batch : rework llama_batch_allocr (#14153)

* batch : rework llama_batch_allocr ggml-ci * cont : move validation inside class ggml-ci * cont : move output counting to class ggml-ci * cont : minor ggml-ci * batch : add TODOs ggml-ci
2025-06-13 13:47:55 +03:00 · 2025-06-13 13:47:55 +03:00 · 60c666347b
commit 60c666347b
parent b7cc7745e3
7 changed files with 162 additions and 106 deletions
--- a/src/llama-kv-cache-unified.cpp
+++ b/src/llama-kv-cache-unified.cpp
@ -674,6 +674,7 @@ void llama_kv_cache_unified::apply_ubatch(uint32_t head_cur, const llama_ubatch

            cells.pos_set(head_cur + idx, ubatch.pos[idx]);

+            // TODO: fix indexing [UBATCH_IDX]
            for (int32_t i = 0; i < ubatch.n_seq_id[s]; i++) {
                cells.seq_add(head_cur + idx, ubatch.seq_id[s][i]);
            }