cont : bump max nodes to 8x model tensors

ggerganov · ggerganov · commit 5883f0149189 · 2025-07-18T14:12:10.000+03:00
diff --git a/src/llama-context.cpp b/src/llama-context.cpp
@@ -1312,7 +1312,7 @@ uint32_t llama_context::output_reserve(int32_t n_outputs) {
 //
 
 uint32_t llama_context::graph_max_nodes() const {
-    return std::max<uint32_t>(1024u, 6u*model.n_tensors());
+    return std::max<uint32_t>(1024u, 8u*model.n_tensors());
 }
 
 llm_graph_result * llama_context::get_gf_res_reserve() const {

Original file line number	Diff line number	Diff line change
`@@ -1312,7 +1312,7 @@ uint32_t llama_context::output_reserve(int32_t n_outputs) {`
`1312`	`1312`	`//`
`1313`	`1313`
`1314`	`1314`	`uint32_t llama_context::graph_max_nodes() const {`
`1315`		`- return std::max<uint32_t>(1024u, 6u*model.n_tensors());`
	`1315`	`+ return std::max<uint32_t>(1024u, 8u*model.n_tensors());`
`1316`	`1316`	`}`
`1317`	`1317`
`1318`	`1318`	`llm_graph_result * llama_context::get_gf_res_reserve() const {`