Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

graph : utilize ggml_build_forward_select() to avoid reallocations (#18898)

Browse Source 
* graph : avoid branches between embedding and token inputs

* models : make deepstack graphs (e.g. Qwen3 VL) have constant topology

* ci : enable -DGGML_SCHED_NO_REALLOC=ON for server CI

* cont : pad token embeddings to n_embd_inp
This commit is contained in:
Georgi Gerganov
2026-01-23 18:22:34 +02:00
committed by GitHub
parent cb6caca191
commit 557515be1e
7 changed files with 69 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/models/gemma3n-iswa.cpp
+2 -2
View File
@@ -245,12 +245,12 @@ ggml_tensor * llm_build_gemma3n_iswa::view_2d_slice(ggml_tensor * x, int idx) {
// equivalent to get_per_layer_inputs() in python code
// output shape: [n_embd_altup, n_layer, n_tokens]
ggml_tensor * llm_build_gemma3n_iswa::get_per_layer_inputs() {
auto inp = std::make_unique<llm_graph_input_embd>();
auto inp = std::make_unique<llm_graph_input_embd>(n_embd);
ggml_tensor * inp_per_layer;
if (ubatch.token) {
inp->tokens = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, ubatch.n_tokens);
ggml_set_input(inp->tokens);
res->t_tokens = inp->tokens;
res->t_inp_tokens = inp->tokens;
inp_per_layer = ggml_get_rows(ctx0, model.tok_embd_per_layer, inp->tokens);
inp_per_layer = ggml_reshape_3d(ctx0, inp_per_layer, n_embd_altup, n_layer, n_tokens);
inp_per_layer = ggml_scale(ctx0, inp_per_layer, sqrtf((float) n_embd_altup));