Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

feat(dataflow): Use raw contents in joins if any message uses raw contents #4781

Merged
merged 4 commits into from
Apr 5, 2023
Merged

feat(dataflow): Use raw contents in joins if any message uses raw contents #4781

Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
f51ee3f
Fix raw contents in joins
ukclivecox Apr 4, 2023
1cc0236
Review fixes
ukclivecox Apr 4, 2023
a5ddf6a
fp16 exception and simplified builder
ukclivecox Apr 5, 2023
4a7be44
revert fp16 handling of content types
ukclivecox Apr 5, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 9 additions & 4 deletions scheduler/data-flow/src/main/kotlin/io/seldon/dataflow/kafka/protocol.kt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@ package io.seldon.dataflow.kafka

import com.google.protobuf.kotlin.toByteString
import io.seldon.mlops.inference.v2.V2Dataplane
import java.lang.UnsupportedOperationException
import java.nio.ByteBuffer
import java.nio.ByteOrder

Expand Down Expand Up @@ -100,7 +101,9 @@ fun convertRequestToRawInputContents(request: V2Dataplane.ModelInferRequest): V2
.toList()
}.toByteArray()
}
DataType.FP16, // as data is stored as FP32 we treat the same. Its unclear if Triton would handdle this correctly though
DataType.FP16 -> { // as data is stored as FP32 we treat the same. Its unclear if Triton would handle this correctly though
agrski marked this conversation as resolved.
Show resolved Hide resolved
throw UnsupportedOperationException("FP16 is not presently supported in Pipeline joins")
agrski marked this conversation as resolved.
Show resolved Hide resolved
}
DataType.FP32 -> {
input.contents.fp32ContentsList.flatMap {
ByteBuffer
Expand Down Expand Up @@ -136,7 +139,7 @@ fun convertRequestToRawInputContents(request: V2Dataplane.ModelInferRequest): V2
// Add raw contents
builder.addRawInputContents(v.toByteString())
// Clear the contents now we have added the raw inputs
builder.setInputs(idx, input.toBuilder().clearContents())
builder.getInputsBuilder(idx).clearContents()
}
return builder.build()
}
Expand Down Expand Up @@ -230,7 +233,9 @@ fun convertResponseToRawOutputContents(request: V2Dataplane.ModelInferResponse):
.array().toList()
}.toByteArray()
}
DataType.FP16, // Unclear if this is correct as will be stored as 4 byte floats so Triton may not handle this
DataType.FP16 -> { // as data is stored as FP32 we treat the same. Its unclear if Triton would handle this correctly though
throw UnsupportedOperationException("FP16 is not presently supported in Pipeline joins")
}
DataType.FP32 -> {
output.contents.fp32ContentsList.flatMap {
ByteBuffer
Expand Down Expand Up @@ -266,7 +271,7 @@ fun convertResponseToRawOutputContents(request: V2Dataplane.ModelInferResponse):
// Add raw contents
builder.addRawOutputContents(v.toByteString())
// Clear the contents now we have added the raw outputs
builder.setOutputs(idx, output.toBuilder().clearContents())
builder.getOutputsBuilder(idx).clearContents()
}
return builder.build()
}
Expand Down