Instructions to use Snapkitty/hilbert-4b-baseline-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use Snapkitty/hilbert-4b-baseline-GGUF 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 Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M

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 Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M

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 Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M

Use Docker

docker model run hf.co/Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use Snapkitty/hilbert-4b-baseline-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Snapkitty/hilbert-4b-baseline-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Snapkitty/hilbert-4b-baseline-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M

Ollama
How to use Snapkitty/hilbert-4b-baseline-GGUF with Ollama:
```
ollama run hf.co/Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M
```
Unsloth Desktop

How to use Snapkitty/hilbert-4b-baseline-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Snapkitty/hilbert-4b-baseline-GGUF: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:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Snapkitty/hilbert-4b-baseline-GGUF with Docker Model Runner:
```
docker model run hf.co/Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M
```

Lemonade

How to use Snapkitty/hilbert-4b-baseline-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.hilbert-4b-baseline-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use Snapkitty/hilbert-4b-baseline-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Snapkitty/hilbert-4b-baseline-GGUF: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://127.0.0.1:8080/v1
hermes config set model.default Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Snapkitty/hilbert-4b-baseline-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Snapkitty/hilbert-4b-baseline-GGUF: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://127.0.0.1:8080/v1 \
  --custom-model-id "Snapkitty/hilbert-4b-baseline-GGUF: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"

Hilbert 4B Baseline · GGUF

💬 Feedback wanted. Ran it? Share your hardware, speed and decision-agreement numbers, or report a mismatch. Open a discussion. Part of our flagship releases.

GGUF quantizations of the frozen 4B decision model behind Hilbert: Qwen/Qwen3.5-4B at the pinned revision 851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a, measured the way Hilbert measures quantization: decision agreement with the BF16 reference, row by row.

Hilbert treats AI workloads as typed decisions. Given a state, a question and 2 to 16 declared options, the model returns a probability for each option from one forward pass. Code, benchmarks and the full row-level evidence: github.com/SNAPKITTYAGENT9NOVA/SemIf-OpenJev (see results/gguf/).

This is the unpruned baseline. The pruned and distilled Hilbert 4B is a separate, upcoming release; these files are its reference point.

Files

File	Quant	Size	SHA-256
`hilbert-4b-baseline.Q8_0.gguf`	Q8_0 (8.51 bpw)	4.61 GB	`4c9aff9532cdfba6e27f0f6c102f6086cb79bca4ef779e1ac17d608dc5a0beab`
`hilbert-4b-baseline.Q4_K_M.gguf`	Q4_K_M (5.13 bpw)	2.78 GB	`e84b5e30c8ac4636e20c11685c1d7d269cbd1b46cc564daa95465d636045dfaa`

Produced from one BF16 conversion with the llama.cpp sources vendored in llama-cpp-python==0.3.35 (text model only, no importance matrix). The exact steps are in conversion-*.json.

Decision quality vs BF16

Scored on CPU with the repository's llama.cpp backend on the authored144 and perturbations108 suites. Prompts matched the BF16 run exactly on every row.

Quant	Suite	Agreement with BF16	Flips	Balanced accuracy	Δ vs BF16 (95% CI)
Q8_0	authored144	99.31%	1 / 144	80.67%	−0.65 pp [−2.22, 0.00]
Q8_0	perturbations108	98.15%	2 / 108	76.58%	−1.41 pp [−3.67, 0.00]
Q4_K_M	authored144	92.36%	11 / 144	84.42%	+3.10 pp [+0.04, +6.60]
Q4_K_M	perturbations108	91.67%	9 / 108	76.08%	−1.90 pp [−6.30, +2.99]

BF16 reference: 81.32% (authored144), 77.99% (perturbations108). Q8_0 is effectively lossless for decisions. Q4_K_M is the compact choice and moves about 8% of decisions; its higher authored144 score reflects borderline rows flipping, not a real gain.

Use it

Decision scoring (the Hilbert way)

git clone https://github.com/SNAPKITTYAGENT9NOVA/SemIf-OpenJev && cd SemIf-OpenJev
pip install -e '.[llamacpp]'
huggingface-cli download Snapkitty/hilbert-4b-baseline-GGUF hilbert-4b-baseline.Q8_0.gguf --local-dir .
semif-score --backend llamacpp --mode direct --gguf hilbert-4b-baseline.Q8_0.gguf \
  --model Qwen/Qwen3.5-4B --revision 851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a \
  --input examples/decisions.jsonl --output decisions.out.jsonl

Chat / generation

llama-cli -hf Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M -cnv
ollama run hf.co/Snapkitty/hilbert-4b-baseline-GGUF:Q4_K_M

Verify the download

sha256sum -c SHA256SUMS

Citation

@misc{snapkitty_hilbert_4b_baseline_gguf_2026,
  title        = {Hilbert 4B Baseline GGUF: Decision-Agreement Evaluation of Quantized Qwen3.5-4B},
  author       = {{Snapkitty Collective LLC}},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/Snapkitty/hilbert-4b-baseline-GGUF}}
}

Please also cite the base model, Qwen3.5-4B, and the SemIf decision-native method it builds on.

License

The weights are a quantization of Qwen3.5-4B and are distributed under the Apache License 2.0, the base model's license. See LICENSE.

Downloads last month: 162

GGUF

Model size

4B params

Architecture

qwen35

Hardware compatibility

4-bit

8-bit

Model tree for Snapkitty/hilbert-4b-baseline-GGUF

Base model

Qwen/Qwen3.5-4B-Base

Finetuned

Qwen/Qwen3.5-4B

Quantized

(546)

this model

Space using Snapkitty/hilbert-4b-baseline-GGUF 1

Collection including Snapkitty/hilbert-4b-baseline-GGUF

Flagship releases · October 2026

Collection

Start here: our most-used releases, each with evidence you can check. Feedback welcome in each repo's Discussions. • 7 items • Updated 7 days ago