[BugFix] rollback max_tokens and min_tokens when continue to infer (#5082)

* [BugFix] rollback max_tokens and min_tokens when continue to infer * [BugFix] rollback max_tokens and min_tokens when continue to infer * [fix] add more logger info: max_tokens --------- Co-authored-by: liqinrui <liqinrui@baidu.com>
2026-04-23 00:17:25 +08:00 · 2025-11-19 18:43:42 +08:00
parent 43f0c7557e
commit a5cd7c9039
2 changed files with 8 additions and 5 deletions
@@ -264,14 +264,12 @@ class Ernie4_5_VLProcessor(Ernie4_5Processor):
        if max_model_len is not None and len(request["prompt_token_ids"]) > max_model_len:
            request["prompt_token_ids"] = request["prompt_token_ids"][: max_model_len - 1]

-        tmp_max_tokens = 0
        if request.get("max_tokens") is None:
            request["max_tokens"] = max(1, max_model_len - len(request["prompt_token_ids"]))
-            tmp_max_tokens = request["max_tokens"]
        else:
            request["max_tokens"] = min(max_model_len - len(request["prompt_token_ids"]), request["max_tokens"])
        if request.get("reasoning_max_tokens") is None:
-            request["reasoning_max_tokens"] = max(int(tmp_max_tokens * 0.8), 1)
+            request["reasoning_max_tokens"] = max(int(request["max_tokens"] * 0.8), 1)
        data_processor_logger.info(f"Processed request {request}")

        if request.get("top_p") is not None and request.get("top_p") < _SAMPLING_EPS: