common : Disable backend sampling if reasoning budget is enabled (#21209)

diff --git a/common/sampling.cpp b/common/sampling.cpp
index 5259c5f3c6f58d49285869cafecd03112870763f..2f60be19432416a2243aa9460596b400c1f9025b 100644 (file)
--- a/common/sampling.cpp
+++ b/common/sampling.cpp
@@ -383,6 +383,12 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, st
         params.backend_sampling = false;
     }
 
+    if (rbudget && params.backend_sampling) {
+        LOG_WRN("%s: backend sampling is not compatible with reasoning budget, disabling\n", __func__);
+
+        params.backend_sampling = false;
+    }
+
     auto * result = new common_sampler {
         /* .params  = */ params,
         /* .grmr    = */ grmr,