matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-14 09:47:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

chore: update mllama to use ollama engine (#10637)

Browse Source
This commit is contained in:
Michael Yang
2025-05-13 17:36:02 -07:00
committed by GitHub
parent 0478d440f0
commit 23125648b8
67 changed files with 785 additions and 4354 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
llama/llama.cpp/src/llama-graph.cpp vendored
View File

@@ -532,12 +532,6 @@ void llm_graph_input_attn_cross::set_input(const llama_ubatch * ubatch) {
}
}
void llm_graph_input_cross_attn_state::set_input(const llama_ubatch * ubatch) {
if (ubatch->embd) {
ggml_backend_tensor_set(cross_attn_state, ubatch->embd, 0, ggml_nbytes(cross_attn_state));
}
}
//
// llm_graph_context
//
@@ -1520,25 +1514,6 @@ llm_graph_input_attn_cross * llm_graph_context::build_attn_inp_cross() const {
return (llm_graph_input_attn_cross *) res->add_input(std::move(inp));
}
ggml_tensor * llm_graph_context::build_inp_cross_attn_state() const {
const int64_t n_embd = hparams.n_embd;
auto inp = std::make_unique<llm_graph_input_cross_attn_state>();
ggml_tensor * cur = nullptr;
inp->cross_attn_state = ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, n_embd, 1601, 4);
ggml_set_input(inp->cross_attn_state);
cur = inp->cross_attn_state;
cb(cur, "inp_cross_attn_state", -1);
res->add_input(std::move(inp));
return cur;
}
ggml_tensor * llm_graph_context::build_attn(
llm_graph_input_attn_cross * inp,
ggml_cgraph * gf,