xFasterTransformer icon indicating copy to clipboard operation
xFasterTransformer copied to clipboard

Add env param KV_CACHE_LOCATION to control kv cache memory numanode location

Open a3213105 opened this issue 8 months ago • 2 comments

Usage: before you run instance export KV_CACHE_LOCATION=#memory_numa_node_id_you_want_to_use_for_kv_cache

by defaults, kv_cache location is the same as other parts of instance.

a3213105 avatar Jun 28 '24 04:06 a3213105