Parent Epic

#1326

Summary

Create a private Brev launchable for GPU-based E2E tests (Ollama local inference, vLLM). Extends Flavor 1 with NVIDIA Container Toolkit and Ollama pre-installed.

What It Adds Over Flavor 1 (#1327)

• NVIDIA Container Toolkit installed and configured
• Docker runtime configured for GPU passthrough
• Ollama pre-installed
• Small test model pre-pulled (e.g., qwen3:0.6b for fast validation)
• GPU compute instance (cheapest available — L4 or T4)

Launchable Settings

• Code: Git repo (NVIDIA/NemoClaw)
• Compute: GPU (L4 or T4, cheapest available)
• Ports: None (CI uses SSH)
• Visibility: Private (Nemoclaw CI/CD org)

Deliverables

1. scripts/brev-launchable-ci-gpu.sh — startup script extending brev-launchable-ci-cpu.sh
2. Brev launchable created via form
3. Integration with nightly-e2e.yaml gpu-e2e job

Depends On

• ci: create Flavor 1 Brev launchable — CI-Ready CPU #1327 (Flavor 1 validated first)