Skip to content

LLM inference in C/C++

C++ ggml

sampler: reduce the size of the probe (#29285)

Max Krasnyansky441df11f65ea
diff --git a/src/llama-sampler.cpp b/src/llama-sampler.cpp
index 61d28ad8a8..b797c50bef 100644
--- a/src/llama-sampler.cpp
+++ b/src/llama-sampler.cpp
@@ -644,7 +644,7 @@ static bool llama_sampler_backend_support(
return true;
}
- auto probe = llama_sampler_backend_probe_graph(smpl, 1024*1024, GGML_DEFAULT_GRAPH_SIZE, true);
+ auto probe = llama_sampler_backend_probe_graph(smpl, 128*1024, GGML_DEFAULT_GRAPH_SIZE, true);
for (int i = 0; i < ggml_graph_n_nodes(probe.gf); i++) {
struct ggml_tensor * op = ggml_graph_node(probe.gf, i);
@@ -764,7 +764,7 @@ static bool llama_sampler_chain_backend_init(
res = res && cur_prefix;
}
- auto probe = llama_sampler_backend_probe_graph(smpl, 1024*1024, GGML_DEFAULT_GRAPH_SIZE, false);
+ auto probe = llama_sampler_backend_probe_graph(smpl, 128*1024, GGML_DEFAULT_GRAPH_SIZE, false);
chain->n_nodes = llama_sampler_backend_probe_n_nodes(probe);
return res;