sleepy/llama.cpp
1
0
Fork 0
Code Issues 16 Pull Requests Actions 136 Packages Projects Releases Wiki Activity

model : add LFM2-ColBert-350M (#18607)

Browse Source 
* model : add LFM2-ColBert-350M

* llama_model_n_embd_out() - returns `hparams.n_embd_out` if set and fallbacks to `hparams.n_embd`
This commit is contained in:
Tarek Dakhran
2026-01-05 19:52:56 +01:00
committed by GitHub
parent df17a4c94f
commit 73d284a250
16 changed files with 118 additions and 60 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-graph.cpp
+7 -3
View File
@@ -2071,14 +2071,18 @@ llm_graph_input_mem_hybrid * llm_graph_context::build_inp_mem_hybrid() const {
void llm_graph_context::build_dense_out(
ggml_tensor * dense_2,
ggml_tensor * dense_3) const {
if (!cparams.embeddings || dense_2 == nullptr || dense_3 == nullptr) {
if (!cparams.embeddings || !(dense_2 || dense_3)) {
return;
}
ggml_tensor * cur = res->t_embd_pooled != nullptr ? res->t_embd_pooled : res->t_embd;
GGML_ASSERT(cur != nullptr && "missing t_embd_pooled/t_embd");
cur = ggml_mul_mat(ctx0, dense_2, cur);
cur = ggml_mul_mat(ctx0, dense_3, cur);
if (dense_2) {
cur = ggml_mul_mat(ctx0, dense_2, cur);
}
if (dense_3) {
cur = ggml_mul_mat(ctx0, dense_3, cur);
}
cb(cur, "result_embd_pooled", -1);
res->t_embd_pooled = cur;
ggml_build_forward_expand(gf, cur);