Public/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-09-19 09:15:04 +02:00
Code Issues Packages Projects Releases Wiki Activity

spec: single device drafter should create meta backend wrapper

Browse Source
This commit is contained in:
Aman Gupta
2026-09-04 20:28:56 +08:00
parent 64a155d242
commit 477ecf18ef
1 changed files with 8 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/speculative.cpp
+8
View File
@@ -2472,6 +2472,14 @@ common_params common_base_params_to_speculative(const common_params & params) {
result.n_gpu_layers = params_spec.n_gpu_layers;
result.tensor_buft_overrides = params_spec.tensor_buft_overrides;
// a draft pinned to a single device doesn't need the meta wrapper an inherited -sm tensor would give it
// (the device list is null-terminated, so a single device means size 2)
const size_t n_devs = std::count_if(params_spec.devices.begin(), params_spec.devices.end(),
[](ggml_backend_dev_t d) { return d != nullptr; });
if (n_devs == 1) {
result.split_mode = LLAMA_SPLIT_MODE_LAYER;
}
if (params_spec.cpuparams.n_threads > 0) {
result.cpuparams.n_threads = params_spec.cpuparams.n_threads;
result.cpuparams_batch.n_threads = params_spec.cpuparams_batch.n_threads;