-
Notifications
You must be signed in to change notification settings - Fork 824
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
refactor(dataflow): Refactor conversion to binary contents (#4786)
* Remove unused import * Fix indentation * Remove unnecessary braces in short-form conditionals * Refactor per-data type conversion to bytes to extension methods This has two key benefits. Firstly, the same conversion logic was used for both requests and responses. Thus, we are removing duplication and minimising the risk of differences creeping in. Secondly, the business is easier to read, as the handling is moved outside. * Move business logic functions above implementation details * Rename file to better reflect logic * Refactor typed -> binary contents function as extension methods * Use run context method to ensure no mutations happen on the original object * Reword comments for concision * Reorder enum values into related groups * Add comment on FP16 data type * Add value for required byte order for binary contents This order is defined by the specification, so we should not allow for it to change. https://github.com/triton-inference-server/server/blob/c47b2392a4f81a35f88d416f3356eec60ff83dbf/docs/protocol/extension_binary_data.md#binary-tensor-request * Restrict visibility of per-data type to-binary methods to private
- Loading branch information
Showing
3 changed files
with
224 additions
and
294 deletions.
There are no files selected for viewing
220 changes: 220 additions & 0 deletions
220
scheduler/data-flow/src/main/kotlin/io/seldon/dataflow/kafka/BinaryContent.kt
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,220 @@ | ||
| /* | ||
| Copyright 2023 Seldon Technologies Ltd. | ||
| Licensed under the Apache License, Version 2.0 (the "License"); | ||
| you may not use this file except in compliance with the License. | ||
| You may obtain a copy of the License at | ||
| http://www.apache.org/licenses/LICENSE-2.0 | ||
| Unless required by applicable law or agreed to in writing, software | ||
| distributed under the License is distributed on an "AS IS" BASIS, | ||
| WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| See the License for the specific language governing permissions and | ||
| limitations under the License. | ||
| */ | ||
|
|
||
| package io.seldon.dataflow.kafka | ||
|
|
||
| import com.google.protobuf.kotlin.toByteString | ||
| import io.seldon.mlops.inference.v2.V2Dataplane.InferTensorContents | ||
| import io.seldon.mlops.inference.v2.V2Dataplane.ModelInferRequest | ||
| import io.seldon.mlops.inference.v2.V2Dataplane.ModelInferResponse | ||
| import java.nio.ByteBuffer | ||
| import java.nio.ByteOrder | ||
|
|
||
| // FP16 is only supported for binary contents, as Protobuf has no corresponding type. | ||
| enum class DataType { | ||
| BOOL, | ||
| BYTES, | ||
| UINT8, UINT16, UINT32, UINT64, | ||
| INT8, INT16, INT32, INT64, | ||
| FP16, FP32, FP64, | ||
| } | ||
|
|
||
| private val binaryContentsByteOrder = ByteOrder.LITTLE_ENDIAN | ||
|
|
||
| fun ModelInferRequest.withBinaryContents(): ModelInferRequest { | ||
| return this.toBuilder().run { | ||
| inputsList.forEachIndexed { idx, input -> | ||
| val v = when (DataType.valueOf(input.datatype)) { | ||
| DataType.UINT8 -> input.contents.toUint8Bytes() | ||
| DataType.UINT16 -> input.contents.toUint16Bytes() | ||
| DataType.UINT32 -> input.contents.toUint32Bytes() | ||
| DataType.UINT64 -> input.contents.toUint64Bytes() | ||
| DataType.INT8 -> input.contents.toInt8Bytes() | ||
| DataType.INT16 -> input.contents.toInt16Bytes() | ||
| DataType.INT32 -> input.contents.toInt32Bytes() | ||
| DataType.INT64 -> input.contents.toInt64Bytes() | ||
| DataType.BOOL -> input.contents.toBoolBytes() | ||
| DataType.FP16, // may need to handle this separately in future | ||
| DataType.FP32 -> input.contents.toFp32Bytes() | ||
| DataType.FP64 -> input.contents.toFp64Bytes() | ||
| DataType.BYTES -> input.contents.toRawBytes() | ||
| } | ||
|
|
||
| // Add binary data and clear corresponding contents. | ||
| addRawInputContents(v.toByteString()) | ||
| getInputsBuilder(idx).clearContents() | ||
| } | ||
|
|
||
| build() | ||
| } | ||
| } | ||
|
|
||
| fun ModelInferResponse.withBinaryContents(): ModelInferResponse { | ||
| return this.toBuilder().run { | ||
| outputsList.forEachIndexed { idx, output -> | ||
| val v = when (DataType.valueOf(output.datatype)) { | ||
| DataType.UINT8 -> output.contents.toUint8Bytes() | ||
| DataType.UINT16 -> output.contents.toUint16Bytes() | ||
| DataType.UINT32 -> output.contents.toUint32Bytes() | ||
| DataType.UINT64 -> output.contents.toUint64Bytes() | ||
| DataType.INT8 -> output.contents.toInt8Bytes() | ||
| DataType.INT16 -> output.contents.toInt16Bytes() | ||
| DataType.INT32 -> output.contents.toInt32Bytes() | ||
| DataType.INT64 -> output.contents.toInt64Bytes() | ||
| DataType.BOOL -> output.contents.toBoolBytes() | ||
| DataType.FP16, // may need to handle this separately in future | ||
| DataType.FP32 -> output.contents.toFp32Bytes() | ||
| DataType.FP64 -> output.contents.toFp64Bytes() | ||
| DataType.BYTES -> output.contents.toRawBytes() | ||
| } | ||
|
|
||
| // Add binary data and clear corresponding contents. | ||
| addRawOutputContents(v.toByteString()) | ||
| getOutputsBuilder(idx).clearContents() | ||
| } | ||
|
|
||
| build() | ||
| } | ||
| } | ||
|
|
||
| private fun InferTensorContents.toUint8Bytes(): ByteArray = this.uintContentsList | ||
| .flatMap { | ||
| ByteBuffer | ||
| .allocate(1) | ||
| .put(it.toByte()) | ||
| .array() | ||
| .toList() | ||
| } | ||
| .toByteArray() | ||
|
|
||
| private fun InferTensorContents.toUint16Bytes(): ByteArray = this.uintContentsList | ||
| .flatMap { | ||
| ByteBuffer | ||
| .allocate(UShort.SIZE_BYTES) | ||
| .order(binaryContentsByteOrder) | ||
| .putShort(it.toShort()) | ||
| .array() | ||
| .toList() | ||
| }.toByteArray() | ||
|
|
||
| private fun InferTensorContents.toUint32Bytes(): ByteArray = this.uintContentsList | ||
| .flatMap { | ||
| ByteBuffer | ||
| .allocate(UInt.SIZE_BYTES) | ||
| .order(binaryContentsByteOrder) | ||
| .putInt(it) | ||
| .array() | ||
| .toList() | ||
| } | ||
| .toByteArray() | ||
|
|
||
| private fun InferTensorContents.toUint64Bytes(): ByteArray = this.uint64ContentsList | ||
| .flatMap { | ||
| ByteBuffer | ||
| .allocate(ULong.SIZE_BYTES) | ||
| .order(binaryContentsByteOrder) | ||
| .putLong(it) | ||
| .array() | ||
| .toList() | ||
| } | ||
| .toByteArray() | ||
|
|
||
| private fun InferTensorContents.toInt8Bytes(): ByteArray = this.intContentsList | ||
| .flatMap { | ||
| ByteBuffer | ||
| .allocate(1) | ||
| .put(it.toByte()) | ||
| .array() | ||
| .toList() | ||
| } | ||
| .toByteArray() | ||
|
|
||
| private fun InferTensorContents.toInt16Bytes(): ByteArray = this.intContentsList | ||
| .flatMap { | ||
| ByteBuffer | ||
| .allocate(Short.SIZE_BYTES) | ||
| .order(binaryContentsByteOrder) | ||
| .putShort(it.toShort()) | ||
| .array() | ||
| .toList() | ||
| } | ||
| .toByteArray() | ||
|
|
||
| private fun InferTensorContents.toInt32Bytes(): ByteArray = this.intContentsList | ||
| .flatMap { | ||
| ByteBuffer | ||
| .allocate(Int.SIZE_BYTES) | ||
| .order(binaryContentsByteOrder) | ||
| .putInt(it) | ||
| .array() | ||
| .toList() | ||
| } | ||
| .toByteArray() | ||
|
|
||
| private fun InferTensorContents.toInt64Bytes(): ByteArray = this.int64ContentsList | ||
| .flatMap { | ||
| ByteBuffer | ||
| .allocate(Long.SIZE_BYTES) | ||
| .order(binaryContentsByteOrder) | ||
| .putLong(it) | ||
| .array() | ||
| .toList() | ||
| } | ||
| .toByteArray() | ||
|
|
||
| private fun InferTensorContents.toFp32Bytes(): ByteArray = this.fp32ContentsList | ||
| .flatMap { | ||
| ByteBuffer | ||
| .allocate(Float.SIZE_BYTES) | ||
| .order(binaryContentsByteOrder) | ||
| .putFloat(it) | ||
| .array() | ||
| .toList() | ||
| } | ||
| .toByteArray() | ||
|
|
||
| private fun InferTensorContents.toFp64Bytes(): ByteArray = this.fp64ContentsList | ||
| .flatMap { | ||
| ByteBuffer | ||
| .allocate(Double.SIZE_BYTES) | ||
| .order(binaryContentsByteOrder) | ||
| .putDouble(it) | ||
| .array() | ||
| .toList() | ||
| } | ||
| .toByteArray() | ||
|
|
||
| private fun InferTensorContents.toBoolBytes(): ByteArray = this.boolContentsList | ||
| .flatMap { | ||
| ByteBuffer | ||
| .allocate(1) | ||
| .put(if (it) 1 else 0) | ||
| .array() | ||
| .toList() | ||
| } | ||
| .toByteArray() | ||
|
|
||
| private fun InferTensorContents.toRawBytes(): ByteArray = this.bytesContentsList | ||
| .flatMap { | ||
| ByteBuffer | ||
| .allocate(it.size() + Int.SIZE_BYTES) | ||
| .order(binaryContentsByteOrder) | ||
| .putInt(it.size()) | ||
| .put(it.toByteArray()) | ||
| .array() | ||
| .toList() | ||
| } | ||
| .toByteArray() |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.