1 Lab Dashboard

Claude edited this page 2026-02-23 19:41:13 +00:00

Table of Contents

• Lab Dashboard
• Access
• Architecture
• InfluxDB
• Measurements
• Go Application Structure
• Dashboard Views
• Seeding InfluxDB

Lab Dashboard

Live monitoring of LEM training and generation progress.

Access

• URL: lab.lthn.io
• Source: ~/infrastructure/lab/ (Go application)

Architecture

lem_generate.py / core ml → InfluxDB → Lab Go collector → HTML dashboard

InfluxDB

• URL: http://10.69.69.165:8181 (snider-linux)
• Database: training

Measurements

Measurement	Source	Tags	Fields
gold_gen	lem_generate.py	i, w, d, v	seed_id, gen_time, chars
golden_gen_progress	lem_generate.py	worker	completed, target, pct
golden_set_stats	pipeline.py metrics	—	total_examples, domains, voices, avg_gen_time
expansion_gen	lem_expand.py	i, w, d, r	seed_id, gen_time, chars, model
expansion_progress	lem_expand.py	worker	completed, target, pct
training_loss	lem_train_15k.py	—	train_loss, val_loss, iteration, learning_rate
training_status	lem_train_15k.py	—	model, status (running/complete/failed), total_iters

Go Application Structure

File	Purpose
internal/handler/templates/golden-set.html	Golden set dashboard page
internal/collector/influxdb.go	Queries InfluxDB for gold_gen and golden_set_stats
internal/model/model.go	GoldenSetSummary, DomainStat, VoiceStat, WorkerStat types

Dashboard Views

• Golden set progress — completion percentage, generation rate, ETA
• Domain breakdown — examples per domain
• Voice breakdown — examples per voice variant
• Worker stats — per-worker generation rate and status
• Training curves — loss over iterations (when training active)

Seeding InfluxDB

If InfluxDB data needs to be rebuilt from DuckDB:

# Seed from DuckDB golden_set
core ml seed-influx --force

# Push current stats
core ml metrics