stelliferous
Releases (2)
Definitions (2,086)
tensor_file.bend
- def marked
- def header
- def refused
- def npy_counted
- def npy_parsed
- def npy_received
- def npy_headed
- def npy_opened
- def load_npyAn FP32 array saved by NumPy, with its shape.
- type Writing
- def bytes
- def encode
- def read_chunkThe cursor retains the array; only the current chunk becomes a list.
- def pending
- def continued
- def after_write
- def write_chunk
- def finished
- def write_all
- def unsaved
- def written
- def createdThe prefix is written first; a raw file has none.
- def saving
- def saveWrites the values in row-major order and returns the tensor.
- def joined
- def tupleNumPy writes (), (3,) and (2, 3).
- def spaces
- def framedVersion 1: the header ends in a newline and the data starts at a multiple of 64 bytes.
- def npy_header
- def npy_saving
- def save_npyWrites a .npy file that NumPy loads with the tensor's shape.
tensor_indexed.bend
- def value
- def read
- def destination
- type Call
- def run
- type BlockThe same mapping over blocks of consecutive positions.
- def block_index
- type Blocks
- def next_block
- def block_step
- def run_blocks
- def run_whole
- def block_limitBlocks are used only up to the machine capacity, where the block size divides a position in U32 without overflow.
- def run_tiledtotal positions in blocks(context) blocks of size(context) positions.
tensor_lanes.bend
- def identity
- def read_result
- def read
- def write
- def combine_tile
- type Binary
- def pair_reads
- def pair_gather
- def pair_scatter
- def pair_result
- def pair_finishThe checked caller has already validated both views and owners.
- def pair_map
- def pair_buffers
- def pair_when
- def combine
- type Selected
- def select_tile
- def select_reads
- def select_gather
- def select_scatter
- def select_joined
- def select_rejected
- def select_chosen
- def select_finishThe final 0..7 values use the existing certified binary kernels.
- def select_result
- def select
tensor_layout.bend
- def extents
- def naturals
- def copied
- def gather_views
- def gather_statusCopy the elements selected by source_view, in its row-major order, into a new tensor whose storage is laid out as view_shape and labelled target_shape.
- def result_of
- def swapped
- def copy_bandsThe only layout operation that keeps its argument: (original, copy).
- def copy
- def permuted
- def transpose_view
- def transposeThe result's axis i is the source axis order[i].
- def sliced
- def slice_view
- def sliceElements start, start + step, ...
- def same_count
- def relabel_ready
- def relabel
- def reshapeSame elements in the same row-major order under new extents.
- def prepend_repeatedNearest-neighbour upsampling repeats each element of the last two axes scale times, by copying from a view whose repeated axes have stride zero.
- def repeated_axes
- def with_offset
- def repeated
- def scaled_extents
- def upsample_view
- type Upsample
- def upsample_leaf
- def upsample_supported
- def upsample_bands