From 33d7b70c881b1f9ebabd381daf3ad9cb283ce028 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Tue, 3 Dec 2024 10:58:43 +0200
Subject: [PATCH] server : do not speculate during prompt processing

ggml-ci
---
 examples/server/server.cpp | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/examples/server/server.cpp b/examples/server/server.cpp
index 8eca14b86..b91671de6 100644
--- a/examples/server/server.cpp
+++ b/examples/server/server.cpp
@@ -2322,6 +2322,10 @@ struct server_context {
                     continue;
                 }
 
+                if (slot.state != SLOT_STATE_GENERATING) {
+                    continue;
+                }
+
                 llama_token id = slot.sampled;
 
                 struct common_speculative_params params_spec;