Instructions to use toolathlonEval/Northstar-Mini-EvalPack with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use toolathlonEval/Northstar-Mini-EvalPack with Transformers:

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

pipe = pipeline("text-generation", model="toolathlonEval/Northstar-Mini-EvalPack")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("toolathlonEval/Northstar-Mini-EvalPack")
model = AutoModelForCausalLM.from_pretrained("toolathlonEval/Northstar-Mini-EvalPack", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use toolathlonEval/Northstar-Mini-EvalPack with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "toolathlonEval/Northstar-Mini-EvalPack"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "toolathlonEval/Northstar-Mini-EvalPack",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/toolathlonEval/Northstar-Mini-EvalPack

SGLang

How to use toolathlonEval/Northstar-Mini-EvalPack 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 "toolathlonEval/Northstar-Mini-EvalPack" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "toolathlonEval/Northstar-Mini-EvalPack",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "toolathlonEval/Northstar-Mini-EvalPack" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "toolathlonEval/Northstar-Mini-EvalPack",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use toolathlonEval/Northstar-Mini-EvalPack with Docker Model Runner:
```
docker model run hf.co/toolathlonEval/Northstar-Mini-EvalPack
```

Northstar Mini

Northstar Mini is a compact instruction-tuned language model designed for local assistants, structured generation, and retrieval-grounded responses. This evaluation pack documents the release candidate selected after the Aurora training run.

Training Snapshot

The candidates share the same tokenizer and architecture. They differ only by training progress, so the bundled evaluation suite can be used to select the best balanced checkpoint.

Evaluation Results

All values are normalized scores where higher is better.

Category	Benchmark	Atlas-Base	Beacon-Chat	Northstar Mini
Reasoning	Math Reasoning	0.612	0.645	0.684
Reasoning	Logical Reasoning	0.774	0.803	0.821
Understanding	Reading Comprehension	0.699	0.721	0.744
Understanding	Question Answering	0.654	0.688	0.713
Generation	Code Generation	0.603	0.671	0.702
Generation	Dialogue Quality	0.627	0.662	0.681
Generation	Summarization	0.718	0.751	0.769
Multilingual	Translation	0.756	0.794	0.812
Knowledge	Knowledge Retrieval	0.663	0.701	0.726
Alignment	Instruction Following	0.711	0.759	0.781
Alignment	Safety Alignment	0.724	0.749	0.758
Reliability	Factual Consistency	0.681	0.718	0.735

Usage

Use the standard Transformers causal language modeling interface. For deterministic extraction tasks, start with temperature 0.2; for conversational tasks, start with temperature 0.7.

License

Northstar Mini is released under the Apache License 2.0.

Contact

Questions about this evaluation pack can be sent to models@northstar-labs.example.

Downloads last month: 12