vLLM before 0.31.0 skips the min_tokens <= max_tokens check for /inference/v1/generate requests with a null max_tokens, so the EngineCore input thread later dies on ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results