server : fix EOS token detection with disabled cache (ggml-org#5938)

ggerganov · hazelnutcloud · commit 2aba0cb697ea · 2024-03-10T15:38:20.000+08:00
diff --git a/examples/server/server.cpp b/examples/server/server.cpp
@@ -1123,7 +1123,7 @@ struct server_context {
             });
         }
 
-        if (!slot.cache_tokens.empty() && result.tok == llama_token_eos(model)) {
+        if (result.tok == llama_token_eos(model)) {
             slot.stopped_eos    = true;
             slot.has_next_token = false;
 

Original file line number	Diff line number	Diff line change
`@@ -1123,7 +1123,7 @@ struct server_context {`
`1123`	`1123`	`});`
`1124`	`1124`	`}`
`1125`	`1125`
`1126`		`- if (!slot.cache_tokens.empty() && result.tok == llama_token_eos(model)) {`
	`1126`	`+ if (result.tok == llama_token_eos(model)) {`
`1127`	`1127`	`slot.stopped_eos = true;`
`1128`	`1128`	`slot.has_next_token = false;`
`1129`	`1129`