learning inference

this repository runs inference for Llama2-7B

todo: [x] kv caching [] continuous batching [] paging [] quantization [] spec decoding [] kernels + kernel fusion and more