GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
ridcl
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
ridcl
Model Tree
Qwen/Qwen3-VL-4B-Instruct
Input Modalities
Output Modalities
Supported Functionality
VQA model that also supports bounding boxes.
Run via vLLM:
markdown
vllm serve ridcl/paperwerk-vqa \--max-model-len 32768 \--enable-auto-tool-choice \--tool-call-parser hermes \--tensor-parallel-size 2
Examples input:
markdown
{"role": "user","content": [{"type": "image_url","image_url": {"url": "data:image/jpeg;base64,..."},},{"type": "text", "text": "Extract values:\n- contract_date\n- names of the parties\n- Where is the company registered?"},],}
Example output (bbox format: [xmin, ymin, xmax, ymax], 0..1000 scale):
markdown
[{"query": "contract_date","value": "1 April 2025","box_2d": [466, 693, 544, 705],"index": 0,},{"query": "names of the parties","value": "CoolCo B.V.","box_2d": [151, 279, 257, 292],"index": 0,},{"query": "names of the parties","value": "John Doe","box_2d": [151, 403, 261, 415],"index": 0,},{"query": "names of the parties","value": "Jane Doe","box_2d": [151, 509, 268, 522],"index": 0,},{"query": "Where is the company registered?","value": "Amsterdam","box_2d": [164, 302, 281, 315],"index": 0,},]