zjunlp/OceanGPT-14B-v0.1

💡 Model description

This repo contains a large language model (OceanGPT) for ocean science tasks trained with KnowLM. It should be noted that the OceanGPT is constantly being updated, so the current model is not the final version.

OceanGPT-14B is based on Qwen1.5-14B and trained on a bilingual dataset in Chinese and English.

🔍 Intended uses

You can download the model to generate responses or contact the email for the online test demo.

🛠️ How to use OceanGPT

We wil provide several examples soon and you can modify the input according to your needs.

from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda" # the device to load the model onto

model = AutoModelForCausalLM.from_pretrained(
    "zjunlp/OceanGPT-14B-v0.1",
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("zjunlp/OceanGPT-14B-v0.1")

prompt = "Which is the largest ocean in the world?"
messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(device)

generated_ids = model.generate(
    model_inputs.input_ids,
    max_new_tokens=512
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

🛠️ How to evaluate your model in OceanBench

We wil provide several examples soon and you can modify the input according to your needs.

*Note: We are conducting the final checks on OceanBench and will be uploading it to Hugging Face soon.

>>> from datasets import load_dataset

>>> dataset = load_dataset("zjunlp/OceanBench")

📚 How to cite

@article{bi2023oceangpt,
  title={OceanGPT: A Large Language Model for Ocean Science Tasks},
  author={Bi, Zhen and Zhang, Ningyu and Xue, Yida and Ou, Yixin and Ji, Daxiong and Zheng, Guozhou and Chen, Huajun},
  journal={arXiv preprint arXiv:2310.02031},
  year={2023}
}