language:

• en
• zh library_name: transformers tags:
• 长文本处理
• chatglm
• llama datasets:
• THUDM/LongWriter-6k pipeline_tag: text-generation

LongWriter-glm4-9b

🤗 [LongWriter数据集] • 💻 [Github仓库] • 📃 [LongWriter论文]

LongWriter-glm4-9b基于glm-4-9b训练而成，能够一次性生成10,000字以上的文本。

运行环境：与glm-4-9b-chat相同（需transformers>=4.43.0）。

模型部署的简单示例：

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
tokenizer = AutoTokenizer.from_pretrained("THUDM/LongWriter-glm4-9b", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("THUDM/LongWriter-glm4-9b", torch_dtype=torch.bfloat16, trust_remote_code=True, device_map="auto")
model = model.eval()
query = "撰写一篇10000字的中国旅游指南"
response, history = model.chat(tokenizer, query, history=[], max_new_tokens=32768, temperature=0.5)
print(response)

您也可以使用vllm部署模型，该框架可在一分钟内生成10,000字以上的文本。示例代码如下：

from vllm import LLM, SamplingParams
model = LLM(
    model= "THUDM/LongWriter-glm4-9b",
    dtype="auto",
    trust_remote_code=True,
    tensor_parallel_size=1,
    max_model_len=32768,
    gpu_memory_utilization=1,
)
tokenizer = model.get_tokenizer()
stop_token_ids = [tokenizer.eos_token_id, tokenizer.get_command("<|user|>"), tokenizer.get_command("<|observation|>")]
generation_params = SamplingParams(
    temperature=0.5,
    top_p=0.8,
    top_k=50,
    max_tokens=32768,
    repetition_penalty=1,
    stop_token_ids=stop_token_ids
)
query = "撰写一篇10000字的中国旅游指南"
input_ids = tokenizer.build_chat_input(query, history=[], role='user').input_ids[0].tolist()
outputs = model.generate(
    sampling_params=generation_params,
    prompt_token_ids=[input_ids],
)
output = outputs[0]
print(output.outputs[0].text)

许可证：glm-4-9b许可证

引用

如果您觉得我们的工作有帮助，请考虑引用LongWriter：

@article{bai2024longwriter,
  title={LongWriter: 从长文本LLM中释放10,000+字生成能力}, 
  author={白雨石 and 张家杰 and 吕昕 and 郑林志 and 朱思齐 and 侯磊 and 董宇霄 and 唐杰 and 李涓子},
  journal={arXiv预印本 arXiv:2408.07055},
  year={2024}
}