simple_LLM

Some simple or fake implementations of LLM infra function

todo:

training:

parallel_implementation

inference (vllm features): clean_vllm

• chunked
• paged attention

cuda algorithm

• flash attention