Instructions to use nuroai/auden-12b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use nuroai/auden-12b 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 nuroai/auden-12b:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf nuroai/auden-12b:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf nuroai/auden-12b:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf nuroai/auden-12b: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 nuroai/auden-12b:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf nuroai/auden-12b: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 nuroai/auden-12b:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf nuroai/auden-12b:Q4_K_M

Use Docker

docker model run hf.co/nuroai/auden-12b:Q4_K_M

LM Studio
Jan

vLLM

How to use nuroai/auden-12b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "nuroai/auden-12b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "nuroai/auden-12b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/nuroai/auden-12b:Q4_K_M

Ollama
How to use nuroai/auden-12b with Ollama:
```
ollama run hf.co/nuroai/auden-12b:Q4_K_M
```
Unsloth Desktop

How to use nuroai/auden-12b with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf nuroai/auden-12b: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": "nuroai/auden-12b:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use nuroai/auden-12b with Docker Model Runner:
```
docker model run hf.co/nuroai/auden-12b:Q4_K_M
```

Lemonade

How to use nuroai/auden-12b with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull nuroai/auden-12b:Q4_K_M

Run and chat with the model

lemonade run user.auden-12b-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use nuroai/auden-12b with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf nuroai/auden-12b: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 nuroai/auden-12b:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use nuroai/auden-12b with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf nuroai/auden-12b: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 "nuroai/auden-12b: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"

auden-12b / repro_commands.md

akhilponnada

Add repro_commands.md

754770b verified 3 months ago

preview code

Raw

History Blame Contribute Delete

3.03 kB

Repro Commands — Auden v1 vs stock Gemma 4 12B

All published numbers: llama.cpp commit 76da2450a4f2cc9ce6c7fc8229e25dc0a4b41e5d, evaluator auden-docbench-v1-normalized (built into the project's scripts/eval_auden_docbench.py), OCR-text input, Q4_K_M quants. Commands below assume the project repo root as working directory; the server runs locally (127.0.0.1) — substitute <YOUR_SERVER> if you serve remotely.

1. Serve the model (one server per column)

# Auden v1 (this repo)
llama-server \
  -m q4_k_m/auden-12b-Q4_K_M.gguf \
  --mmproj q4_k_m/mmproj-BF16.gguf \
  --jinja -ngl 999 --host 127.0.0.1 --port 8091

# Stock Gemma 4 12B baseline (public unsloth/gemma-4-12b-it-GGUF)
llama-server \
  -m gemma-4-12b-it-Q4_K_M.gguf \
  --mmproj mmproj-BF16.gguf \
  --jinja -ngl 999 --host 127.0.0.1 --port 8093

Sanity gate before any eval: one POST /v1/chat/completions with the Auden tool schemas and a small receipt OCR text must return finish_reason: tool_calls with an extract_fields call. Wait for /health = 200 first; never score against a still-loading server.

2. Run the three evals (swap `--server-url`/`--out-dir` per model)

# a. CORD test 100 (real receipts, OCR text; CORD-v2 test split, CC-BY-4.0)
python scripts/eval_auden_docbench.py \
  --manifest data/real/cord/manifest.jsonl \
  --corpus-root data/real/cord \
  --server-url http://127.0.0.1:<PORT>/v1/chat/completions \
  --model-id <MODEL_ID> \
  --out-dir outputs/eval/<RUN>/<MODEL>_cord_ocr_norm \
  --input-mode ocr --limit 100

# b. Abstention 120
python scripts/eval_auden_docbench.py \
  --manifest data/eval/abstention_v1/manifest.jsonl \
  --corpus-root data/eval/abstention_v1 \
  --server-url http://127.0.0.1:<PORT>/v1/chat/completions \
  --model-id <MODEL_ID> \
  --out-dir outputs/eval/<RUN>/<MODEL>_abstention_norm \
  --input-mode ocr

# c. Synthetic 500
python scripts/eval_auden_docbench.py \
  --manifest data/eval/synthetic_v1_500/manifest.jsonl \
  --corpus-root data/eval/synthetic_v1_500 \
  --server-url http://127.0.0.1:<PORT>/v1/chat/completions \
  --model-id <MODEL_ID> \
  --out-dir outputs/eval/<RUN>/<MODEL>_synthetic_500_norm \
  --input-mode ocr

Note: the eval script exits non-zero when scores fall below its pass thresholds (expected for the stock baseline); the report and raw outputs are still written.

3. Failure analyzer (CORD output)

python scripts/analyze_cord_failures.py \
  --raw outputs/eval/<RUN>/<MODEL>_cord_ocr_norm/raw_outputs.jsonl \
  --out-dir outputs/eval/<RUN>/<MODEL>_cord_ocr_norm/failure_analysis

4. Provenance

Auden v1 numbers: phase5_rerun eval set; stock-Gemma baseline: base_phase5 eval set. Reports and raw outputs are archived in the project's private artifact store along with memory.md/docs/results.md (full run history).
v0 (historical comparison column) is reproducible from its preserved LoRA adapter (r=16, alpha=32) by re-merging onto google/gemma-4-12B-it and re-quantizing with llama.cpp.

Repro Commands — Auden v1 vs stock Gemma 4 12B

1. Serve the model (one server per column)

2. Run the three evals (swap --server-url/--out-dir per model)

3. Failure analyzer (CORD output)

4. Provenance

2. Run the three evals (swap `--server-url`/`--out-dir` per model)