modeling_llama_fastv_kvcache.py

[cool-xiang]

Thank you for your wonderful work! I noticed this code: modeling_llama_fastv_kvcache.py, This is for saving the full kv-caches of the visual token. Can this code run now? I see a few todos in the code line.

Thank you very much!

[chenllliang]

Hi, this part of code is discarded and not tested, we move to the huggingface's llava to implement kv-cache. For kv-cache, please refer to this comment for more details.