Views
No views yet
| Benchmark | Base Qwen3-4B-2507 | Our Model | Improvement |
|---|---|---|---|
| MMLU | 63.4% | 66.8% | +3.4% |
| GSM8K | 71.2% | 76.5% | +5.3% |
| HumanEval | 51.2% | 54.7% | +3.5% |
| HellaSwag | 80.8% | 82.3% | +1.5% |
| TruthfulQA | 51.7% | 58.2% | +6.5% |
<thinking> tags1from transformers import AutoModelForCausalLM, AutoTokenizer
2
3model = AutoModelForCausalLM.from_pretrained("Supra-Nexus/supra-nexus-o1-instruct")
4tokenizer = AutoTokenizer.from_pretrained("Supra-Nexus/supra-nexus-o1-instruct")
5
6# Example usage
7messages = [{"role": "user", "content": "Explain quantum computing"}]
8text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
9inputs = tokenizer([text], return_tensors="pt")
10
11outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7)
12response = tokenizer.decode(outputs[0], skip_special_tokens=True)1from vllm import LLM, SamplingParams
2
3llm = LLM(model="Supra-Nexus/supra-nexus-o1-instruct")
4sampling_params = SamplingParams(temperature=0.7, top_p=0.95, max_tokens=512)
5
6prompts = ["Explain the theory of relativity"]
7outputs = llm.generate(prompts, sampling_params)1@software{supra_nexus_o1_2025,
2 title = {Supra Nexus O1: Transparent Reasoning with Qwen3-4B-2507},
3 author = {Supra Foundation},
4 year = {2025},
5 month = {September},
6 url = {https://github.com/Supra-Nexus/o1},
7 note = {Based on Qwen3-4B-2507 (July 2025)}
8}