"//waymo/ml/compiler/mlir/...",
        # Allow visibility from the mlir language server.
        "//learning/brain/mlir/mlir_lsp_server/...",
        "//research/language_modeling/sentence_explorer/ondevice/...",
        "//learning/brain/research/babelfish/inference/speech_tflite/mlir/...",
    ],
)

td_library(
    name = "tensorflow_lite_ops_td_files",
    srcs = [
        "ir/tfl_op_enums.td",

	if _, ok := metadata[strings.ToLower(xhttp.ContentType)]; !ok {
		metadata[strings.ToLower(xhttp.ContentType)] = "binary/octet-stream"
	}

	// https://github.com/google/security-research/security/advisories/GHSA-76wf-9vgp-pj7w
	for k := range metadata {
		if equals(k, xhttp.AmzMetaUnencryptedContentLength, xhttp.AmzMetaUnencryptedContentMD5) {
			delete(metadata, k)
		}
	}

			// Do not need to send any internal metadata
			// values to client.
			toRemove = append(toRemove, k)
			continue
		}

		// https://github.com/google/security-research/security/advisories/GHSA-76wf-9vgp-pj7w
		if equals(k, xhttp.AmzMetaUnencryptedContentLength, xhttp.AmzMetaUnencryptedContentMD5) {
			toRemove = append(toRemove, k)
			continue
		}
	}

	// compatible with the internal representation used here:
	//
	// Vincent Lefèvre: "The Generic Multiple-Precision Floating-
	// Point Addition With Exact Rounding (as in the MPFR Library)"
	// http://www.vinc17.net/research/papers/rnc6.pdf

	if debugFloat {
		validateBinaryOperands(x, y)
	}

	// compute exponents ex, ey for mantissa with "binary point"
	// on the right (mantissa.0) - use int64 to avoid overflow

	// whole checksums for many years now, no need to allocate.

	fi.Metadata = make(map[string]string, len(j.MetaUser)+len(j.MetaSys))
	for k, v := range j.MetaUser {
		// https://github.com/google/security-research/security/advisories/GHSA-76wf-9vgp-pj7w
		if equals(k, xhttp.AmzMetaUnencryptedContentLength, xhttp.AmzMetaUnencryptedContentMD5) {
			continue
		}
		if equals(k, "x-amz-storage-class") && v == storageclass.STANDARD {

http://deeplearning.cs.cmu.edu/pdfs/Hochreiter97_lstm.pdf
S. Hochreiter and J. Schmidhuber. 'Long Short-Term Memory'. Neural Computation,
9(8):1735-1780, 1997.
The peephole implementation is based on:
https://research.google.com/pubs/archive/43905.pdf
Hasim Sak, Andrew Senior, and Francoise Beaufays. 'Long short-term memory
recurrent neural network architectures for large scale acoustic modeling.'
INTERSPEECH, 2014.

		if stringsHasPrefixFold(k, ReservedMetadataPrefixLower) {
			continue
		}

		if equals(k, xhttp.AmzBucketReplicationStatus) {
			continue
		}

		// https://github.com/google/security-research/security/advisories/GHSA-76wf-9vgp-pj7w
		if equals(k, xhttp.AmzMetaUnencryptedContentLength, xhttp.AmzMetaUnencryptedContentMD5) {
			continue
		}
		meta[k] = v
	}

	if oi.ContentEncoding != "" {

*   Adds `beta` hyperparameter to
    [FTRL](https://www.tensorflow.org/api_docs/python/tf/keras/optimizers/Ftrl)
    optimizer classes (Keras and others) to match
    [FTRL paper](https://research.google.com/pubs/archive/41159.pdf).
*   `Optimizer.__init__` now accepts a `gradient_aggregator` to allow for
    customization of how gradients are aggregated across devices, as well as