trainproof 0.5.0

A deterministic linter for ML training runs: dataset, tokenizer, and epoch logs.

trainproof

A deterministic linter for ML training runs. Point it at your dataset, your tokenizer, or your first-epoch logs — it returns a PASS / WARN / FAIL verdict with named findings and cited evidence, before you burn days of GPU time on a run that was doomed at step 50.

No ML judging ML. No invented "confidence 97%". Every rule is a deterministic threshold in one auditable module, and every finding cites the numbers that triggered it.

What it is: the reliability layer between your training code and the GPU bill — checks that run before training (is this run safe to start?), during training (should it keep going?), and after (is it reproducible; did it match a known-good baseline?).

The one rule it never breaks: trainproof does not infer causes, does not invent confidence scores, and does not guess. It reports deterministic findings backed by evidence, or it stays silent. Every feature earns its place by answering a single question — if it were gone, would someone lose GPU hours?

pip install trainproof

See a verdict in 60 seconds

This repo ships the real logs of five QLoRA fine-tuning runs (Qwen2.5-3B, RTX 5080 — see the gallery below). Judge one right now:

trainproof epoch examples/gallery/lr_hot/trainer_state.json --format hf

[FAIL] Critical checks failed:
  [FAIL] Loss curve is diverging.
         Evidence: End loss 7.492 vs Min loss 1.398
  [WARN] Gradient norm spikes detected.
         Evidence: Max gn 2649.75 > 10.0x median (0.55)

Why this exists

The author lost a real 11-hour fine-tune to a failure nothing warned about. Pointed retroactively at that run's 1MB Coqui Trainer log (2,501 logged steps), trainproof's verdict: FAIL — diverging. The loss reached its minimum at 82% of the run and ended 1.9x above it. Translation: the final two hours of GPU time made the model measurably worse, and the checkpoint worth keeping had already existed for hours. No tool in the stack said a word.

The fault-injection gallery

To validate the rules, the same QLoRA fine-tune (Qwen2.5-3B-Instruct, 4-bit, LoRA r=16, 300 steps on Alpaca-cleaned) was run five times — once healthy, four times with exactly one knob deliberately broken. Real runs, real logs, all shipped in examples/gallery/:

Run	Sabotage	Verdict	Key evidence
healthy	none	PASS	loss 1.52 → 0.94, stable gradients
lr_hot	LR x100 (2e-2)	FAIL	diverging: end 7.49 vs min 1.40; grad spike 2650 vs median 0.55
lr_zero	LR = 0	FAIL	dead run: first-5 median 1.52 vs last-5 1.49 (<5% improvement); lr=0 on 100% of steps
fp16_nan	fp16 + hot LR, no clipping	FAIL	diverging: end 7.21 vs min 1.09 (grad scaling absorbed the intended NaN — the run diverged instead; reported as observed)
bad_labels	labels shuffled per-sequence	WARN only (single-run) — caught by trainproof compare (v0.3)	grad spike 23.3 vs median 1.09

The honest finding: loss curves cannot see corrupted data

The bad_labels run — whose shuffled labels make real learning impossible — reduced its loss by 62% (18.9 → 5.75). The model was genuinely learning: not the task, but the marginal token statistics of the garbage. From its own loss curve, that is indistinguishable from healthy training (neural networks famously fit random labels). No single-run, loss-only rule can catch this class of failure — its real signature is relative: a loss floor ~6x higher than a known-good run of the same task (5.59 vs 0.94).

That finding produced v0.3: trainproof compare <run> <baseline> — deterministic ratio rules against the healthy baseline you already have — which catches bad_labels at a 6x loss-floor ratio, in 3 seeds out of 3. The full study was repeated with three random seeds (15 runs): see EVIDENCE_MATRIX.md for every verdict, including the honest miss (compare alone overlooks one lr_zero seed — the single-run zero-LR fatality rule owns that case; the two commands cover each other's blind spots). The gallery also improved the tool itself twice: the dead-run rule and the total-zero-LR fatality rule both exist because runs escaped earlier rule versions. See ROADMAP.md.

The commands

# 1. Dataset preflight (speech/TTS pack): audio integrity, transcript quality,
#    duplicates, text-vs-audio duration mismatches
trainproof data /path/to/dataset_or_manifest.jsonl

# 2. Tokenizer preflight: vocabulary coverage, OOV rate, sequence blowouts,
#    suspicious splits on numbers/dates
trainproof tokenizer my_tokenizer.model transcripts.txt

# 3. Training-run verdict: NaN/divergence/dead-run detection, gradient spikes,
#    LR sanity, throughput — from log files, any framework
trainproof epoch logs/run.jsonl            # exit code 1 on FAIL: CI-ready

# 4. Compare against a baseline
#    Catch relative pathologies like the `bad_labels` run that evade single-run rules.
trainproof compare examples/gallery/bad_labels/trainer_state.json examples/gallery/healthy/trainer_state.json

========================================
TRAINPROOF VERDICT
========================================
[FAIL] Critical checks failed:
  [FAIL] loss floor ratio exceeded limit
         Evidence: Run floor 5.592 vs Baseline floor 0.937 (ratio 6.0x > 2.0)
  [FAIL] end loss ratio exceeded limit
         Evidence: Run end 5.750 vs Baseline end 1.082 (ratio 5.3x > 2.0)
========================================

Each command prints the verdict, writes a self-contained HTML report, and sets the process exit code — so it works as a CI gate out of the box.

Live guardian (v0.4)

Don't wait for the post-mortem — catch a doomed run while it is still burning GPU. Add one line to a HuggingFace Trainer:

from transformers import Trainer
from trainproof.integrations.hf import TrainproofCallback

trainer = Trainer(
    ...,
    callbacks=[TrainproofCallback(policy="stop_on_fail")],  # or policy="warn"
)

Run against a real diverging QLoRA fine-tune (learning rate 100x too high), the guardian aborts it 20 steps into a 300-step schedule — on its own:

{'loss': '1.784', 'grad_norm': '9.634',  'learning_rate': '0.007'}
{'loss': '4.282', 'grad_norm': '53.76',  'learning_rate': '0.009'}
{'loss': '10.6',  'grad_norm': '13.34',  'learning_rate': '0.011'}
{'loss': '31.67', 'grad_norm': '76.67',  'learning_rate': '0.013'}
...
TRAINPROOF ABORT - stopping training at step 20. Findings:
  [FAIL] Loss curve is diverging.
         Evidence: End loss 22.952 vs Min loss 1.358
  [FAIL] Loss never improved over the run (dead run).
         Evidence: median of first 5 losses 1.502 vs last 5 22.952

  scheduled steps : 300
  stopped at step : 20
  run saved       : 93% of the scheduled steps never ran

On a two-day pre-training run, that fraction is days of GPU time. Or watch a growing log file from outside the process (CI-friendly, exits non-zero on FAIL):

trainproof watch logs/run.jsonl --interval 10 --until-fail
# [21:37:44] warming up (5 records)
# [21:37:44] n_records=15 verdict=PASS findings=1

The default is safe. policy="warn" (the default) only observes and reports — it never interrupts your run, so you can leave it on even for experiments you expect to fail. Aborting is strictly opt-in via policy="stop_on_fail", the one mode that takes an irreversible action. trainproof does not make that decision for you unless you ask.

The guardian applies the same deterministic rules as trainproof epoch, so it inherits their documented single-run limitations.

Pre-flight (v0.5): stop a run before it starts

The guardian saves most of a doomed run; preflight saves 100% because it never starts. Catch broken datasets and tokenizer misconfigurations instantly.

trainproof preflight data/dataset.jsonl --tokenizer mistralai/Mistral-7B-v0.1 --max-len 4096

========================================
TRAINPROOF VERDICT
========================================
[FAIL] Critical checks failed:
  [FAIL] Empty or whitespace-only text found.
         Evidence: 1 records (indices [1]...)
========================================

Checks: malformed JSONL, empty text, exact duplicate text, tokenizer structural checks (EOS/PAD/BOS), and context length overflows.

Supported log formats

• HuggingFace Trainer (trainer_state.json)
• Coqui Trainer text logs (ANSI-colored trainer_0_log.txt)
• Generic JSONL / CSV (columns: step, loss, lr, grad_norm, time — all optional)

Auto-detected; override with --format hf|coqui|jsonl|csv. TensorBoard event files are planned (Lightning console captures are TTY dumps, not logs, and will not be supported).

Philosophy

1. Deterministic. A rule fires or it doesn't. Thresholds live in one module, commented, tunable.
2. Evidence-cited. Every finding names the steps and values that triggered it.
3. Honest about limits. What the tool cannot detect is documented in the README, not discovered by the user in production.

Family

trainproof judges training runs. Its sibling ttsproof judges TTS model outputs (structural audio checks, equivalence-aware WER/CER, published method with DOI) — and trainproof builds on it for the speech dataset checks.

Author

Panagiotis (Panos) Gkilis — portfolio · bedvibe.studio

MIT license.