Vllm.Config.CUDAGraphMode (VLLM v0.3.0)

Copy Markdown View Source

Constants for the cudagraph mode in CompilationConfig.

Meanwhile, the subset enum NONE, PIECEWISE and FULL are also treated as concrete runtime mode for cudagraph runtime dispatching.

Summary

Types

t()

@opaque t()

Functions

full(ref)

@spec full(SnakeBridge.Ref.t()) :: {:ok, term()} | {:error, Snakepit.Error.t()}

full_and_piecewise(ref)

@spec full_and_piecewise(SnakeBridge.Ref.t()) ::
  {:ok, term()} | {:error, Snakepit.Error.t()}

full_decode_only(ref)

@spec full_decode_only(SnakeBridge.Ref.t()) ::
  {:ok, term()} | {:error, Snakepit.Error.t()}

name(ref)

@spec name(SnakeBridge.Ref.t()) :: {:ok, term()} | {:error, Snakepit.Error.t()}

new(opts \\ [])

@spec new(keyword()) :: {:ok, SnakeBridge.Ref.t()} | {:error, Snakepit.Error.t()}

Constructs CUDAGraphMode.

none(ref)

@spec none(SnakeBridge.Ref.t()) :: {:ok, term()} | {:error, Snakepit.Error.t()}

piecewise(ref)

@spec piecewise(SnakeBridge.Ref.t()) :: {:ok, term()} | {:error, Snakepit.Error.t()}

value(ref)

@spec value(SnakeBridge.Ref.t()) :: {:ok, term()} | {:error, Snakepit.Error.t()}