model: qwen4exp: reduce number of graph splits (#27880)

2026-09-19 17:24:57 +02:00 · 2026-08-28 19:24:46 +02:00
parent b387ddfd84
commit 6fe7498016
2 changed files with 27 additions and 10 deletions
@@ -2360,9 +2360,12 @@ struct llama_model_qwen4exp : public llama_model_base {
                        int64_t   channels,
                            int   il);

+        ggml_tensor * build_inp_ple(
+  const llama_memory_hybrid_idx_context * mctx_hyb);
+
        ggml_tensor * build_ple(
             llm_graph_input_rs * inp,
-  const llama_memory_hybrid_idx_context * mctx_hyb,
+                    ggml_tensor * emb,
                    ggml_tensor * hidden,
                            int   il);