nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 22 Packages Projects Releases Wiki Activity

graph : fix graph reuse logic when n_pos_per_embd > 1 (#18566)

Browse Source
This commit is contained in:
Georgi Gerganov
2026-01-03 23:59:06 +02:00
committed by GitHub
parent e57f52334b
commit c69c7ebc90
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-graph.cpp
+2 -2
View File
@@ -32,7 +32,7 @@ bool llm_graph_input_embd::can_reuse(const llm_graph_params & params) {
bool res = true; bool res = true;
res &= (!tokens && !params.ubatch.token) || (tokens && tokens->ne[0] == params.ubatch.n_tokens); res &= (!tokens && !params.ubatch.token) || (tokens && tokens->ne[0] == params.ubatch.n_tokens);
res &= (!embd && !params.ubatch.embd) || (embd && embd->ne[0] == params.ubatch.n_tokens); res &= (!embd && !params.ubatch.embd) || (embd && embd->ne[1] == params.ubatch.n_tokens);
return res; return res;
} }
@@ -62,7 +62,7 @@ void llm_graph_input_pos::set_input(const llama_ubatch * ubatch) {
bool llm_graph_input_pos::can_reuse(const llm_graph_params & params) { bool llm_graph_input_pos::can_reuse(const llm_graph_params & params) {
bool res = true; bool res = true;
res &= pos->ne[0] == params.ubatch.n_tokens; res &= pos->ne[0] == params.ubatch.n_tokens*n_pos_per_embd;
return res; return res;
} }