Switch deepseek-ocr-vllm.py to stable vLLM (>=0.15.1), drop nightly index

Fix config_name=None breaking push_to_hub + use cu129 nightly index

Add --config/--create-pr support to deepseek-ocr-vllm.py

Pin datasets>=4.0.0, drop pyarrow pin

Pin pyarrow<18 to fix datasets compatibility

Fix deepseek-ocr-vllm.py: use official vLLM offline pattern (P0)

Remove extraneous f-string prefixes (ruff F541)

Replace deprecated HF_TRANSFER with HF_XET_HIGH_PERFORMANCE

Add prompt mode presets and remove test script

prompt

Add UV_TORCH_BACKEND=auto to HF Jobs example

Fix vLLM nightly wheel index URL

Simplify HF Jobs token syntax to -s HF_TOKEN

Switch DeepSeek-OCR scripts to vLLM nightly wheels

Add DeepSeek-OCR vLLM scripts (PR #27247 testing)