Skip to main content

Module kv_cache_append

Module kv_cache_append 

Source
Expand description

kv_cache_append_head_major op-diff harness — see crate::op_diff.

Appends new_tokens head-major K/V vectors into a pre-filled cache at position cache_len. Pure data movement (exact); the compared output is [cache_k, cache_v] after the append.

Structs§

KvCacheAppendOp