CudaTokenEmbeddingOp.ixx File Reference

CUDA implementation of the TokenEmbedding operation. More...

#include <cuda_fp16.h>
#include <string>
#include <stdexcept>
#include <cstdint>
#include <format>
#include <sstream>
import Cuda.Debug;
import Compute.OperationRegistrarHelpers;
import Compute.CudaTensorDataType;
import Compute.CudaDeviceMemoryResource;
import Compute.CudaTokenEmbeddingOp:Dispatch;
import Dnn.TensorDataType;
import Compute.ExecutionContext;
import Dnn.Tensor;
import Dnn.Components.TokenEmbeddingConfig;
import Compute.DeviceType;
import Dnn.ITensor;
import Compute.OperationType;
import Dnn.TensorTypes;
import Dnn.Component;
import Dnn.TensorDataTypeTraits;
import Compute.UnaryOperation;
import Compute.IExecutionContext;

Classes
class	Mila::Dnn::Compute::Cuda::TokenEmbedding::CudaTokenEmbeddingOp< TInput, TPrecision >
class	Mila::Dnn::Compute::Cuda::TokenEmbedding::CudaTokenEmbeddingOpRegistrar

Namespaces
namespace	Mila
	Mila main API namespace.
namespace	Mila::Dnn
namespace	Mila::Dnn::Compute
namespace	Mila::Dnn::Compute::Cuda
namespace	Mila::Dnn::Compute::Cuda::TokenEmbedding

Detailed Description

CUDA implementation of the TokenEmbedding operation.

Pure vocabulary lookup: output[b,t,:] = wte[X[b,t],:]. No positional information. Positional encoding is handled downstream by a dedicated encoding component (RoPE, ALiBi, or Learned).

Template Parameters

TInput	Data type of token index input (INT32).
TPrecision	Precision of embedding output (FP32 or FP16).