reduce kv cache from available memory
This commit is contained in:
+1
-1
@@ -914,7 +914,7 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
|
||||
dev_info_set = (struct device_info *)malloc(n_world * sizeof(struct device_info));
|
||||
dev_info_set[0] = dev_info;
|
||||
llama_gather_device_info(lctx, dev_info_set);
|
||||
device_print_props(dev_info_set, n_world, model);
|
||||
device_print_props(dev_info_set, n_world, model, cparams);
|
||||
} else {
|
||||
llama_send_device_info(lctx, &dev_info);
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user