Role Name

ansible-vllm

Requirements

None, currently.

Role Variables

|| Name || Default || Description || | VLLM_MODEL | None | The path to the model file, e.g. lmsys/vicuna-7b-v1.5 | | VLLM_MAX_NUM_BATCHED_TOKENS | 8192 | The maximum number of tokens to batch together | | VLLM_MAX_MODEL_LENGTH | 8192 | The maximum length of the model | | VLLM_API_SERVER_ENABLE | true | Whether to enable the API server | | VLLM_API_SERVER_LOCAL_ONLY | false | Whether to only listen on localhost | | VLLM_API_SERVER_HOST | 0.0.0.0 | The host to listen on (0.0.0.0 = listen on any address, 127.0.0.1 listen on localhost only) | | VLLM_API_SERVER_PORT | 8000 | The port to listen on | | VLLM_TENSOR_PARALLEL_SIZE | 1 | The number of GPUs to use for tensor parallelism; you can increase if using distributed compute |

Dependencies

Example Playbook

Including an example of how to use your role (for instance, with variables passed in as parameters) is always nice for users too:

- hosts: servers
  tasks:
    - ansible.builtin.import_role: 
        name: ansible-vllm

License

BSD

Author Information

Edwin Skidmore (edwins@arizona.edu)