diff --git a/common/common.cpp b/common/common.cpp
index 44bb76618..df3cba119 100644
--- a/common/common.cpp
+++ b/common/common.cpp
@@ -224,6 +224,7 @@ bool gpt_params_parse(int argc, char ** argv, gpt_params & params) {
                 break;
             }
             sparams.temp = std::stof(argv[i]);
+            sparams.temp = std::max(sparams.temp, 0.0f);
         } else if (arg == "--tfs") {
             if (++i >= argc) {
                 invalid_param = true;
diff --git a/common/sampling.cpp b/common/sampling.cpp
index 4d0084a3c..c4996c985 100644
--- a/common/sampling.cpp
+++ b/common/sampling.cpp
@@ -168,12 +168,12 @@ llama_token llama_sampling_sample(
     }
 
     if (temp < 0.0) {
-        // greedy sampling, no probs
-        id = llama_sample_token_greedy(ctx_main, &cur_p);
-    } else if (temp == 0.0) {
         // greedy sampling, with probs
         llama_sample_softmax(ctx_main, &cur_p);
         id = cur_p.data[0].id;
+    } else if (temp == 0.0) {
+        // greedy sampling, no probs
+        id = llama_sample_token_greedy(ctx_main, &cur_p);
     } else {
         if (mirostat == 1) {
             const int mirostat_m = 100;
diff --git a/examples/speculative/speculative.cpp b/examples/speculative/speculative.cpp
index 07799fcbd..5da020e25 100644
--- a/examples/speculative/speculative.cpp
+++ b/examples/speculative/speculative.cpp
@@ -118,7 +118,7 @@ int main(int argc, char ** argv) {
     std::vector<seq_draft> drafts(n_seq_dft);
 
     params.sparams.grammar.clear(); // the draft samplers will copy the target sampler's grammar
-    params.sparams.temp = 0.0f;
+    params.sparams.temp = -1.0f;    // force greedy sampling with probs for the draft model
 
     for (int s = 0; s < n_seq_dft; ++s) {
         drafts[s].ctx_sampling = llama_sampling_init(params.sparams);