Ultra-Lightweight Multilingual LLM Series

BAAR2-150M
Compact. Real-Time.

Engineered with 150M parameters to deliver sub-second, real-time multilingual text generation on edge hardware and budget GPUs without compute bottlenecks.

View Benchmarks BAAR Visit baar.uk
Languages: KO EN JA ZH ES PT DE RU HI VI TH
MODEL PORTFOLIO

BAAR2 Model Architecture Lineup

Tailored lightweight model family designed from ultra-edge IoT to high-density real-time servers.

ULTRA-EDGE / IOT Private · Coming Soon

BAAR2-130M

Extreme compact micro-model designed for embedded IoT hardware and tight micro-controllers.

130M Parameters Status: In Training
CURRENT
FOUNDATION / BALANCED Active

BAAR2-150M

Current active release with 11-language coverage, achieving 300~600ms latency on a single Tesla T4.

150M Parameters Ready for Inference
EXTENDED REASONING Private · Coming Soon

BAAR2-300M

Engineered for dense reasoning, corporate RAG alignment, and advanced instruction-following capabilities.

300M Parameters Status: In Evaluation
EMPIRICAL BENCHMARKS

Live Inference Benchmark Logs (150M)

HARDWARE: SINGLE TESLA T4 (FP16)
Task / Category Prompt Generated Output Latency
Korean General QA 대한민국의 수도는 어디인가요? 대한민국의 수도는 서울입니다. 개인적으로, 이 지역의 수도가 너 ~1451.0ms
English General QA What is artificial intelligence? A comprehensive machine that is designed to perform tasks that normally require ~671.2ms
KO-to-EN Translation (1) 안녕이 영어로 뭐야? 안녕은 영어로 Hello 입니다. ~390.0ms
KO-to-EN Translation (2) 사과가 영어로 뭐야? 사과는 영어로 apple 입니다. ~359.4ms
EN-to-KO Translation What is 'Seoul' in Korean? A company that is a company that operates in the United ~571.1ms
Japanese General QA 日本の首都はどこですか？ 日本の首都は、 東京都双方にあり ~661.0ms
This model is a pre-trained base checkpoint. For strict factual alignment or specific domain tasks, SFT or RAG is recommended.
LICENSING TERMS

License & Commercial Adoption

Non-Commercial & Educational Only Strictly restricted to academic and personal research. Any commercial monetization without a license is prohibited.

Commercial & Enterprise Licensing To integrate into paid products, monetize via APIs, or deploy in enterprise workflows, please request a commercial license.

ENTERPRISE INQUIRIES

Request Commercial Terms

Share your deployment target (130M, 150M, 300M) and volume requirements for custom enterprise terms.

Send Inquiry
CAREERS & PARTNERSHIPS

Opportunities & Technical Collaboration

Demonstrating proven practical expertise in custom lightweight architecture scaling from 130M to 300M under tight compute limits. Actively open to AI Engineering / Research positions and venture investment/partnerships.

Send Offer BAAR Official Site (baar.uk)
01 / ADOPTION Model Licensing [email protected]
02 / CAREERS Hiring & Roles [email protected]
03 / VENTURE Investment & Collab [email protected]