stelliferous

Search uses release 0.0.2.0 · 2,086 definitions

Releases (2)

0.0.3.0 invalid
expected a fresh constructor name (duplicate declaration: Text)

at checkpoint_pickle.bend:14
8 files do not load
0.0.2.0 indexed

Definitions (2,086)

tensor_layout.bend

def upsample_nearest
def settle
def count_or_zero
def extent_at
def settled
def placed
def same_extents
def placeThe part written through a view of the target storage: one checked copy, which rejects a view whose extents differ from the part.
def place_status
def set_slicetarget[..., start : start + part extent, ...] = part along axis: the part is written through the slice of the target that it replaces.
type SurveyConcatenation along an axis.
def surveyed
def prepend_survey
def surveySummarizes the extents of every part and returns the parts in order.
def with_extent
def append_at
def append_part
def appended_all
def append_parts
def join_selected
def join_discovered
def join_compatible_selectedThe final branch is one Bool.
def join_compatible
def join_status
def joined
def lead
def concat_first
def concat_led
def concatParts agree on every extent except axis; the result stacks them in order.
def broadcast_statusBroadcasting to given extents: one checked copy from a stretched view.
def broadcast_toThe tensor repeated to the extents target under NumPy broadcasting.
def fold_sliceReduction along one axis.
def fold_stepThe views are built from the extents at every step: a loop that carried one View value through its iterations would make the compiler share every View constructor of the program.
def fold_slices
def folded_valid
def folded
def reduce_axis_countunit is the shape with extent 1 at the axis: the layout of the output and the extents of every slice.
def reduce_axis_status
def reduce_axisThe result has the extents of the tensor without axis.
type BorderPadding.
def border_name
def blank
def edge_stripThe elements at one end of axis, repeated count times: (strip, tensor).
def pad_edge_axis
def pad_edge_axes
def interiorThe view of the padded extents that holds the tensor: on every axis, the extent of the tensor starting after the elements added in front.
def all_zero
def pad_constantConstant padding: storage of the padded extents filled with the value, then the tensor written through the interior view.
def pad_border
def padded_count
def pad_valid
def padbefore[i] elements are added in front of axis i and after[i] behind it.

tensor_model.bend

def Tensor
type Store
def ModelA step: given the store, a value and the store that remains.
def Model.pure
def Model.next
def Model.bind
def missing
def passed
def scan
def taken
def takeThe tensor of that name leaves the store.
def kept
def keepPuts a copy under the name, for a later take, and gives the tensor back.
def run
def value
def resultRuns the steps on the store and gives the result; what remains in the store is dropped.
def joined
def copied_entry
def copyTwo stores with the same tensors, to run a model more than once.
def entry
def load_each
def named
def listed
def opened
def loadA folder without names.txt gives an empty store; every take then fails by name.
def named_rest
def first
def names
def save_each
def listing
def save_listed
def saveGives the store back; a tensor that could not be written comes back failed.

tensor_operations.bend

type Outcome
def valid_buffer
def pair_buffers
def copy_when
def copy
def scatter_whenoutput[address of element i of target] = input[offset of source + i]: a contiguous source written through a strided target view, in row-major order.
def scatter
def combine_buffers
def combine_kernel
def combine_when_with
def combine_with
def combine
def mappedA mapper is the ordered map of count values from position on a certified Buffer.
def lanesThe same map with the values of each round of eight computed together: an operation written with FP32 arithmetic alone then compiles to vector code.
def mapped_lanes
def apply_context_when