@@ -217,11 +217,12 @@ static std::vector<training_sample> load_jsonl(
217217 break ;
218218 }
219219 }
220- if (last_assistant_content.find (" Error:" ) != std::string::npos ||
221- last_assistant_content.find (" error:" ) != std::string::npos) {
222- LOG_DBG (" %s: skipping line %d — assistant response contains error marker\n " , __func__, lineno);
223- continue ;
224- }
220+ // // this should be done on the python side...
221+ // if (last_assistant_content.find("Error:") != std::string::npos ||
222+ // last_assistant_content.find("error:") != std::string::npos) {
223+ // LOG_DBG("%s: skipping line %d — assistant response contains error marker\n", __func__, lineno);
224+ // continue;
225+ // }
225226 }
226227
227228 // Split into prompt (no loss) + last assistant response (loss).
@@ -262,11 +263,12 @@ static std::vector<training_sample> load_jsonl(
262263 }
263264 } else if (j.contains (" prompt" ) && j.contains (" response" )) {
264265 response_text = j[" response" ].get <std::string>();
265- if (response_text.find (" Error:" ) != std::string::npos ||
266- response_text.find (" error:" ) != std::string::npos) {
267- LOG_DBG (" %s: skipping line %d — response contains error marker\n " , __func__, lineno);
268- continue ;
269- }
266+ // // this should be done on the python side...
267+ // if (response_text.find("Error:") != std::string::npos ||
268+ // response_text.find("error:") != std::string::npos) {
269+ // LOG_DBG("%s: skipping line %d — response contains error marker\n", __func__, lineno);
270+ // continue;
271+ // }
270272 prompt_text = j[" prompt" ].get <std::string>();
271273 } else if (j.contains (" text" )) {
272274 response_text = j[" text" ].get <std::string>();
0 commit comments