activeCycle 2026.1
Automated Experiment Lineage & GPU Telemetry Testbed
Building lightweight GitHub Actions runners that automatically capture VRAM spikes, seed integrity, and model hash lineage per pull request.
POD 06
Systems, HPC & MLOps
Engineering research infrastructure, automated reproducibility testbeds, and distributed GPU profiling tools for lab members. We enforce reproducible experiment tracking where model weights, seed randomness, hardware memory bounds, and code revisions are immutably logged for every pull request.
Core Research Questions
- What minimal container telemetry capture is required to guarantee bit-level replication of CUDA kernels?
- How can multi-tenant GPU memory fragmentation be minimized during automated test execution?
- What network bandwidth thresholds govern model partition splits across heterogeneous edge-server environments?
Scientific Methodology
Prometheus GPU scrape exporters and VRAM peak tracking
Automated container build runners with pinned dependency hashes
Heterogeneous RPC model layer offload benchmarking
Immutable experiment artifact hashing
proposedCycle 2026.2
Heterogeneous Edge-to-Server Model Offloading Benchmarks
Profiling network latency thresholds for splitting neural pipeline layers between mobile nodes and local GPU servers over varying bandwidth constraints.
open source2026-02-10
AIAC MLOps Experiment Runner Template & GitHub Actions Suite
Turnkey CI/CD GitHub Actions workflow for machine learning repositories that captures GPU VRAM usage via Prometheus exporters, validates deterministic random seeds, and computes dataset SHA-256 hashes on every commit.