From: Georgi Gerganov <redacted>
Date: Fri, 2 May 2025 17:54:13 +0000 (+0300)
Subject: context : fix reorder logic (#13267)
X-Git-Tag: upstream/0.0.5318~51
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=a75cb30dc9e63488c3614e2d5a9fe2306eaf47cd;p=pkg%2Fggml%2Fsources%2Fllama.cpp

context : fix reorder logic (#13267)

ggml-ci
---

diff --git a/src/llama-context.cpp b/src/llama-context.cpp
index a88b9a5f..45591be9 100644
--- a/src/llama-context.cpp
+++ b/src/llama-context.cpp
@@ -1050,6 +1050,9 @@ int llama_context::decode(llama_batch & inp_batch) {
     // finalize the batch processing
     kv_guard.commit();
 
+    // set to total number of outputs in the batch, for use in llama_get_logits_ith
+    n_outputs = n_outputs_all;
+
     // set output mappings
     {
         bool sorted_output = true;
@@ -1103,9 +1106,6 @@ int llama_context::decode(llama_batch & inp_batch) {
         }
     }
 
-    // set to total number of outputs in the batch, for use in llama_get_logits_ith
-    n_outputs = n_outputs_all;
-
     // wait for the computation to finish (automatically done when obtaining the model output)
     //synchronize();