skainet-io-core/sk.ainet.io

Package-level declarations

Types

ByteArrayMemoryChunk

class ByteArrayMemoryChunk(data: ByteArray, offset: Int = 0, val size: Long = data.size.toLong()) : MemoryChunk

FallbackMappedMemoryChunk

class FallbackMappedMemoryChunk(val path: String, val fileOffset: Long, data: ByteArray, dataOffset: Int = 0, val size: Long = (data.size - dataOffset).toLong()) : MappedMemoryChunk

Fallback MappedMemoryChunk implementation backed by a heap ByteArray.

JsBlobRandomAccessSource

class JsBlobRandomAccessSource : RandomAccessSource

JavaScript/Browser implementation of RandomAccessSource using Blob API.

class JsBlobRandomAccessSource : RandomAccessSource

WASM JS implementation of RandomAccessSource using Blob API.

JvmFileBackedResolver

object JvmFileBackedResolver

JVM file-backed buffer resolver using memory-mapped I/O.

JvmMappedMemoryChunk

class JvmMappedMemoryChunk : MappedMemoryChunk

JVM implementation of MappedMemoryChunk using FileChannel.map.

JvmRandomAccessSource

class JvmRandomAccessSource : RandomAccessSource

JVM implementation of RandomAccessSource using FileChannel.

MappedMemoryChunk

interface MappedMemoryChunk : MemoryChunk, AutoCloseable

A MemoryChunk backed by a memory-mapped file region.

MappedRandomAccessSource

class MappedRandomAccessSource : RandomAccessSource

A RandomAccessSource backed by a memory-mapped file via JvmMappedMemoryChunk.

interface MemoryChunk

interface ModelReader : AutoCloseable

ParametersLoader

interface ParametersLoader

RandomAccessSource

interface RandomAccessSource : AutoCloseable

A source that supports random access reads at arbitrary positions.

data class TensorInfo(val name: String, val shape: Shape, val dtype: String, val offset: Long, val size: Long, val extra: Map<String, Any> = emptyMap())

Functions

inline suspend fun <T : DType, V> ParametersLoader.load(ctx: ExecutionContext, noinline onTensorLoaded: (String, Tensor<T, V>) -> Unit)