prod
GT
NameVersionTypeStatus Eval scoreSizePushed
order-classifier v3.1 model production 94.2% 2.3 GB2026-05-22
order-classifier v3.0 model staging 91.7%2.3 GB2026-05-18
order-classifier v2.9 model stable 89.1%2.1 GB2026-04-30
customer-embeddings v1.2 model stable 780 MB2026-05-10
orders-2024-Q4 v2 dataset active 48 GB2026-04-20
bpe-tokenizer v1 tokenizer stable 4.2 MB2026-03-15
inference-config v1.3 config active 2 KB2026-05-22

Showing 7 of 23 artifacts

Type
model
safetensors · 2.3 GB
Eval score
94.2%
↑ 2.5% from v3.0
Hallucination rate
1.4%
↓ 0.7% from v3.0
Pushed by
@gyt
2026-05-22 · 14:22
Lineage
DATASET orders-2024-Q4 v2 · 48 GB bpe-tokenizer v1 TRAINING RUN run-2026-05-22 lr=2e-5 · epochs=3 · 8×A100 MODEL order-classifier v3.1 · 94.2% ● production
Version history
VersionStatusEvalHallucinationPushed
v3.1 production 94.2%1.4%2026-05-22
v3.0 staging 91.7%2.1%2026-05-18
v2.9 stable 89.1%3.0%2026-04-30
Baseline
order-classifier v3.1
production · pushed 2026-05-22
vs
Candidate
order-classifier v3.2
candidate · pushed 2026-05-23
Metricv3.1 (baseline)v3.2 (candidate)Δ
Accuracy94.2%95.8%↑ +1.6%
F1 Score0.9310.948↑ +0.017
Latency p50131 ms127 ms↑ −4 ms
Latency p99352 ms344 ms↑ −8 ms
Hallucination rate1.4%1.1%↑ −0.3%
Cost / 1K tokens$0.0041$0.0044↓ +$0.0003
Eval suite runtime4m 12s4m 08s
Policy gates — v3.2
Accuracy > 93%
Latency p99 < 400 ms
Hallucination < 1.5%
Cost delta < 20%
✓ All 4 gates passed
Pushed
14:22
Validated
14:23
Eval passed
14:25 · 4/4
Staging
15:30
Production
16:48
14:22pushed by @gyt — order-classifier v3.1 (2.3 GB, safetensors)
14:23validation passedschema OK · format OK · size within limit · no unsafe tensors detected
14:25eval suite: 4/4 gates passedaccuracy 94.2% ✓ · latency p99 352ms ✓ · hallucination 1.4% ✓ · cost delta 0% ✓
15:30promoted to staging by @gyt
16:15QA sign-off: approved by @qa-lead
16:48deployed to production · 0 downtime · previous v3.0 archived
Current production: order-classifier v3.1 · deployed 2026-05-22 16:48 ·
Latency p50
131 ms
↓ 3 ms from yesterday
Drift score
0.04
Normal · threshold 0.15
Hallucination rate
1.4%
↓ 0.1% from last week
Requests / day
48,230
↑ 12% from last week
Output drift — 30 days
warn 0.15 0.08 0.00
Recent alerts
No critical alerts in the last 7 days
Latency spike detected — p99 spike to 580 ms 2026-05-19 14:00–15:30 · resolved
Model deployed: order-classifier v3.1 → production 2026-05-22 16:48 · auto
Month total
$2,847
Budget: $4,000 · 71% used
$2,847 spent$1,153 remaining
Projected end-of-month
$3,713
↓ $287 under budget
Avg cost / 1K tokens
$0.0038
↓ 7% from last month
Cost by model — May 2026
ModelRequestsTotal tokensCostCost / 1K tokensShare
order-classifier v3.11.34M336M $1,943$0.0041
68%
customer-embeddings v1.2348K89M $642$0.0020
23%
Infrastructure (storage + egress) $262
9%
Moyrin — UI design specification · v0.1