44670
44670
Hi, as we explained in the `README.md`, Chinese, English, Japanese, Korean, French, Germany are the languages that we supported.
> It should be come from llama.
base版本已经在README里说明了，是用50%数据微调的版本，模型需要进一步用垂直领域数据微调才可使用，不建议直接用于对话任务。 此外，请参考GitHub的说明和`examples/hello.py`，按照Prompt的格式使用，如果不遵守prompt格式，模型可能无法正常工作。
Thanks for your kind words. Our goal for the model is to adhere to the system prompt, while maintaining safety. For Falcon, we will be mainly focused on 180B in...
Thanks for pointing out, we may add 40B back into our consideration.
Hello, we have released a new version of the falcon-40b model: https://huggingface.co/OpenBuddy/openbuddy-falcon-40b-v16.1-4k . In this version, we used rope scaling to support 4k length. However, this model seems to be...
`llm = torch.nn.DataParallel(llm, device_ids=list(range(num_gpus)))` 感觉这句话是多余的，vllm会自动同步到多个卡上。
1048k也太夸张了。 我们有计划做合理范围内，有长文认知能力的模型。
有可能是库版本问题， pip install "transformers>=4.34.0"试试。 如果还不行的话，可以去modelscope repo的issue区域问问看。
3bit量化版的话需要3090*2 如果是全量部署，需要3090*8