open weights???
#43
by
alanchan808
- opened
It may be a stupid question. If the model is not open source, how HF Transformers use the open weights without knowing the model architecture? Please help me understand or point me to some good reads. Thanks a lot.
Well that's because it is open source. As stated on their website https://docs.mistral.ai/models/ , their models are Open-weight models. Mistral 7b & 8x7b models are both under Apache 2.0 license, I hope I did not misunderstand your question !
You can run model with safetensor format weights, because the network structure is already in the weights file. You just don't know the details of the training program and the dataset that trained the weights.