Inference Server

Starting the server

Always use the inference entry point — never vllm serve or python -m vllm.entrypoints.openai.api_server directly. The entry point runs setup_vllm_env() which configures environment variables (LoRA, multiprocessing) before vLLM is imported.

# With a TOML config
uv run inference @ path/to/config.toml

# With CLI overrides
uv run inference --model.name Qwen/Qwen3-0.6B --model.max_model_len 2048 --model.enforce_eager

# Combined
uv run inference @ path/to/config.toml --server.port 8001 --gpu-memory-utilization 0.5

Custom endpoints

The server extends vLLM with:

• •/v1/chat/completions/tokens — accepts token IDs as prompt input (used by multi-turn RL rollouts)
• •/update_weights — hot-reload model weights from the trainer
• •/load_lora_adapter — load LoRA adapters at runtime
• •/init_broadcaster — initialize weight broadcast for distributed training

Testing the server

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Qwen/Qwen3-0.6B",
    "messages": [{"role": "user", "content": "Hi"}],
    "max_tokens": 50
  }'

Key files

• •src/prime_rl/inference/server.py — entry point, env var setup
• •src/prime_rl/inference/config.py — InferenceConfig and all sub-configs
• •src/prime_rl/inference/vllm/server.py — FastAPI routes and vLLM monkey-patches
• •configs/debug/infer.toml — minimal debug config