Context: the project archive (KXKM_Batterie_Parallelator-main) had
no git history locally; a fresh repository is needed to host it on
git.saillant.cc (electron/KXKM_Batterie_Parallelator).

Approach: initialize a new repo on branch main, stage the archive
content, and harden .gitignore before the first commit.

Changes:
- Import the full project tree: firmware/, firmware-idf/,
  firmware-rs/, iosApp/, kxkm-bmu-app/, kxkm-api/, hardware/,
  docs/, specs/, scripts/, models/, tests/
- Keep project dotfiles tracked despite the trailing '.*' ignore
  rule: .github/, .claude/, .superpowers/, .gitattributes,
  .markdownlint.json
- Extend .gitignore: firmware/src/credentials.h (local secrets,
  template kept), kxkm-bmu-app/**/build/ (66 MB compiled iOS
  framework), .remember/ (session data)

Impact: the project can now be maintained on the self-hosted Gitea
forge with a clean, secret-free initial history.

2026-07-04 12:32:28 +02:00

2.6 KiB

Raw Permalink Blame History

Runbook Deployment KXKM-AI

Statut: operational draft
Host: kxkm@kxkm-ai
Container cible: mascarade-platformio

1. Pre-checks

SSH access must be valid: key-based auth, no interactive prompt.
Target container must exist and be running.
Dataset file must be available on host: consolidated.parquet.

2. Remote checks

scripts/ml/remote_kxkm_ai_pipeline.sh check
scripts/ml/remote_kxkm_ai_pipeline.sh discover-dataset

Expected:

container visible in docker ps
dataset either found on host or reported missing in container path

3. Bootstrap repo in container

scripts/ml/remote_kxkm_ai_pipeline.sh bootstrap-container

This syncs tracked repository files to:

/workspace/KXKM_Batterie_Parallelator

4. Inject dataset (if missing)

scripts/ml/remote_kxkm_ai_pipeline.sh inject-dataset \
  --dataset-path /abs/path/on/host/models/consolidated.parquet

5. Run pipeline in container

scripts/ml/remote_kxkm_ai_pipeline.sh run-container

Expected artifacts:

models/phase2_metrics.json
models/fpnn_soh_v2_quantized.onnx

6. Blockers and recovery

CONTAINER_DATASET_MISSING: run inject step with valid host path.
Python dependency missing in container: install dependency in container, then rerun bootstrap + run.
Any runtime error: keep logs and update plan/refactor-safety-core-web-remote-1.md with blocker and next action.

7. Evidence to store

command used
date/time
resulting artifact paths
CI link or remote session log reference

Dependency gate checked during bootstrap

pandas
numpy
torch
onnxruntime
pyarrow
onnx

Latest verified remote run

Host: kxkm@kxkm-ai
Container: mascarade-platformio
Verified at: 2026-03-30 15:50 UTC
Artifacts generated:
- models/fpnn_soh_v2_quantized.onnx (16.0 KB)
- models/phase2_metrics.json
Final verdict:
- technical unblock: yes
- quality gate pass: no (quantized_mape_degradation_le_5pp=false)

Recovery notes

torch.onnx.OnnxExporterError: Module onnx is not installed!: install onnx in the target container, then rerun quantization/finalize.
overall_gate_pass=false: infrastructure is healthy; switch to quantization remediation (calibration strategy, percentile clipping, quant format) instead of re-debugging SSH/container plumbing.

Promoted quantization baseline

The first remediation iteration passed the gate and is now the promoted remote baseline:

--quant-format qdq
--calib-strategy stratified
--calib-samples 500
--percentile-clip 1 99

Promoted result:

overall_gate_pass=true
quantized size 15.99 KB
quantized degradation +3.0446 pp

2.6 KiB Raw Permalink Blame History