server : support max_completion_tokens request property (#19831)

"max_tokens" is deprectated in favor of "max_completion_tokens" which
sets the upper bound for reasoning+output token.

Closes: #13700

diff --git a/tools/server/server-task.cpp b/tools/server/server-task.cpp
index a137427c69c02d4a91df8a00f82bcc49addd5901..739e30a7046ea432764945fcaecf4aa740918b1a 100644 (file)
--- a/tools/server/server-task.cpp
+++ b/tools/server/server-task.cpp
@@ -204,7 +204,8 @@ task_params server_task::params_from_json_cmpl(
     params.cache_prompt     = json_value(data,       "cache_prompt",       defaults.cache_prompt);
     params.return_tokens    = json_value(data,       "return_tokens",      false);
     params.return_progress  = json_value(data,       "return_progress",    false);
-    params.n_predict        = json_value(data,       "n_predict",          json_value(data, "max_tokens", defaults.n_predict));
+    auto max_tokens         = json_value(data,       "max_tokens",         defaults.n_predict);
+    params.n_predict        = json_value(data,       "n_predict",          json_value(data, "max_completion_tokens", max_tokens));
     params.n_indent         = json_value(data,       "n_indent",           defaults.n_indent);
     params.n_keep           = json_value(data,       "n_keep",             defaults.n_keep);
     params.n_discard        = json_value(data,       "n_discard",          defaults.n_discard);