From 0c1df14b5f8d992805cb22d0b77b44092a18aeab Mon Sep 17 00:00:00 2001
From: Douglas Hanley <thesecretaryofwar@gmail.com>
Date: Sat, 12 Jul 2025 06:21:02 -0400
Subject: [PATCH] server : fix pooled embedding output (#14645)

---
 tools/server/server.cpp | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/tools/server/server.cpp b/tools/server/server.cpp
index 57b917f2f..d4dffb39c 100644
--- a/tools/server/server.cpp
+++ b/tools/server/server.cpp
@@ -2581,12 +2581,14 @@ struct server_context {
                 continue;
             }
 
-            const float * embd = llama_get_embeddings_seq(ctx, batch.seq_id[i][0]);
-            if (embd == NULL) {
+            const float * embd = nullptr;
+            if (llama_pooling_type(slot.ctx) == LLAMA_POOLING_TYPE_NONE) {
                 embd = llama_get_embeddings_ith(ctx, i);
+            } else {
+                embd = llama_get_embeddings_seq(ctx, batch.seq_id[i][0]);
             }
 
-            if (embd == NULL) {
+            if (embd == nullptr) {
                 SLT_ERR(slot, "failed to get embeddings, token = %d, seq_id = %d\n", batch.token[i], batch.seq_id[i][0]);
 
                 res->embedding.push_back(std::vector<float>(n_embd, 0.0f));
@@ -2594,12 +2596,12 @@ struct server_context {
             }
 
             // normalize only when there is pooling
-            // TODO: configurable
             if (llama_pooling_type(slot.ctx) != LLAMA_POOLING_TYPE_NONE) {
                 common_embd_normalize(embd, embd_res.data(), n_embd, 2);
                 res->embedding.push_back(embd_res);
+                break;
             } else {
-                res->embedding.push_back({ embd, embd + n_embd });
+                res->embedding.emplace_back(embd, embd + n_embd);
             }
         }