叠加权重时超出python迭代范围

bibibabibo26 commented 4 months ago

你好，在我合成llama-base和你的roi-delta权重时报错，可以告诉我是什么原因怎么解决吗？ RecursionError: maximum recursion depth exceeded while calling a Python object

jshilong commented 4 months ago

https://huggingface.co/shilongz/debug 这个是一个merge好，原本用于demo的checkpoint

bibibabibo26 commented 4 months ago

https://huggingface.co/shilongz/debug 这个是一个merge好，原本用于demo的checkpoint

谢谢！

bibibabibo26 commented 4 months ago

https://huggingface.co/shilongz/debug 这个是一个merge好，原本用于demo的checkpoint 你好，在我运行app.py时，调试到加载tokenizer时报错RecursionError: maximum recursion depth exceeded in comparison。我融合权重文件时也是调试到delta_tokenizer = AutoTokenizer.from_pretrained(delta_path)报错RecursionError，请问项目环境的transformers版本是多少？或者可能是其他什么原因造成错误？

def build_model(self, model_name): ######################## base model define ######################## print('Start loading model...') disable_torch_init() model_name = os.path.expanduser(model_name) self.tokenizer = AutoTokenizer.from_pretrained(model_name) from gpt4roi.models.spi_llava import SPILlavaMPTForCausalLM

    # TODO add detector for normal conversation
    self.model = SPILlavaMPTForCausalLM.from_pretrained(
        model_name,
        low_cpu_mem_usage=True,
        torch_dtype=torch.float16,
        use_cache=True,
    ).cuda()

jshilong / GPT4RoI

叠加权重时超出python迭代范围 #46