/llama_decomposed

Llama 2 model code with hack to reload it's code while keeping the model in memory

Primary LanguagePython

Watchers