stelliferous

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

  • checkpoint_pickle.bend
  • tensor_batch_norm.bend
  • tensor_checkpoint.bend
  • tensor_f32.bend
  • tensor_layer.bend
  • tensor_loss.bend
  • tensor_optimizer.bend
  • tensor_training.bend
• 0.0.2.0 indexed

Definitions (2,086)

tensor_f32.bend

• def greater
• def greater_equal
• def equal
• def not_equal
• def less_scalarScalar comparisons share the eight-lane map and preserve the tensor layout.
• def less_equal_scalar
• def greater_scalar
• def greater_equal_scalar
• def equal_scalar
• def not_equal_scalar
• def select
• def clamp
• def activateActivations: Act.silu() and the other values of lib/math_activation.bend carry their own planner cost.
• def relu
• def relu6
• def leaky_relu
• def sigmoid
• def silu
• def tanh
• def hard_sigmoid
• def hard_swish
• def gelu_tanhPyTorch's gelu(approximate='tanh').
• def conv2dOrdered convolution over CHW or NCHW; the activation runs on each finished leaf while it is in cache.
• def conv2d_customconv2d with a custom activation of the given scalar operations per element.
• def applyCustom elementwise functions are passed as direct templates, never records, with their scalar operations per element.
• def apply_indexedThe function also receives the element's row-major index.
• def arangestart, start + 1, ...
• def sum
• def max
• def min
• def meansum divided by the number of elements; an empty tensor gives NaN.
• def reshape
• def transposeThe result's axis i is the argument's axis order[i].
• def slicecount elements from start along axis, every step-th one.
• def set_slicetarget with part written at start along axis.
• def Parts
• def singleBuild the parts of concat: F.cons(a, F.cons(b, F.single(c))).
• def cons
• def concat
• def upsample_nearestRepeats each element of the last two axes scale times in each direction.
• def max_pool2d_withOver the last two axes; positions outside the input count as -inf.
• def max_pool2dWindows of kernel by kernel that do not overlap: stride kernel, padding 0.
• def broadcast_toThe tensor repeated to shape under NumPy broadcasting; add, sub, mul, max and min broadcast their operands in the same way.
• def sum_axisReductions along one axis remove that axis.
• def max_axis
• def min_axis
• def expElementwise functions share the eight-value traversal used by activations.
• def log
• def softmaxKeep a singleton reduction axis for broadcasting.
• def mean_axissum_axis divided by the extent of the axis.
• def padbefore[i] new elements in front of axis i and after[i] behind it, all equal to value.
• def pad_edgeThe same, each new element repeating the nearest element of the tensor.

tensor_file.bend

• def Tensor
• def Read
• def Written
• def chunkValues per File effect.
• def from_bits
• def to_bits
• type Words
• type TargetThe buffer of the tensor, the values found so far, and a word in progress.
• type Reading
• type Outcome
• def gather
• def push
• def trickle
• def quadsFour bytes at a time while the bytes are whole words; the values are newest first.
• def words
• def placedValues beyond the count of the shape are counted and not written.
• def fill
• def absorbed
• def received
• def taken
• def closed
• def read_allA file longer than its shape stops at the fuel and fails the count.
• def fresh
• def fuel
• def fault
• def sized
• def whole
• def aligned
• def checked
• def loaded
• def opened
• def counted
• def loadThe file holds the values of the shape in row-major order and nothing else.
• type Header
• def codes
• def magic
• def starts
• def at
• def contains
• def afterThe text after the first occurrence of the pattern.
• def pushed
• def closed_shape
• def extentsExtents up to the closing parenthesis: digits, commas and spaces.
• def shaped
• def typed
• def described
• def region
• def versionedVersion 1 has a two-byte header length; later versions have four bytes.