model : register t_layer_inp for qwen3next (#25141)

* Fix input assignment in layer processing loop

Fix DFLASH for qwen-coder-next

* add line break

Added tensor for attention normalization in Qwen3 model.

diff --git a/src/models/qwen3next.cpp b/src/models/qwen3next.cpp
index 97200a44072ff5d4623650bc8f89d78dd3e1216d..09b66423d5a8250720bebf26cdf61a9f31deefd1 100644 (file)
--- a/src/models/qwen3next.cpp
+++ b/src/models/qwen3next.cpp
@@ -121,6 +121,8 @@ llama_model_qwen3next::graph::graph(const llama_model & model, const llm_graph_p
     ggml_tensor * inp_out_ids = build_inp_out_ids();
 
     for (int il = 0; il < n_layer; ++il) {
+        res->t_layer_inp[il] = inpL;
+
         ggml_tensor * inpSA = inpL;
 
         cur = build_norm(inpL, model.layers[il].attn_norm, nullptr, LLM_NORM_RMS, il);