pymllm.configs.server_config
============================

.. py:module:: pymllm.configs.server_config


Classes
-------

.. autoapisummary::

   pymllm.configs.server_config.ServerConfig


Module Contents
---------------

.. py:class:: ServerConfig

   Centralized runtime configuration for the MLLM server.


   .. py:attribute:: model_path
      :type:  Optional[pathlib.Path]
      :value: None



   .. py:attribute:: tokenizer_path
      :type:  Optional[pathlib.Path]
      :value: None



   .. py:attribute:: tokenizer_mode
      :type:  Literal['auto', 'slow', 'fast']
      :value: 'auto'



   .. py:attribute:: load_format
      :type:  Literal['auto', 'safetensors']
      :value: 'auto'



   .. py:attribute:: trust_remote_code
      :type:  bool
      :value: False



   .. py:attribute:: download_dir
      :type:  Optional[pathlib.Path]
      :value: None



   .. py:attribute:: context_length
      :type:  Optional[int]
      :value: None



   .. py:attribute:: dtype
      :type:  Literal['auto', 'float16', 'bfloat16', 'float32']
      :value: 'auto'



   .. py:attribute:: host
      :type:  str
      :value: '127.0.0.1'



   .. py:attribute:: port
      :type:  int
      :value: 30000



   .. py:attribute:: fastapi_root_path
      :type:  str
      :value: ''



   .. py:attribute:: api_key
      :type:  Optional[str]
      :value: None



   .. py:attribute:: admin_api_key
      :type:  Optional[str]
      :value: None



   .. py:attribute:: served_model_name
      :type:  Optional[str]
      :value: None



   .. py:attribute:: file_storage_path
      :type:  pathlib.Path


   .. py:attribute:: cors_allow_origins
      :type:  list[str]
      :value: ['*']



   .. py:attribute:: mem_fraction_static
      :type:  Optional[float]
      :value: None



   .. py:attribute:: max_running_requests
      :type:  Optional[int]
      :value: 1



   .. py:attribute:: max_queued_requests
      :type:  Optional[int]
      :value: None



   .. py:attribute:: max_total_tokens
      :type:  Optional[int]
      :value: None



   .. py:attribute:: chunked_prefill_size
      :type:  Optional[int]
      :value: None



   .. py:attribute:: max_prefill_tokens
      :type:  Optional[int]
      :value: None



   .. py:attribute:: schedule_policy
      :type:  Literal['auto', 'fcfs']
      :value: 'fcfs'



   .. py:attribute:: schedule_conservativeness
      :type:  float
      :value: 1.0



   .. py:attribute:: sleep_on_idle
      :type:  bool
      :value: False



   .. py:attribute:: stream_interval
      :type:  int
      :value: 1



   .. py:attribute:: stream_output
      :type:  bool
      :value: True



   .. py:attribute:: base_gpu_id
      :type:  int
      :value: 0



   .. py:attribute:: attention_backend
      :type:  Literal['auto', 'flashinfer']
      :value: 'auto'



   .. py:attribute:: gdn_decode_backend
      :type:  Literal['auto', 'flashinfer', 'mllm_kernel', 'pytorch']
      :value: 'auto'



   .. py:attribute:: sampling_backend
      :type:  Optional[str]
      :value: None



   .. py:attribute:: disable_cuda_graph
      :type:  bool
      :value: False



   .. py:attribute:: enable_torch_compile
      :type:  bool
      :value: False



   .. py:attribute:: torch_compile_max_bs
      :type:  int
      :value: 32



   .. py:attribute:: random_seed
      :type:  Optional[int]
      :value: 42



   .. py:attribute:: reasoning_parser
      :type:  Optional[str]
      :value: None



   .. py:attribute:: tool_call_parser
      :type:  Optional[str]
      :value: None



   .. py:attribute:: log_level
      :type:  Literal['debug', 'info', 'warning', 'error', 'critical']
      :value: 'info'



   .. py:attribute:: enable_metrics
      :type:  bool
      :value: False



   .. py:attribute:: show_time_cost
      :type:  bool
      :value: False



   .. py:attribute:: decode_log_interval
      :type:  int
      :value: 40



   .. py:attribute:: enable_shared_queue
      :type:  bool
      :value: False



   .. py:attribute:: disable_radix_cache
      :type:  bool
      :value: False



   .. py:attribute:: radix_cache_page_size
      :type:  int
      :value: 1



   .. py:attribute:: enable_mamba_cache
      :type:  bool
      :value: False



   .. py:attribute:: tensor_transport_mode
      :type:  str
      :value: 'default'



   .. py:attribute:: cuda_ipc_pool_size_mb
      :type:  int
      :value: 512



   .. py:attribute:: cuda_ipc_recycle_interval
      :type:  float
      :value: 0.1



   .. py:attribute:: extra_options
      :type:  dict[str, Any]


   .. py:method:: __post_init__()