Enhance device compatibility by adding MPS support and ensuring tensor operations respect device context

Handle LlamaFlashAttention2 import for compatibility with newer transformers versions

Update model path in config.json to match new repository location

Copying actual model file

Initial copy

initial commit