Denpex ML Diagnostics

Root-cause GPU/ML training crashes without leaving your IDE. Denpex is the same seven-layer diagnosis engine that powers denpex.com — this extension brings the full cloud experience into VS Code, including over Remote-SSH on your training nodes.

Features

Forensic diagnosis — select a traceback (or use a terminal buffer / clipboard) and run Denpex: Diagnose Terminal Error. You get a full forensic report: root cause, confidence calibration, causal chain, competing hypotheses, hardware telemetry findings, advisories, and an escalating fix ladder (primary → fallback → emergency → verify → prevent).

Predictive failure detection — Denpex watches your training terminals for pre-crash signals (NCCL transport retries, Xid/ECC events, loss spikes, thermal throttling) and warns you before the crash: "High risk of impending crash detected. Confidence: 84%." Detection is fully local; nothing is uploaded unless you click Diagnose Now.

Multi-node topology awareness — debugging a distributed run across several Remote-SSH terminals? Denpex: Diagnose Across All Terminals coalesces every terminal's output and sends a topology map so the engine can compute blast radius and isolate the initiating node.

Closed-loop remediation — every fix in the report has an Execute button that runs it in a dedicated terminal (with a confirmation prompt you can disable). Restart from Checkpoint re-launches your training with your configured resume command.

Hardware RMA payloads — Denpex: Generate Hardware RMA Payload pulls nvidia-smi ECC/row-remap counters and dmesg Xid events from the current machine and produces a vendor-ready JSON ticket for Dell/NVIDIA/Supermicro support.

Cloud sync — set your API key (Denpex: Set API Key) and every diagnosis lands in your denpex.com dashboard, counts against your real plan quota, and appears in your incident history. The status bar shows your plan and remaining diagnoses.

Getting started

1. Install the extension.
2. Run a few diagnoses free — no account needed.
3. When you want more: Denpex: Set API Key (create a key at denpex.com/dashboard), or Denpex: Redeem Trial Code if you have one. Upgrading happens on denpex.com/pricing — the extension picks up your new plan automatically.

Settings

Privacy

Logs are sent to the Denpex API only when you explicitly run a diagnose command. Predictive monitoring runs entirely locally. Clipboard contents are only used when you pick them from the menu.