From: Kevin Wang <redacted>
Date: Mon, 8 Jul 2024 06:31:55 +0000 (-0400)
Subject: common : avoid unnecessary logits fetch (#8358)
X-Git-Tag: upstream/0.0.4488~1147
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=ffd00797d81ef7db1528b9e10adbdc333ade6495;p=pkg%2Fggml%2Fsources%2Fllama.cpp

common : avoid unnecessary logits fetch (#8358)
---

diff --git a/common/sampling.cpp b/common/sampling.cpp
index 9f332fe5..e9db2715 100644
--- a/common/sampling.cpp
+++ b/common/sampling.cpp
@@ -282,8 +282,6 @@ static llama_token llama_sampling_sample_impl(
         GGML_ASSERT(!original_logits.empty());
     }
     llama_token id = 0;
-    // Get a pointer to the logits
-    float * logits = llama_get_logits_ith(ctx_main, idx);
 
     if (temp < 0.0) {
         // greedy sampling, with probs
@@ -324,6 +322,9 @@ static llama_token llama_sampling_sample_impl(
     }
 
     if (ctx_sampling->grammar != NULL && !is_resampling) {
+        // Get a pointer to the logits
+        float * logits = llama_get_logits_ith(ctx_main, idx);
+
         // Create an array with a single token data element for the sampled id
         llama_token_data single_token_data = {id, logits[id], 0.0f};
         llama_token_data_array single_token_data_array = { &single_token_data, 1, false };