Billy D.
bc4b230dd9
New: - vllm_tuning_pipeline.py: A/B benchmark different vLLM configs, logs latency/TPS/TTFT to MLflow (vllm-tuning experiment) - vllm_tuning_pipeline.yaml: compiled KFP YAML Updated: - voice_pipeline.py: per-step NamedTuple outputs with latency tracking, new log_pipeline_metrics MLflow component - voice_pipeline.yaml, tts_pipeline.yaml, rag_pipeline.yaml: recompiled
88 lines
3.3 KiB
YAML
88 lines
3.3 KiB
YAML
# PIPELINE DEFINITION
|
|
# Name: text-to-speech-pipeline
|
|
# Description: Simple text to speech pipeline
|
|
# Inputs:
|
|
# text: str
|
|
components:
|
|
comp-synthesize-speech:
|
|
executorLabel: exec-synthesize-speech
|
|
inputDefinitions:
|
|
parameters:
|
|
text:
|
|
parameterType: STRING
|
|
tts_url:
|
|
defaultValue: http://ai-inference-serve-svc.ai-ml.svc.cluster.local:8000/tts
|
|
isOptional: true
|
|
parameterType: STRING
|
|
outputDefinitions:
|
|
parameters:
|
|
audio_b64:
|
|
parameterType: STRING
|
|
latency_s:
|
|
parameterType: NUMBER_DOUBLE
|
|
deploymentSpec:
|
|
executors:
|
|
exec-synthesize-speech:
|
|
container:
|
|
args:
|
|
- --executor_input
|
|
- '{{$}}'
|
|
- --function_to_execute
|
|
- synthesize_speech
|
|
command:
|
|
- sh
|
|
- -c
|
|
- "\nif ! [ -x \"$(command -v pip)\" ]; then\n python3 -m ensurepip ||\
|
|
\ python3 -m ensurepip --user || apt-get install python3-pip\nfi\n\nPIP_DISABLE_PIP_VERSION_CHECK=1\
|
|
\ python3 -m pip install --quiet --no-warn-script-location 'kfp==2.12.1'\
|
|
\ '--no-deps' 'typing-extensions>=3.7.4,<5; python_version<\"3.9\"' &&\
|
|
\ python3 -m pip install --quiet --no-warn-script-location 'httpx' && \"\
|
|
$0\" \"$@\"\n"
|
|
- sh
|
|
- -ec
|
|
- 'program_path=$(mktemp -d)
|
|
|
|
|
|
printf "%s" "$0" > "$program_path/ephemeral_component.py"
|
|
|
|
_KFP_RUNTIME=true python3 -m kfp.dsl.executor_main --component_module_path "$program_path/ephemeral_component.py" "$@"
|
|
|
|
'
|
|
- "\nimport kfp\nfrom kfp import dsl\nfrom kfp.dsl import *\nfrom typing import\
|
|
\ *\n\ndef synthesize_speech(\n text: str,\n tts_url: str = \"http://ai-inference-serve-svc.ai-ml.svc.cluster.local:8000/tts\"\
|
|
\n) -> NamedTuple(\"TTSResult\", [(\"audio_b64\", str), (\"latency_s\",\
|
|
\ float)]):\n \"\"\"Convert text to speech using TTS service.\"\"\"\n\
|
|
\ import base64\n import time\n import httpx\n from collections\
|
|
\ import namedtuple\n\n start = time.perf_counter()\n with httpx.Client(timeout=120.0)\
|
|
\ as client:\n response = client.post(\n f\"{tts_url}/v1/audio/speech\"\
|
|
,\n json={\n \"input\": text,\n \
|
|
\ \"voice\": \"en_US-lessac-high\",\n \"response_format\"\
|
|
: \"wav\"\n }\n )\n audio_b64 = base64.b64encode(response.content).decode(\"\
|
|
utf-8\")\n latency = time.perf_counter() - start\n\n TTSResult = namedtuple(\"\
|
|
TTSResult\", [\"audio_b64\", \"latency_s\"])\n return TTSResult(audio_b64,\
|
|
\ latency)\n\n"
|
|
image: python:3.13-slim
|
|
pipelineInfo:
|
|
description: Simple text to speech pipeline
|
|
name: text-to-speech-pipeline
|
|
root:
|
|
dag:
|
|
tasks:
|
|
synthesize-speech:
|
|
cachingOptions:
|
|
enableCache: true
|
|
componentRef:
|
|
name: comp-synthesize-speech
|
|
inputs:
|
|
parameters:
|
|
text:
|
|
componentInputParameter: text
|
|
taskInfo:
|
|
name: synthesize-speech
|
|
inputDefinitions:
|
|
parameters:
|
|
text:
|
|
parameterType: STRING
|
|
schemaVersion: 2.1.0
|
|
sdkVersion: kfp-2.12.1
|