Views
No views yet
| Model Name | Context Length | Hugging Face Link | ModelScope Link | Notes |
|---|---|---|---|---|
| JT-Math-8B-Base | 32K | Link | Link | The foundational base model. Ideal for custom fine-tuning. |
| Model | GSM8K | Math | CMath (zh) | Average |
|---|---|---|---|---|
| Qwen2.5-Base-32B | 92.8 | 57.7 | 85.4 | 78.6 |
| Llama-3.1-Base-405B | 89.0 | 53.8 | 77.4 | 73.4 |
| DeepSeek-Math-Base-7B | 64.2 | 36.2 | 71.7 | 57.4 |
| DeepSeek-Coder-V2-Lite-Base | 68.3 | 38.1 | 77.8 | 61.4 |
| Qwen2.5-Math-7B | 91.6 | 55.4 | 85.0 | 77.3 |
| JT-Math-8B-Base | 87.5 | 60.1 | 90.2 | 79.2 |
JT-Math-8B-Base model.1from transformers import AutoModelForCausalLM, AutoTokenizer
2
3model_name = "JT-LM/JT-Math-8B-Base"
4
5tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
6model = AutoModelForCausalLM.from_pretrained(
7 model_name,
8 torch_dtype="auto",
9 device_map="auto",
10 trust_remote_code=True,
11)
12
13prompt = "Janet’s ducks lay 16 eggs per day. She eats three for breakfast every morning and bakes muffins for her friends every day with four. She sells the remainder at the farmers' market daily for $2 per fresh duck egg. How much in dollars does she make every day at the farmers' market?"
14text = f"Question:\n{prompt}\nAnswer:\n"
15model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
16
17gen_kwargs = {
18 "do_sample": False,
19 "max_new_tokens": 8192,
20}
21generated_ids = model.generate(
22 **model_inputs,
23 **gen_kwargs
24)
25output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()
26
27response = tokenizer.decode(output_ids, skip_special_tokens=True)
28print("response:", response)1@article{jiutian-math2025,
2 title={JIUTIAN MATH: A MULTI-STAGE FRAMEWORK FOR ADVANCED MATHEMATICAL REASONING IN LARGE LANGUAGE MODELS},
3 author={Yifan Hao, Fangning Chao, Yaqian Hao, Zhaojun Cui, Huan Bai, Haiyu Zhang, Yankai Liu, Chao Deng, Junlan Feng},
4 journal={arXiv:2507.19748},
5 year={2025}
6}