Preflight

1#!/usr/bin/env bash
2# Fast, automatable subset of the G1-G10 checks in SKILL.md.
3# Output contract: every result line starts with its G-number.
4#   G1/G9 emit PASS/FAIL/WARN verdicts; G3/G4 emit INFO lines
5#   (raw readings needing judgment — see SKILL.md G3/G4). G7 emits
6#   PASS/WARN for hardware capability only — it does NOT verify
7#   the kernel target arch (sm_121a), which stays a manual step,
8#   see gotcha-checks.md G7.
9# Non-automatable gotchas (G2 flash-attn presence/Unsloth
10# override check, G6 process survey, G8 upstream-issue lookup,
11# G10 config review) are not covered here — see
12# references/gotcha-checks.md for those.
13# Usage: bash preflight.sh
14set -uo pipefail
15
16echo "== G1: CUDA 12/13 ABI =="
17g1_cuda_ver=$(python3 -c "import torch; print(torch.version.cuda or '')" 2>/dev/null)
18if [ -z "$g1_cuda_ver" ]; then
19  echo "G1 SKIP: torch not importable"
20elif [[ "$g1_cuda_ver" == 13* ]]; then
21  echo "G1 PASS: torch built against CUDA $g1_cuda_ver"
22else
23  echo "G1 FAIL: torch built against CUDA $g1_cuda_ver (need 13.x — see gotcha G1)"
24fi
25
26echo "== G3: UMA headroom (raw reading) =="
27free -g | awk 'NR==2 {print "G3 INFO: free/used (GB):", $4, $3}'
28
29echo "== G4: thermal snapshot (raw reading) =="
30nvidia-smi --query-gpu=temperature.gpu,power.draw --format=csv,noheader 2>/dev/null \
31  | sed 's/^/G4 INFO: /' || echo "G4 SKIP: nvidia-smi not available"
32
33echo "== G7: SM121 capability (kernel target arch NOT checked here) =="
34python3 -c "
35import torch
36cap = torch.cuda.get_device_capability()
37print('G7 PASS: SM121 hardware capability confirmed (partial — verify the kernel target arch (sm_121a) manually, see gotcha-checks.md G7)' if cap == (12, 1) else f'G7 WARN: unexpected capability {cap}')
38" 2>/dev/null || echo "G7 SKIP: torch not importable"
39
40echo "== G9: container vs bare pip =="
41if [ -f /.dockerenv ] || [ -f /run/.containerenv ]; then
42  echo "G9 PASS: running inside a container (Docker/Podman marker file present)"
43elif grep -qE '(docker|containerd|kubepods)' /proc/1/cgroup 2>/dev/null; then
44  echo "G9 PASS: running inside a container (cgroup marker present)"
45elif [ -r /proc/1/cgroup ]; then
46  echo "G9 UNKNOWN: no container marker matched — this does not prove a bare host (cgroup v2 and some runtimes hide the marker); verify manually before assuming bare-host, prefer an NGC or Unsloth container if unsure"
47else
48  echo "G9 SKIP: /proc/1/cgroup not readable"
49fi
50
51echo "Full details and remaining gotchas: references/gotcha-checks.md"