Instructions to use luminousresearch/L0-PolyCore-4B-Base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use luminousresearch/L0-PolyCore-4B-Base with Transformers:

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

pipe = pipeline("text-generation", model="luminousresearch/L0-PolyCore-4B-Base")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("luminousresearch/L0-PolyCore-4B-Base")
model = AutoModelForCausalLM.from_pretrained("luminousresearch/L0-PolyCore-4B-Base")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use luminousresearch/L0-PolyCore-4B-Base with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "luminousresearch/L0-PolyCore-4B-Base"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "luminousresearch/L0-PolyCore-4B-Base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/luminousresearch/L0-PolyCore-4B-Base

SGLang

How to use luminousresearch/L0-PolyCore-4B-Base 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 "luminousresearch/L0-PolyCore-4B-Base" \
    --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": "luminousresearch/L0-PolyCore-4B-Base",
		"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 "luminousresearch/L0-PolyCore-4B-Base" \
        --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": "luminousresearch/L0-PolyCore-4B-Base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use luminousresearch/L0-PolyCore-4B-Base with Docker Model Runner:
```
docker model run hf.co/luminousresearch/L0-PolyCore-4B-Base
```

L0-PolyCore-4B-Base

File size: 1,350 Bytes

de560ca

---
license: other
language:
- en
base_model:
- Qwen/Qwen3-4B-Base
base_model_relation: finetune
tags:
- code
- c
- clang
- cpp
- c++
- qlora
- cpt
library_name: transformers
pipeline_tag: text-generation
---
## Training Data

This model was trained on a dataset of curated C/C++ code from multiple licenses (GPL-2.0, Apache-2.0, MIT, public domain, and some source-available licenses, etc.).
The original authors are not affiliated with or responsible for this model.

## Base Model

Base model: [Qwen/Qwen3-4B-Base](https://huggingface.co/Qwen/Qwen3-4B-Base)

## Fine-tuning Method

- Adapter: QLoRA
- Method: CPT
- Precision: trained with 4-bit base weights + BF16 compute, then merged to safetensors

## Training Details

- Training time: ~74 hours
- Hardware: 1x NVIDIA RTX 5060 Ti

## Notes

- This is an **L0 base model**, it is not instruction-tuned and may be more verbose with strict formatting request compared to an instruct model.
- Recommended usage is raw code continuation, or pairing with an external template strategy.

## Intended use

- Code generation for C/C++
- Fast code completion
- Examples and prototyping

## Constraints

- May produce incorrect code
- May reproduce identifiable upstream code fragments (including license headers) when prompted.
- Verify outputs, especially for memory safety and security-sensitive code.