Instructions to use NIM-AI/NIM-1-3B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use NIM-AI/NIM-1-3B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="NIM-AI/NIM-1-3B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("NIM-AI/NIM-1-3B", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use NIM-AI/NIM-1-3B 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 NIM-AI/NIM-1-3B:Q8_0
# Run inference directly in the terminal:
llama cli -hf NIM-AI/NIM-1-3B:Q8_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf NIM-AI/NIM-1-3B:Q8_0
# Run inference directly in the terminal:
llama cli -hf NIM-AI/NIM-1-3B:Q8_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 NIM-AI/NIM-1-3B:Q8_0
# Run inference directly in the terminal:
./llama-cli -hf NIM-AI/NIM-1-3B:Q8_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 NIM-AI/NIM-1-3B:Q8_0
# Run inference directly in the terminal:
./build/bin/llama-cli -hf NIM-AI/NIM-1-3B:Q8_0

Use Docker

docker model run hf.co/NIM-AI/NIM-1-3B:Q8_0

LM Studio
Jan

vLLM

How to use NIM-AI/NIM-1-3B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "NIM-AI/NIM-1-3B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NIM-AI/NIM-1-3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/NIM-AI/NIM-1-3B:Q8_0

SGLang

How to use NIM-AI/NIM-1-3B with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "NIM-AI/NIM-1-3B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NIM-AI/NIM-1-3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "NIM-AI/NIM-1-3B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NIM-AI/NIM-1-3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use NIM-AI/NIM-1-3B with Ollama:
```
ollama run hf.co/NIM-AI/NIM-1-3B:Q8_0
```
Unsloth Desktop

How to use NIM-AI/NIM-1-3B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf NIM-AI/NIM-1-3B:Q8_0

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": "NIM-AI/NIM-1-3B:Q8_0"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use NIM-AI/NIM-1-3B with Docker Model Runner:
```
docker model run hf.co/NIM-AI/NIM-1-3B:Q8_0
```

Lemonade

How to use NIM-AI/NIM-1-3B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull NIM-AI/NIM-1-3B:Q8_0

Run and chat with the model

lemonade run user.NIM-1-3B-Q8_0

List all available models

lemonade list

Hermes Agent

How to use NIM-AI/NIM-1-3B with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf NIM-AI/NIM-1-3B:Q8_0

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 NIM-AI/NIM-1-3B:Q8_0

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use NIM-AI/NIM-1-3B with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf NIM-AI/NIM-1-3B:Q8_0

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 "NIM-AI/NIM-1-3B:Q8_0" \
  --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"

NIM-1 (3B)

NIM-1 is a high-efficiency 3-billion parameter local reasoning model engineered for rapid, deterministic task execution, code intelligence, and structured agentic workflows. Built to deliver flagship reasoning density within consumer hardware limits, NIM-1 runs completely offline with ultra-low latency.

Highlights

High-Density Reasoning: Tuned on high-signal chain-of-thought demonstrations to handle multi-step logic, Python algorithms, and complex instruction following.
Consumer-Ready Edge Inference: Consumes under 4 GB of VRAM in 8-bit quantization, delivering 60–90 tokens/sec on entry-level GPUs (such as RTX 3060/4060) or Apple Silicon.
Zero-Degradation Precision: Packaged with Q8_0 GGUF quantization for maximum output stability and numerical fidelity.
Autonomous Agent Ready: Built with native support for tool orchestration, structured schema outputs, and multi-turn conversational memory.

Quickstart with Ollama

Run NIM-1 instantly from Hugging Face:

ollama run hf.co/NIM-AI/NIM-1-3B:NIM-1-3B-Q8_0.gguf

Or build and run directly from the local repository:

ollama create nim-1 -f Modelfile
ollama run nim-1

Model Specifications

Parameter	Specification
Model Architecture	Dense Transformer (Decoder-only)
Total Parameters	3.09 Billion
Context Window	2,048 tokens (extensible to 32k)
Quantization Format	GGUF (`Q8_0`)
Inference Footprint	~3.4 GB VRAM / System Memory
Chat Template	ChatML format (`<

Architectural & Training Methodology

NIM-1 was trained using parameter-efficient fine-tuning (QLoRA) with custom Triton-accelerated backpropagation kernels:

Curated High-Signal Supervision: Trained against an curated dataset of algorithmic logic puzzles, step-by-step math derivations, and software architecture patterns.
Quantized Fine-Tuning: Trained using 4-bit base model weight caching, rank-16 target projection adapters (q, k, v, o, gate, up, down), and fused cross-entropy loss.
High-Fidelity Export: LoRA adapters merged back into 16-bit floating-point space before single-pass quantization to Q8_0 GGUF format to avoid quantization degradation.

Hardware Compatibility

Environment	Performance	VRAM / Memory
NVIDIA RTX 4060 (8 GB)	~70–90 tok/s	~3.4 GB
Apple Silicon (M-Series, 16 GB)	~60–80 tok/s	~3.6 GB Unified
Modern x86 CPU (AVX-512)	~18–28 tok/s	~4.2 GB RAM

Reproducing Training & Packaging

# Clone the repository
git clone [https://github.com/NIM-AI/NIM-1.git](https://github.com/NIM-AI/NIM-1.git)
cd NIM-1

# Install requirements
pip install -r requirements.txt

# Run the training pipeline
python scripts/train_student.py

# Export weights to GGUF format
python scripts/export_gguf.py

Downloads last month: 41

GGUF

Model size

3B params

Architecture

qwen2

Hardware compatibility

8-bit

Model tree for NIM-AI/NIM-1-3B

Base model

Qwen/Qwen2.5-3B

Finetuned

Qwen/Qwen2.5-3B-Instruct

Quantized

(298)

this model