Summary: 62 instances, 50 unique

Text	Count
// TODO: [AD] Similarly for the embedding projection used in ALBERT.	1
// TODO: Add argument that controls this.	1
// @TODO: move stbi__convert_format16 to here	1
// TODO: figure out why the string equality is being ignored	1
// TODO: Re-enable this once we have settled on a single toolchain.	4
// TODO: Specify number of loops, rather than looping forever.	1
// TODO: Extend handling to different tensor types.	1
/// TODO: [DOC] Add a documentation string and fix the parameter descriptions.	1
// TODO: Incorporate file offset.	2
// TODO: loss threshold clipping	1
// TODO: Come up with better names for these protocols	1
// TODO: Optimize bounding box to only provided updated part.	1
// TODO: remove computeEmbeddingsVJP once inout differentiation is supported!	1
// TODO: need to persist the activation function.  Until then, manually set it to tanh.	1
// TODO: Reduceinto and +=	1
// TODO: Replace the following, so as to not waste memory for non-projection cases.	1
// TODO: read/write activation, strides, and padding from checkpoint file	1
// TODO: Extend handling to different tensor types.	1
// @TODO: move stbi__convert_format to here	1
// TODO: Proper error propagation for this.	2
// TODO: Find a way for this to be checked only at initialization, not during training or	1
// TODO: Remove the following tensor workaround when above is fixed.	1
// TODO: Add BGR byte reordering.	1
// TODO: Find a nice way to support caching.	1
// @TODO: special case RGB-to-Y (and RGBA-to-YA) for 8-bit-to-16-bit case to keep more precision	1
// TODO: Handle splitting into multiple shards.	1
// TODO: avoid copies?	1
// TODO: Proper error propagation for this.	2
// TODO: Find a more efficient way of calculating this without casting to bytes twice.	1
// TODO: Write this directly to disk, rather than accumulating it in memory.	1
// TODO: regularization	1
// TODO: Make dropout generic over the probability type.	3
// TODO: Split out the optimizer as a separate specification.	1
// FIXME: Unless the palette length is 256xRGB the GIF is invalid for some reason...	1
// TODO: Add documentation.	1
// TODO: Convert to a generic constraint once TF-427 is resolved.	4
/// TODO: Consider promoting this into a public protocol in swift-apis?	1
// TODO: Remove this when differentiation uses per-instance zeros	1
// TODO: Expand beyond using a single binary shard.	1
// TODO: Adapt the following to the colors in the input image.	1
// TODO: [AD] Avoid using token type embeddings for RoBERTa once optionals are supported in AD.	1
// FIXME: Need to generalize `DatasetUtilities.downloadResource` to accept	1
// TODO: Use parallelism to grab the samples in parallel.	1
// TODO: Re-enable this for the stock toolchain when it can be realigned with VectorProtocol.	1
// TODO: Re-enable this once we have settled on a single toolchain.	2
// TODO: [AD] Change to optional once supported.	1
// TODO: Convert this to suitable protocol with Discriminator configuration	1
// TODO: specialize where value is Int32 such that keyForValue is actually an	1
// TODO: read/write activation, strides, padding, and dialations from checkpoint file	1
// TODO: need to persist a sample generator and set this URL	1