llama : remove KV cache defragmentation logic (#15473)

ggml-ci
2025-08-22 12:22:13 +03:00
parent ad5c975c2d
commit 9ebebef62f
16 changed files with 32 additions and 440 deletions
--- a/src/llama-cparams.h
+++ b/src/llama-cparams.h
@@ -24,7 +24,6 @@ struct llama_cparams {
    float yarn_attn_factor;
    float yarn_beta_fast;
    float yarn_beta_slow;
-    float defrag_thold;

    bool embeddings;
    bool causal_attn;