT
How to use nickprock/archai-jev-zagreus-0.4b-ita with PEFT:
```
Task type is invalid.
```
Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use nickprock/archai-jev-zagreus-0.4b-ita 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 nickprock/archai-jev-zagreus-0.4b-ita
# Run inference directly in the terminal:
llama cli -hf nickprock/archai-jev-zagreus-0.4b-ita

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf nickprock/archai-jev-zagreus-0.4b-ita
# Run inference directly in the terminal:
llama cli -hf nickprock/archai-jev-zagreus-0.4b-ita

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf nickprock/archai-jev-zagreus-0.4b-ita
# Run inference directly in the terminal:
./llama-cli -hf nickprock/archai-jev-zagreus-0.4b-ita

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf nickprock/archai-jev-zagreus-0.4b-ita
# Run inference directly in the terminal:
./build/bin/llama-cli -hf nickprock/archai-jev-zagreus-0.4b-ita

Use Docker

docker model run hf.co/nickprock/archai-jev-zagreus-0.4b-ita

LM Studio
Jan

vLLM

How to use nickprock/archai-jev-zagreus-0.4b-ita with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "nickprock/archai-jev-zagreus-0.4b-ita"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "nickprock/archai-jev-zagreus-0.4b-ita",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/nickprock/archai-jev-zagreus-0.4b-ita

Ollama
How to use nickprock/archai-jev-zagreus-0.4b-ita with Ollama:
```
ollama run hf.co/nickprock/archai-jev-zagreus-0.4b-ita
```
Unsloth Desktop
Docker Model Runner
How to use nickprock/archai-jev-zagreus-0.4b-ita with Docker Model Runner:
```
docker model run hf.co/nickprock/archai-jev-zagreus-0.4b-ita
```

Lemonade

How to use nickprock/archai-jev-zagreus-0.4b-ita with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull nickprock/archai-jev-zagreus-0.4b-ita

Run and chat with the model

lemonade run user.archai-jev-zagreus-0.4b-ita-{{QUANT_TAG}}

List all available models

lemonade list

Atomic Chat

Archai JEV Zagreus 0.4B Ita (System One Decision Engine)

Archai JEV Zagreus 0.4B Ita is an ultra-lightweight, native Italian "System One" decision engine fine-tuned from mii-llm/zagreus-0.4B-ita.

Designed specifically for sub-10ms CPU inference in edge and Rust/C++ environments, this model evaluates native Italian inputs and generates deterministic output tokens (A, B, C, D, TRUE, FALSE) across various reasoning, safety, and classification tasks.

Key Highlights

Native Italian Dataset Suite: Trained on a consolidated mixture of 7 native Italian benchmark datasets (EvalITA, HateCheck-IT, RiTA-NLP, STS-B).
Dynamic Calibration Loss: Optimized via dynamic joint Cross-Entropy and Brier Score minimization, ensuring well-calibrated confidence probabilities.
Ultra-Compact Footprint: Converted to Q8_0 GGUF format with a file size of only 465 MB, allowing instant loading into RAM.

Model Details

Developer: nickprock
Base Model: mii-llm/zagreus-0.4B-ita (400M Parameters)
Architecture: LlamaForCausalLM
Fine-Tuning Method: QLoRA ($r=16, \alpha=32$)
Quantization Format: GGUF (Q8_0, 465 MB)
Target Token Mapping:
- Multiple Choice Target: A (32), B (33), C (34), D (35)
- Boolean Verification Target: TRUE (21260), FALSE (31451)

Training & Evaluation Results

Metrics Summary

Metric	Epoch 1	Epoch 2	Epoch 3	Final Test Set
Train Loss	0.6461	0.4760	0.4150	—
Validation Loss	0.4700	0.4105	0.3962	—
Validation Accuracy	77.04%	81.49%	82.18%	—
Test Accuracy	—	—	—	77.90%
Test Brier Score	—	—	—	0.2407

Dataset Mixture (Italian NLP Suite)

Safety & Moderation: Paul/hatecheck-italian + evalitahf/hatespeech_detection
Multi-Choice Reasoning: RiTA-nlp/ai2_arc_ita (ARC-Challenge + ARC-Easy)
Sentiment Analysis: evalitahf/sentiment_analysis
Natural Language Inference (NLI): evalitahf/textual_entailment
Support Pertinence: evalitahf/faq
Semantic Similarity: stsb_multi_mt (it)

Files in Repository

adapter/: LoRA adapters and tokenizer configurations.
archai-jev-zagreus-0.4b-q8.gguf: Quantized 8-bit full model (~465 MB) ready for llama.cpp and Rust integrations.

Downloads last month: 470

GGUF

Model size

0.4B params

Architecture

llama

Hardware compatibility

We're not able to determine the quantization variants.

View all variants

Model tree for nickprock/archai-jev-zagreus-0.4b-ita

Base model

mii-llm/zagreus-0.4B-ita

Adapter

(1)

this model

nickprock
/

archai-jev-zagreus-0.4b-ita