How much accuracy did the distilled model retain compared to the teacher model?

The distilled Nova Lite student model achieved 95.085 percent F1-Score overall, retaining 98 percent of the teacher model's performance. The teacher Nova Pro model scored 97.0 percent overall, with English at 97.8 percent and Japanese at 96.2 percent.

What was the training process and timeline?

IBS Software's team of nine researchers and engineers spent approximately 4 months. The student model was trained for 4 epochs over 70 steps, reducing loss from 0.05 to 0.008, using 500 manually annotated cargo email messages (350 English, 150 Japanese) with 23 entity types.

What entities does the system extract from cargo emails?

The system identifies 23 entity types including air waybill (AWB) numbers, flight numbers and routes, weights (gross, chargeable, dimensional), dimensions and volume, commodity descriptions, shipper and consignee information, special handling codes, and delivery instructions.

Back to articles

IBS Software cuts cargo AI inference costs by 14× with Amazon Bedrock distillation

Amazon AI Blog14h ago5 min read

Key takeaway

IBS Software deployed a bilingual cargo logistics system using Amazon Bedrock's model distillation to compress a teacher model into a smaller student model, achieving 95.085 percent accuracy on named-entity recognition while cutting inference costs by 14×. The approach retained 98 percent of the teacher model's performance and now processes cargo emails in real time, demonstrating that managed distillation can balance cost and accuracy for production multilingual systems.

Summaries like this, in your inbox every morning.

3 Key Points

What happened
IBS Software built a bilingual named-entity recognition system for cargo logistics email processing in English and Japanese, using Amazon Bedrock's model distillation to compress Amazon Nova Pro into the smaller Nova Lite model. The distilled Nova Lite achieved 95.085 percent F1-Score accuracy while reducing production inference costs by 14×.
Why it matters
Cargo logistics companies handle thousands of bilingual messages daily, extracting critical data like air waybill numbers, flight details, and weights. The traditional trade-off between accuracy and cost made this difficult; IBS Software's approach shows that managed distillation can deliver near-teacher performance (retaining 98 percent of the teacher's accuracy) at a fraction of the operational expense, potentially enabling similar multilingual NER deployments across industries that need cost-effective, real-time processing.
What to watch
The system identified 23 entity types across two languages and showed a 2.565 percent accuracy gap on Japanese text versus English, primarily due to complex kanji combinations and smaller Japanese training data (150 versus 350 email messages). IBS Software addressed this by augmenting Japanese training data with synthetic examples and applying post-processing rules for known patterns.

FAQ

How much accuracy did the distilled model retain compared to the teacher model?: The distilled Nova Lite student model achieved 95.085 percent F1-Score overall, retaining 98 percent of the teacher model's performance. The teacher Nova Pro model scored 97.0 percent overall, with English at 97.8 percent and Japanese at 96.2 percent.
What was the training process and timeline?: IBS Software's team of nine researchers and engineers spent approximately 4 months. The student model was trained for 4 epochs over 70 steps, reducing loss from 0.05 to 0.008, using 500 manually annotated cargo email messages (350 English, 150 Japanese) with 23 entity types.
What entities does the system extract from cargo emails?: The system identifies 23 entity types including air waybill (AWB) numbers, flight numbers and routes, weights (gross, chargeable, dimensional), dimensions and volume, commodity descriptions, shipper and consignee information, special handling codes, and delivery instructions.

Discussion

No discussion yet for this article

Stay ahead with AI news

Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.

Get Started Free

Free · takes 30 seconds · unsubscribe anytime

1 minute a day. The AI essentials.

200+ sources · Email / LINE / Slack

Get it free →