transformers 推理如何设置推理参数?

[linxinqiang90]

model_path = "PaddleOCR-VL-1.5"
model = AutoModelForImageTextToText.from_pretrained(model_path, torch_dtype=torch.bfloat16).to(DEVICE).eval()
processor = AutoProcessor.from_pretrained(model_path)

inputs = self.processor.apply_chat_template(
    messages,
    add_generation_prompt=True,
    tokenize=True,
    return_dict=True,
    return_tensors="pt",
    images_kwargs={
        "size": {"shortest_edge": shortest_edge,
                 "longest_edge": longest_edge}},
).to(self.device)

with torch.no_grad():
    outputs = self.model.generate(**inputs,  use_layout_detection=True,  max_new_tokens=max_new_tokens)

1. use_layout_detection
2. format_block_content
3. merge_layout_blocks
4. ...

除max_new_tokens外，试了几个都不支持，请问该如何设置？