llama-batch: add n_keep_tail in split_equal for recurrent models (#25278)

2026-07-08 15:55:19 +08:00
parent f296fdfbed
commit 230ea9d214
10 changed files with 102 additions and 38 deletions
@@ -206,7 +206,7 @@ llama_memory_context_ptr llama_kv_cache_iswa::init_batch(llama_batch_allocr & ba

        std::vector<llama_ubatch> ubatches;
        while (true) {
-            auto ubatch = balloc.split_equal(n_ubatch, !unified);
+            auto ubatch = balloc.split_equal(n_ubatch, !unified, 0);

            if (ubatch.n_tokens == 0) {
                break;