sleepy/llama.cpp
1
0
Fork 0
Code Issues 8 Pull Requests 1 Actions 177 Packages Projects Releases Wiki Activity

ggml: add GATED_DELTA_NET op (#19504)

Browse Source 
* ggml: add GATED_DELTA_NET op

* remove the transpose

* add KDA

* add qwen35 dense

* llama : check for fused gated delta net backend support

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Aman Gupta
2026-03-07 15:41:10 +08:00
committed by GitHub
parent 6fce5c6a7d
commit c5a778891b
15 changed files with 627 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/models/qwen35moe.cpp
+1 -2
View File
@@ -332,8 +332,7 @@ ggml_tensor * llm_build_qwen35moe ::build_layer_attn_linear(
cb(k_conv, "k_conv_predelta", il);
cb(v_conv, "v_conv_predelta", il);
// Choose between build_delta_net_chunking, build_delta_net_recurrent, and build_delta_net_autoregressive based on n_tokens
std::pair<ggml_tensor *, ggml_tensor *> attn_out; // pair of (output, new_state)
std::pair<ggml_tensor *, ggml_tensor *> attn_out;
if (n_seq_tokens == 1) {
attn_out = build_delta_net_autoregressive(q_conv, k_conv, v_conv, gate, beta, state, il);
} else {