1from transformers import AutoModelForCausalLM, AutoTokenizer
2device = "cuda" # the device to load the model onto
3
4model = AutoModelForCausalLM.from_pretrained(
5 "sii-research/InnoSpark-72B-0710",
6 torch_dtype="auto",
7 device_map="auto"
8)
9tokenizer = AutoTokenizer.from_pretrained("sii-research/InnoSpark-72B-0710")
10
11prompt = "Introduce yourself in detail."
12messages = [
13 {"role": "system", "content": "You are InnoSpark（启创）, created by Shanghai Innovation Institute （上海创智学院） and East China Normal University(华东师范大学). You are a helpful assistant."},
14 {"role": "user", "content": prompt}
15]
16text = tokenizer.apply_chat_template(
17 messages,
18 tokenize=False,
19 add_generation_prompt=True
20)
21model_inputs = tokenizer([text], return_tensors="pt").to(device)
22
23generated_ids = model.generate(
24 model_inputs.input_ids,
25 max_new_tokens=512
26)
27generated_ids = [
28 output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
29]
30
31response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
We recommend deploying our model using 4 A100 GPUs. You can run the vllm server-side with the following code in terminal:
1import requests
2import json
3
4def Innospark_stream(inputs,history):
5 url = 'http://loaclhost:6000/v1/chat/completions'
6
7 history+=[{"role": "user", "content": inputs},]
8
9 headers = {"User-Agent": "vLLM Client"}
10
11 pload = {
12 "model": "InnoSpark",
13 "stream": True,
14 "messages": history
15 }
16 response = requests.post(url,
17 headers=headers,
18 json=pload,
19 stream=True)
20
21 for chunk in response.iter_lines(chunk_size=1,
22 decode_unicode=False,
23 delimiter=b"\n"):
24 if chunk:
25 string_data = chunk.decode("utf-8")
26 try:
27 json_data = json.loads(string_data[6:])
28 delta_content = json_data["choices"][0]["delta"]["content"]
29 assistant_reply+=delta_content
30 yield delta_content
31 except KeyError as e:
32 delta_content = json_data["choices"][0]["delta"]["role"]
33 except json.JSONDecodeError as e:
34 history+=[{
35 "role": "assistant",
36 "content": assistant_reply,
37 "tool_calls": []
38 },]
39 delta_content='[DONE]'
40 assert '[DONE]'==chunk.decode("utf-8")[6:]
41
42inputs='hi'
43history=[]
44for response_text in Innospark_stream(inputs,history):
45 print(response_text,end='')
1. 📚 InnoSpark Model Series
- 4 models with different parameter scales: min(0.5B), turbo(7B), plus(72B) and their corresponding inference model R versions
2. 🔍 ELMES Evaluation System
- Education Language Model Evaluation System
- Automated evaluation system for educational tasks
- Helps continuously optimize large model capabilities in teaching scenarios
3. 🛠️ COCLP Data Cleaning Pipeline
- Corpus Cleansing Pipeline
- Visual node-based framework based on ComfyUI
- Supports OCR, audio/video transcription, format conversion, PII removal, text filtering, and other functions
- GitHub: COCLP
4. ⭐ HPC-RM Reward Model
- Helpful, Personalization, and Creativity Reward Model
- Provides scoring in three educational dimensions: helpfulness, personalization, and creativity
- Includes corresponding model scoring and human scoring datasets
1@misc{song2025cultivatinghelpfulpersonalizedcreative,
2 title={Cultivating Helpful, Personalized, and Creative AI Tutors: A Framework for Pedagogical Alignment using Reinforcement Learning},
3 author={Siyu Song and Wentao Liu and Ye Lu and Ruohua Zhang and Tao Liu and Jinze Lv and Xinyun Wang and Aimin Zhou and Fei Tan and Bo Jiang and Hao Hao},
4 year={2025},
5 eprint={2507.20335},
6 archivePrefix={arXiv},
7 primaryClass={cs.LG},
8 url={https://arxiv.org/abs/2507.20335},
9}
1@misc{wei2025elmesautomatedframeworkevaluating,
2 title={ELMES: An Automated Framework for Evaluating Large Language Models in Educational Scenarios},
3 author={Shou'ang Wei and Xinyun Wang and Shuzhen Bi and Jian Chen and Ruijia Li and Bo Jiang and Xin Lin and Min Zhang and Yu Song and BingDong Li and Aimin Zhou and Hao Hao},
4 year={2025},
5 eprint={2507.22947},
6 archivePrefix={arXiv},
7 primaryClass={cs.CY},
8 url={https://arxiv.org/abs/2507.22947},
9}
This project is jointly developed by East China Normal University and Shanghai Innovation Institute. The reward model was trained using the SiiRL training framework provided by Shanghai Innovation Institute.
Please refer to the relevant model pages for specific license information.
