Getting StartedΒΆ

Start here for a practical path from environment setup to a working Diffulex run.

  • Quickstart: install, run a limited benchmark, call the Python API, and start the HTTP server.

  • Installation: prepare Python, CUDA, vLLM, checkpoint paths, and documentation build dependencies.

Before running the examples, prepare:

  • a Python 3.11+ environment;

  • CUDA-visible NVIDIA GPUs;

  • local model checkpoint directories;

  • vLLM if the selected config uses vLLM-backed layers or baselines.

Use DATASET_LIMIT or --dataset-limit for the first run. Increase limits only after model loading and a small generation are successful.