Lucifergene

"""
ASR E2E Integration Tests — Playwright

Comprehensive end-to-end tests for the Audio Speech Recognition (ASR) feature
in the Gen AI Studio Playground. Tests client-side validation, happy path
transcription, cancellation flows, BFF error scenarios, UI gating, message
bar sync, settings interaction, and reset/new chat behavior.

Prerequisites:
  1. OGX/LlamaStack port-forward:  oc port-forward -n <ns> svc/lsd-genai-playground-service 8321:8321

Lucifergene

Multimodal Vision POC — Findings Report

SPIKE: RHOAIENG-62272 (BFF server-side temporary file storage) Date: 2026-05-19 Cluster: crimson-rhoai.h6fk.p3.openshiftapps.com Namespace: avik-gpu-test

Executive Summary

Chris/Eder's proposed architecture — a stateless BFF proxying file operations to the OGX Files API — has been validated end-to-end on a live GPU cluster with a real vision model. The POC proves that:

Lucifergene

Architecture Office Hours Proposal: BFF Server-Side Temporary File Storage

SPIKE: RHOAIENG-62272

Feature: RHAISTRAT-1527 — Multimodal Support in Gen AI Studio Playground

Author: Avik Kundu

Date: 2026-05-13

Lucifergene

GPU ASR Deployment: Whisper on vLLM 0.18 (RHOAI 3.4, A10G)

Cluster: crimson-rhoai (RHOAI 3.4.0, OpenShift 4.19, NVIDIA A10G 24GB) Image: quay.io/aipcc/rhaiis/cuda-ubi9:3.4.0-1777444689 (vLLM 0.18.0+rhaiv.7) Model: registry.redhat.io/rhelai1/modelcar-whisper-large-v3-turbo-quantized-w4a16 (W4A16 quantized) Endpoint: /v1/audio/transcriptions (OpenAI-compatible) Deployment: KServe ServingRuntime + InferenceService with OCI modelcar sidecar

Performance

Lucifergene

ASR Deployment Validated: Whisper on vLLM (RHOAI 3.4 CPU)

Date: 2026-05-05 Cluster: RHOAI 3.3.1 (amd64 / m5.2xlarge nodes) Image: quay.io/aipcc/rhaiis/cpu-ubi9:3.4.0-1777444682 (vLLM 0.18.0+rhaiv.7) Model: openai/whisper-small Endpoint: /v1/audio/transcriptions (OpenAI-compatible) Deployment Method: KServe ServingRuntime + InferenceService (validated)

Key Findings

Lucifergene

apiVersion: v1
kind: Namespace
metadata:
  name: milvus
---
#
# --- etcd (metadata store) ---
#
apiVersion: v1
kind: Service

Lucifergene

kind: ConfigMap
apiVersion: v1
metadata:
  name: llama-stack-config
  namespace: avik-test
  uid: 2840f5a5-1bdf-4250-a1e7-8c0339f6564d
  resourceVersion: '18684102'
  creationTimestamp: '2026-02-17T12:47:00Z'
  labels:
    llamastack.io/distribution: lsd-genai-playground

Lucifergene

[
  {
    "paths": 20,
    "hours": 7,
    "minutes": 40,
    "description": "In this learning path, I'll demonstrate how anyone can run a distributed AI workload on Red Hat OpenShift using just a few nodes and GPUs. ",
    "label": "RoCE multi-node AI training on Red Hat OpenShift",
    "url": "https://developers.redhat.com/learn/openshift/roce-multi-node-ai-training-red-hat-openshift"
  },
  {

Lucifergene

apiVersion: backstage.io/v1alpha1
kind: User
metadata:
  name: lucifergene
spec:
  profile:
    # Intentional no displayName for testing
    email: avikkundu8d@gmail.com
  memberOf: [temp-team-a]

Lucifergene

---
#BEFORE TASK
apiVersion: tekton.dev/v1beta1
kind: Task
metadata:
  name: before
spec:
  steps:
    - image: quay.io/avik6028/busybox:latest
      name: before