Instructions to use simoneschiavoi/VisionPsy-Nano-DomCal-Slim with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use simoneschiavoi/VisionPsy-Nano-DomCal-Slim 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 simoneschiavoi/VisionPsy-Nano-DomCal-Slim:Q4_0
# Run inference directly in the terminal:
llama cli -hf simoneschiavoi/VisionPsy-Nano-DomCal-Slim:Q4_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf simoneschiavoi/VisionPsy-Nano-DomCal-Slim:Q4_0
# Run inference directly in the terminal:
llama cli -hf simoneschiavoi/VisionPsy-Nano-DomCal-Slim:Q4_0

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 simoneschiavoi/VisionPsy-Nano-DomCal-Slim:Q4_0
# Run inference directly in the terminal:
./llama-cli -hf simoneschiavoi/VisionPsy-Nano-DomCal-Slim:Q4_0

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 simoneschiavoi/VisionPsy-Nano-DomCal-Slim:Q4_0
# Run inference directly in the terminal:
./build/bin/llama-cli -hf simoneschiavoi/VisionPsy-Nano-DomCal-Slim:Q4_0

Use Docker

docker model run hf.co/simoneschiavoi/VisionPsy-Nano-DomCal-Slim:Q4_0

LM Studio
Jan
Ollama
How to use simoneschiavoi/VisionPsy-Nano-DomCal-Slim with Ollama:
```
ollama run hf.co/simoneschiavoi/VisionPsy-Nano-DomCal-Slim:Q4_0
```

Unsloth Studio

How to use simoneschiavoi/VisionPsy-Nano-DomCal-Slim with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for simoneschiavoi/VisionPsy-Nano-DomCal-Slim to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for simoneschiavoi/VisionPsy-Nano-DomCal-Slim to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for simoneschiavoi/VisionPsy-Nano-DomCal-Slim to start chatting

Docker Model Runner
How to use simoneschiavoi/VisionPsy-Nano-DomCal-Slim with Docker Model Runner:
```
docker model run hf.co/simoneschiavoi/VisionPsy-Nano-DomCal-Slim:Q4_0
```

Lemonade

How to use simoneschiavoi/VisionPsy-Nano-DomCal-Slim with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull simoneschiavoi/VisionPsy-Nano-DomCal-Slim:Q4_0

Run and chat with the model

lemonade run user.VisionPsy-Nano-DomCal-Slim-Q4_0

List all available models

lemonade list

Atomic Chat

VisionPsy-Nano DomCal-Slim

The smallest package measured anywhere in this study (325.9 MiB) — 17.0% below QVAC's flagship q4_k_m-imat and 2.5% below QVAC's own smallest configuration — with near-flagship quality: 2W/15L/0T over 17 displayed rows vs QVAC q4_k_m-imat (same-harness re-run) and 13W/3L/1T vs QVAC q4_0 (same-harness).

A GGUF quantized derivative of QVAC's VisionPsy-Nano-460M built with P1 (domcal) × P10 (projcomp) — domain-calibrated imatrix quantization plus multimodal-projector compression (q8 → q6_K). This is not a new trained checkpoint: the ~460M-parameter architecture is unchanged. Part of the VisionPsy-Nano release collection — see the Links section below.

Exact runtime pair

Component	File	Bytes
LM	`visionpsy-nano-460m-q4_0-domcal.gguf`	256379776
mmproj	`MY_q4_0-domcal-mmproj-q6.gguf`	85355072

Package size: 325.9 MiB. Use only this LM/mmproj mapping. SHA-256 checksums for both files ship in this repository as SHA256SUMS — verify after download with sha256sum -c SHA256SUMS.

Model at a glance


Base model	QVAC VisionPsy-Nano-460M (~460M parameters; SigLIP2 vision encoder + SmolLM2-360M backbone)
Techniques	P1 (domcal) — domain-calibrated imatrix quantization; P10 (projcomp) — multimodal projector compressed q8 → q6_K
Quantization	q4_0 LM (byte-identical to the DomCal artifact) + q6_K mmproj (81.4 MiB, down from 103.7)
Calibration data	VQAv2-train + TextVQA-train, ChatML-formatted; training splits only — never benchmark test data, no evaluation images
Total size	325.9 MiB — smallest of the 15 measured configurations: −66.9 MiB (−17.0%) vs QVAC q4_k_m-imat (392.8), −6.3% vs QVAC q4_0 (347.7), −2.5% vs QVAC's own smallest, iq3_xxs-imat (334.2)
Headline	13W/3L/1T vs QVAC q4_0 (same-harness); normalized 60.30 — parity with the 348.2 MiB DomCal package (60.31) from the identical LM; 2W/15L/0T over 17 displayed rows vs QVAC q4_k_m-imat (same-harness re-run); RealWorldQA 60.65, best of all six ledger models
Observed speed	817.5 ms/item — the storage-first option: +14.0 ms/item vs q4_0, +4.3 ms/item (+0.5%) vs imat. The size win, not a speed claim

Why this build exists

The study measured a practical compression floor in the projector: moving the mmproj from Q8 to q6_K removes 21.5% of the projector and 6.4% of the complete package with no statistically detectable quality change (pooled p = 0.49) — and both q5_K implementations failed their quality or embedding-equivalence gates. The floor is q6_K, not "as few bits as possible". DomCal-Slim ships exactly that floor, stacked on the DomCal repair: minimum download and disk footprint at parity with its 348.2 MiB parent.

Vs the two rulers

The primary ruler is QVAC q4_k_m-imat, QVAC's flagship build — re-run in this same harness with a hash-pinned full-17 record (judged rows scored with a qwen3.6-27b API judge, a reconstruction of QVAC's judging protocol validated within ±1 pt of their published card on 6/8 judged benchmarks; QVAC's own card numbers remain labeled context, never medaled). QVAC q4_0 is the secondary same-harness ruler. Negative deltas = smaller/faster.

	QVAC q4_k_m-imat (same-harness re-run)	QVAC q4_0 (same-harness)
W/L/T over 17 displayed rows	2W/15L/0T	13W/3L/1T
Package size Δ (325.9 MiB)	−66.9 MiB (−17.0%)	−21.8 MiB (−6.3%)
Speed Δ (817.5 ms/item)	+4.3 ms/item (+0.5%)	+14.0 ms/item (+1.7%)

Measured results

Normalized 60.30 vs q4_0's 59.48 (+0.82) at 6.3% less package; RealWorldQA 60.65 is the outright row-gold across all six ledger models; MMMU dev 32.67 is a three-way tie at the top with DomCal and QVAC q4_0; MM-IFEval@2048 38.79 beats the q4_0 re-run (34.82). The named losses stay visible: QVAC q4_0 keeps MME (1562.5 vs 1527.1), POPE and MMVet; QVAC q4_k_m-imat keeps OCRBench, DocVQA, ChartQA and InfoVQA. This is the efficiency play, not an imat-beater.

Limitations

Exploratory scope: one seed (17), one harness, one GPU. Counts are descriptive rows, not universal-superiority claims.
Speed honesty: the slowest of the six same-harness packages (+1.7% vs q4_0, +0.5% vs imat) — chosen for footprint, not latency.
Judged rows: qwen3.6-27b via OpenRouter — an attempted same-model reconstruction of QVAC's judge, not their exact serving protocol.
Projector compression is quality-neutral at q6_K (pooled p = 0.49); the rejected q5_K step is documented in the ledger (release repository).

License and attribution

Apache-2.0 derivative. The Apache-2.0 NOTICE distributed with the artifact must be retained, and QVAC's VisionPsy-Nano-460M must be attributed as the base model. Build evidence, evaluation ledger, and reproduction scripts: https://github.com/simoneschiavoi/visionpsy-optimization. All four release packages: https://huggingface.co/simoneschiavoi.

Downloads last month: -

GGUF

Model size

0.4B params

Architecture

llama

Hardware compatibility

4-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for simoneschiavoi/VisionPsy-Nano-DomCal-Slim

Base model

lusxvr/nanoVLM-460M-8k

Finetuned

qvac/VisionPsy-Nano-460M

Quantized

(4)

this model