Instructions to use parallaxlabs/parallax-2-preview with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use parallaxlabs/parallax-2-preview 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 parallaxlabs/parallax-2-preview:Q2_K_L
# Run inference directly in the terminal:
llama cli -hf parallaxlabs/parallax-2-preview:Q2_K_L

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf parallaxlabs/parallax-2-preview:Q2_K_L
# Run inference directly in the terminal:
llama cli -hf parallaxlabs/parallax-2-preview:Q2_K_L

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 parallaxlabs/parallax-2-preview:Q2_K_L
# Run inference directly in the terminal:
./llama-cli -hf parallaxlabs/parallax-2-preview:Q2_K_L

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 parallaxlabs/parallax-2-preview:Q2_K_L
# Run inference directly in the terminal:
./build/bin/llama-cli -hf parallaxlabs/parallax-2-preview:Q2_K_L

Use Docker

docker model run hf.co/parallaxlabs/parallax-2-preview:Q2_K_L

LM Studio
Jan

vLLM

How to use parallaxlabs/parallax-2-preview with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "parallaxlabs/parallax-2-preview"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "parallaxlabs/parallax-2-preview",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/parallaxlabs/parallax-2-preview:Q2_K_L

Ollama
How to use parallaxlabs/parallax-2-preview with Ollama:
```
ollama run hf.co/parallaxlabs/parallax-2-preview:Q2_K_L
```
Unsloth Desktop

How to use parallaxlabs/parallax-2-preview with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf parallaxlabs/parallax-2-preview:Q2_K_L

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": "parallaxlabs/parallax-2-preview:Q2_K_L"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use parallaxlabs/parallax-2-preview with Docker Model Runner:
```
docker model run hf.co/parallaxlabs/parallax-2-preview:Q2_K_L
```

Lemonade

How to use parallaxlabs/parallax-2-preview with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull parallaxlabs/parallax-2-preview:Q2_K_L

Run and chat with the model

lemonade run user.parallax-2-preview-Q2_K_L

List all available models

lemonade list

Hermes Agent

How to use parallaxlabs/parallax-2-preview with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf parallaxlabs/parallax-2-preview:Q2_K_L

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 parallaxlabs/parallax-2-preview:Q2_K_L

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use parallaxlabs/parallax-2-preview with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf parallaxlabs/parallax-2-preview:Q2_K_L

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 "parallaxlabs/parallax-2-preview:Q2_K_L" \
  --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"

Parallax 2 Preview

Parallax 2 Preview is the first public look at Parallax Labs' upcoming Parallax-2 model. It's a 4B finetune of Qwen 3.5, trained specifically to reason through problems carefully, like a frontier model would.

Model Details

Model Description

Parallax-2-preview is a 4B parameter finetune designed to demonstrate structured reasoning at small scale. It uses the reasoning style of frontier models, while remaining fast enough to run on consumer hardware.

Developed by: Parallax Labs
Funded by: Parallax Labs
Shared by: Parallax Labs
Model type: Causal language model (LoRA finetune on Qwen 3.5-4B) & Coder
Language: English
License: Apache 2.0
Finetuned from model: Unsloth/Qwen3.5-4B

Model Sources

Repository: https://huggingface.co/parallaxlabs/parallax-2-preview
Demo: N/A

Uses

Direct Use

Parallax 2 Preview is intended for reasoning-heavy tasks where a short, direct answer isn't the goal.

Downstream Use

Suitable for further finetuning into reasoning-focused applications, education tools, or as a component in multi-model pipelines where structured reasoning is the required output format.

Out-of-Scope Use

Parallax 2 Preview is not intended for:

General-purpose chat or assistant behavior
Complex code generation or multi-file tasks. It can handle coding, but not complex coding.
Long-context tasks (As a Preview model, it was trained at a low context, so its answers can degrade as extended context is used.)
Tasks requiring fast, short responses (the model reasons before answering - see Latency below)
Production tooling where the identity of the model matters (see Known Limitations)

Bias, Risks, and Limitations

Verbose reasoning as default. Parallax-2-Preview reasons before answering, on every prompt. Simple questions that would normally return in 2 seconds may take 10–30 seconds of generation first.
Coding is degraded relative to base. The finetune traded some general code-generation ability for reasoning depth.
Identity override is incomplete without a system prompt. Without the recommended system prompt (below), the model may identify as its base model (Qwen). This is a known behavior of thin LoRA finetunes on strong base models.
Long-context performance is not guaranteed. Trained at 2048 tokens. The base model supports up to 262K via RoPE, but quality degrades significantly beyond ~4K.