This repository contains the full weights and LoRA weights for Zh-MT-LLM v1.0 which fine-tuned with ChatGLM3-6b-base.
Zh-MT-LLM
Zheng He Maritime Large Language Model (Zh-MT-LLM) is a vertical domain maritime Large Language Model developed by the Intelligent Technology Laboratory of Dalian Maritime University for practitioners, trainers and students in the maritime field, providing questions and answers on maritime laws and regulations, maritime education and training, and questions and answers on maritime expertise.
Corresponding to the above three segments, our model has the following three main characteristics:
-
Maritime Laws and Regulations Q&A:
The model is trained on a wide range of maritime laws and regulations, providing consulting services for those in the maritime field.
-
Maritime Education and Training:
The model learns from maritime professional test questions, vocational examination syllabi, and high-quality crew common Q&A to provide training knowledge.
-
Maritime Expertise Q&A:
The model covers ship maintenance, safety management, port operations, maritime logistics, navigation technology, marine environmental protection, and scientific research to answer questions for maritime industry practitioners.
Zh-MT-SFT Dataset
The specific statistics of the dataset used for the above training are as follows:
| Services | Subtasks | Data sets | Data volume |
|---|
| Maritime Laws and Regulations Q&A | Maritime Legal Advice | CrimeKgAssitant | 18,279 |
| Zh-law-qa | 59,244 |
| The Court held | Zh-law-court | 2,684 |
| Sentence projections | Zh-law-predict | 3,004 |
| Maritime education and training | Maritime Education Counseling | Zh-edu-qa | 41,052 |
| Maritime Specialization Question Bank | Zh-edu-qb | 23,531 |
| Maritime Expertise Q&A | Ship Knowledge | Zh-mt-qa | 46,759 |
| Navigational Knowledge |
| Port knowledge |
| Marine knowledge |
| Generic Dialogue | | moss-003-sft-data | 300,000 |
| Total | 494,553 |
Code Usage
You can create a conversation using the Zh-MT-LLM model using the following codes:
1>>>from transformers import AutoTokenizer, AutoModel
2>>>tokenizer = AutoTokenizer.from_pretrained("ZhangFuXi/Zh-MT-LLM",trust_remote_code=True)
3>>>model = AutoModel.from_pretrained("ZhangFuXi/Zh-MT-LLM", trust_remote_code=True).half().cuda()
4>>>model = model.eval()
5>>>response, history = model.chat(tokenizer, "你好", history=[])
6>>>print(response)
Declaration
Due to factors such as the limitation of the number of model parameters and the degree of cleaning of the training data, the model open source in this project may have the following limitations:
-
Because it has not been harmlessly fine-tuned, it may result in discriminatory, harmful and unethical statements.
-
Lacking an accurate understanding of the real world, the model may produce hallucinatory responses that mislead the user.
-
The model's training data may contain biased data, and users should be cautious about potential bias in model responses.
-
Due to the limited number of model parameters, it may not be possible to cover all areas of knowledge, resulting in less accurate or complete responses on some topics.
-
When dealing with factual knowledge in a specific domain, models may provide incorrect answers due to insufficient information or misinterpretation, leading to misinformation or confusion.
Given the limitations of the above model, we request that the code, data, and model of this project not be used for socially harmful purposes and must follow the
MODEL_LICENCE of the base model. We are not responsible for any problems, risks, or adverse consequences arising from the use of Zh-MT-LLM.
Licenses
The use of the source code in this repository complies with the Apache 2.0 License.
