[Question]: Add support for tinyllama 1.1 B model
Nick-infinity opened this issue · comments
Hello, I have added support for llm-export repo to convert tinyllama 1.1B model and then made changes to model template in mnn-llm to load it correct. This model crashes at runtime in mnn. It would be good If I @wangzhaode can guide me with whats fundamentally required to add new model support with hidden dim size 2048. Is only a hidden dim size of 4096 supported as of now?
You can inherit from the Llama2_7b class and modify the hidden_size_, layer_nums_, and key_value_shape_ attributes.