turneram

Results 26 issues of turneram

Add llama2 with KV-cache example

Add llama2 with KV-cache to DLM

Compare llama2 performance with/without KV-cache

Improve performance of GQA sub-operators on GPU

Enable contrib ops for Llama2 in ORT and run end-to-end workflow

Debug end-to-end accuracy of Llama2

Add support for GQA num_heads > kv_num_heads

Add rocMLIR kv-cache attention support

2 comment

Refactor parse_multi_head_attention to use group_query_attention for janus-pro-1b

Run deepseek/Janus_pro_7B in migraphx