can't not run directly

[Xavier1994]

I set VLLM_USE_V1=0 and run the example

python3 -m dripper.server --model_path /models/MinerU-HTML/ --state_machine v2 --port 7986
curl -X POST "http://localhost:7986/extract" -H "Content-Type: application/json" -d '{"html": "...", "url": "https://example.com"}

Traceback (most recent call last):
File "/root/MinerU-HTML/dripper/server.py", line 105, in extract_main
result_list = dripper.process(req.html)
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/MinerU-HTML/dripper/api.py", line 404, in process
generate_outputs = generate(
^^^^^^^^^
File "/root/MinerU-HTML/dripper/inference/inference.py", line 104, in generate
res_list = llm.generate(prompt_list, sampling_params=sampling_params_arg)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/python311/lib/python3.11/site-packages/vllm/entrypoints/llm.py", line 440, in generate
self._validate_and_add_requests(
File "/usr/local/python311/lib/python3.11/site-packages/vllm/entrypoints/llm.py", line 1613, in _validate_and_add_requests
raise e
File "/usr/local/python311/lib/python3.11/site-packages/vllm/entrypoints/llm.py", line 1601, in _validate_and_add_requests
request_id = self._add_request(
^^^^^^^^^^^^^^^^^^
File "/usr/local/python311/lib/python3.11/site-packages/vllm/entrypoints/llm.py", line 1700, in _add_request
engine_request, tokenization_kwargs = self._process_inputs(
^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/python311/lib/python3.11/site-packages/vllm/entrypoints/llm.py", line 1680, in _process_inputs
engine_request = self.processor.process_inputs(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/python311/lib/python3.11/site-packages/vllm/v1/engine/processor.py", line 383, in process_inputs
self._validate_params(params)
File "/usr/local/python311/lib/python3.11/site-packages/vllm/v1/engine/processor.py", line 185, in _validate_params
self._validate_supported_sampling_params(params)
File "/usr/local/python311/lib/python3.11/site-packages/vllm/v1/engine/processor.py", line 150, in _validate_supported_sampling_params
raise ValueError(
ValueError: vLLM V1 does not support per request user provided logits processors.

[Adela-Yu-Coder]

1、Please avoid using the state_machine parameter and keep it as the default value None (not the string "None"). We recommend using the version without state_machine.
Please execute:
unset DRIPPER_STATE_MACHINE
Then run:
python -m dripper.server
or

python -m dripper.server \
    --model_path /path/to/your/model \
    --port 7986

2、Please check your vllm version and ensure vllm==0.11.1 and transformers>=4.57.3.

[zhukefucn]

vllm==0.11.1 和 transformers>=4.57.3，环境变量DRIPPER_STATE_MACHINE=v1，还是报同样错误。unset DRIPPER_STATE_MACHINE后，还是报同样错误。

[ideaflow]

Hello, we have recently decoupled the inference backend from the dripper framework (refer to dripper/inference/imp.py). Currently, we have implemented inference backends for vllm and transformers (you can also refer to #10 to implement a new inference backend). Therefore, you can first try to run the model loading and inference code independently in your own development environment, for example, using vllm:

from vllm import LLM
llm_model = LLM(model=model_path, tensor_parallel_size=1, ...) # add any parameters you need
res = llm_model.generate(["...", "...", ...]) # replace '...' with actual prompts

If the code above runs successfully, then our dripper framework should also work fine.