BAAR Logo

BAAR2-3M

๐ŸŒ Official Website: https://baar.uk/model/
๐Ÿ“„ License: Non-Commercial Only (Contact for Commercial Pricing)


๐Ÿ“Œ Model Overview

BAAR2-3M is an ultra-compact multilingual conversational model designed to enable real-time chat under minimal hardware conditions. With just 3 million (3M) parameters, it brings lightweight, interactive dialogue capabilities across 11 languages to extreme resource-constrained environments, embedded devices, and on-device platforms where standard LLMs cannot operate.

BAAR2-3M์€ ์ตœ์†Œํ•œ์˜ ํ•˜๋“œ์›จ์–ด ์—ฌ๊ฑด๊ณผ ๊ทน๋‹จ์ ์ธ ๋ฆฌ์†Œ์Šค ์ œ์•ฝ ํ™˜๊ฒฝ์—์„œ๋„ 11๊ฐœ ์–ธ์–ด๋กœ ์‹ค์‹œ๊ฐ„ ๋Œ€ํ™”(Chat)๊ฐ€ ๊ฐ€๋Šฅํ•˜๋„๋ก ์„ค๊ณ„๋œ 300๋งŒ(3M) ํŒŒ๋ผ๋ฏธํ„ฐ ๊ทœ๋ชจ์˜ ์ดˆ๊ฒฝ๋Ÿ‰ ๋‹ค๊ตญ์–ด ์–ธ์–ด ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค. ์ผ๋ฐ˜์ ์ธ ๋Œ€ํ˜• ๋ชจ๋ธ ๊ตฌ๋™์ด ๋ถˆ๊ฐ€๋Šฅํ•œ ์ดˆ์ €์‚ฌ์–‘ ์—ฃ์ง€ ๋ฐ ์ž„๋ฒ ๋””๋“œ ๊ธฐ๊ธฐ์—์„œ๋„ ์ฆ‰๊ฐ์ ์ธ ๋Œ€ํ™”ํ˜• ์ถ”๋ก ๊ณผ ์ดˆ๊ฒฝ๋Ÿ‰ ์‹ค์‹œ๊ฐ„ RAG ์‹œ์Šคํ…œ ๊ตฌ์ถ•์„ ์ง€์›ํ•ฉ๋‹ˆ๋‹ค.


๐Ÿ“Š Live Sample Inference

Hardware: Single NVIDIA Tesla T4 | FP16 Inference

Task / Category Prompt Generated Output Latency
ํ•œ๊ตญ์–ด ์ผ๋ฐ˜ QA ๋Œ€ํ•œ๋ฏผ๊ตญ์˜ ์ˆ˜๋„๋Š” ์–ด๋””์ธ๊ฐ€์š”? ๋Œ€ํ•œ๋ฏผ๊ตญ์˜ ์ˆ˜๋„๋Š” ์—ฌ๋Ÿฌ ๊ฐ€์ง€ ์ฃผ์š” ์š”์†Œ๋ฅผ ๋งํ•˜๊ณ  ~298.8ms
์˜์–ด ์ผ๋ฐ˜ QA What is artificial intelligence? It is a complex region of the subject in the ~283.0ms
์ผ์ƒ์–ด ๋ฒˆ์—ญ ์š”์ฒญ ์•ˆ๋…•์ด ์˜์–ด๋กœ ๋ญ์•ผ? ์•ˆ๋…•์€ ์˜์–ด๋กœ Hello ์ž…๋‹ˆ๋‹ค. ~215.0ms
์ผ์ƒ์–ด ๋ฒˆ์—ญ ์š”์ฒญ 2 ์‚ฌ๊ณผ๊ฐ€ ์˜์–ด๋กœ ๋ญ์•ผ? ์‚ฌ๊ณผ๋Š” ์˜์–ด๋กœ apple ์ž…๋‹ˆ๋‹ค. ~212.2ms
์˜->ํ•œ ๋ฒˆ์—ญ ์š”์ฒญ What is 'Seoul' in Korean? A monument in the first time of the park in the U ~275.6ms
์ผ๋ณธ์–ด ์ผ๋ฐ˜ QA ๆ—ฅๆœฌใฎ้ฆ–้ƒฝใฏใฉใ“ใงใ™ใ‹๏ผŸ ๆ—ฅๆœฌใฎ้ฆ–้ƒฝใฏใ€ ๆ—ฅๆœฌใฎ้ฆ–้ƒฝใงใ‚ใ‚Š ~299.2ms
์ค‘๊ตญ์–ด ์ผ๋ฐ˜ QA ไบบๅทฅๆ™บ่ƒฝ็š„ไธป่ฆ็‰น็‚นๆ˜ฏไป€ไนˆ๏ผŸ ไบบๅทฅๆ™บ่ƒฝ็š„ไธป่ฆ็‰น็‚นๆ˜ฏไธ€็งๅ…ฌๅธ ~285.6ms
์ŠคํŽ˜์ธ์–ด ์ผ๋ฐ˜ QA ยฟCuรกl es la capital de Espaรฑa? Espaรฑa de Espaรฑa de Espaรฑa de Espaรฑa de ~286.0ms
๋Ÿฌ์‹œ์•„์–ด ์ผ๋ฐ˜ QA ะšะฐะบะฐั ัั‚ะพะปะธั†ะฐ ะ ะพััะธะธ? ะ ะพััะธะธ ะ ะพััะธะธ โ€” ัั‚ะพ ะฟั€ะตะดะปะฐะณะฐัŽั‰ะธะน ะฟั€ะตะดะป ~270.6ms
ํžŒ๋””์–ด ์ผ๋ฐ˜ QA เคญเคพเคฐเคค เค•เฅ€ เคฐเคพเคœเคงเคพเคจเฅ€ เค•เฅเคฏเคพ เคนเฅˆ? เคญเคพเคฐเคค เค•เฅ€ เคฐเคพเคœเคงเคพเคจเฅ€ เค•เฅเคฏเคพ เคนเฅˆ เคœเคฟเคธเค•เคพ เค‰เคชเคฏเฅ‹เค— เค•เคฟเคฏเคพ เคœเคพ เคธเค•เคคเคพ ~271.1ms

๐Ÿ“„ License & Commercial Terms (๋ผ์ด์„ ์Šค ๋ฐ ์ƒ์—…์  ๋„์ž… ์•ˆ๋‚ด)

  • Non-Commercial Use Only (๋น„์ƒ์—…์  ์—ฐ๊ตฌยท๊ฐœ์ธ ์ด์šฉ ํ•œ์ •):
    This model is strictly provided for non-commercial, research, and educational purposes only. Any commercial use without a separate commercial agreement is strictly prohibited.
    ๋ณธ ๋ชจ๋ธ์€ ์ˆœ์ˆ˜ ์—ฐ๊ตฌ, ๊ต์œก ๋ฐ ๊ฐœ์ธ ๋น„์˜๋ฆฌ ๋ชฉ์ ์— ํ•œํ•ด์„œ๋งŒ ๋ฌด๋ฃŒ๋กœ ์ œ๊ณต๋ฉ๋‹ˆ๋‹ค. ์‚ฌ์ „ ํ—ˆ๊ฐ€ ์—†๋Š” ์ผ์ฒด์˜ ์˜๋ฆฌ์  ์ด์šฉ์€ ์—„๊ฒฉํžˆ ๊ธˆ์ง€๋ฉ๋‹ˆ๋‹ค.

  • Commercial Licensing & Inquiries for Revenue-Generating Entities (๋งค์ถœ ๋ฐœ์ƒ ๊ธฐ์—… ๋ฐ ์ƒ์šฉํ™” ๋„์ž… ๋ฌธ์˜):
    If your organization generates revenue, or if you plan to incorporate this model into commercial products, services, or paid APIs, you are required to purchase a commercial license. Please contact us directly via email for enterprise licensing and integration support.
    ๊ธฐ์—…/์‚ฌ์—…์ฒด์— ๋งค์ถœ์ด ๋ฐœ์ƒํ•˜๊ณ  ์žˆ๊ฑฐ๋‚˜, ๋ณธ ๋ชจ๋ธ์„ ์ƒ์šฉ ์„œ๋น„์Šคยท์œ ๋ฃŒ ์ œํ’ˆยท์ˆ˜์ต ์ฐฝ์ถœ ๋ชฉ์ ์˜ ์‹œ์Šคํ…œ์— ๋„์ž…ํ•˜๊ณ ์ž ํ•˜์‹œ๋Š” ๊ฒฝ์šฐ ๋ฐ˜๋“œ์‹œ ๋ณ„๋„์˜ ์ƒ์šฉ ๋ผ์ด์„ ์Šค๋ฅผ ์ทจ๋“ํ•˜์…”์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๋„์ž… ์กฐ๊ฑด ๋ฐ ๊ธฐ์—…์šฉ ๋ผ์ด์„ ์Šค ๋ฐœ๊ธ‰์€ ์•„๋ž˜ ์ด๋ฉ”์ผ๋กœ ๋ฌธ์˜ํ•ด ์ฃผ์‹œ๊ธฐ ๋ฐ”๋ž๋‹ˆ๋‹ค.

    • ๋„์ž… ๋ฐ ์ƒ์šฉ ๋ผ์ด์„ ์Šค ๋ฌธ์˜ (Commercial Inquiries): admin@099.kr
    • ๊ณต์‹ ์›น์‚ฌ์ดํŠธ (Official Website): https://baar.uk/model/

โš ๏ธ Notes (์•ˆ๋‚ด ์‚ฌํ•ญ)

  • Pre-trained Model: This model is a compact base pre-trained checkpoint. For specialized tasks, factual precision, or task completion, task-specific fine-tuning (SFT) or Retrieval-Augmented Generation (RAG) pipelines are recommended.
  • ๊ธฐ์ดˆ ์‚ฌ์ „ํ•™์Šต ๋ชจ๋ธ: ๋ณธ ๋ชจ๋ธ์€ 3M ๊ทœ๋ชจ์˜ ๋ฒ ์ด์Šค ์‚ฌ์ „ํ•™์Šต ์ฒดํฌํฌ์ธํŠธ์ž…๋‹ˆ๋‹ค. ์ •๊ตํ•œ ์ž‘์—… ์ˆ˜ํ–‰, ์‚ฌ์‹ค ๊ด€๊ณ„ ๊ฒ€์ฆ ๋ฐ ์™„์„ฑ๋„ ๋†’์€ ๋Œ€ํ™” ์ƒ์„ฑ์ด ํ•„์š”ํ•œ ๊ฒฝ์šฐ ๋„๋ฉ”์ธ ํŒŒ์ธํŠœ๋‹(SFT) ๋˜๋Š” ๊ฒ€์ƒ‰ ์ฆ๊ฐ• ์ƒ์„ฑ(RAG) ํŒŒ์ดํ”„๋ผ์ธ ๊ฒฐํ•ฉ์„ ๊ถŒ์žฅํ•ฉ๋‹ˆ๋‹ค.

๐Ÿ’ผ Opportunities & Contact (์ฑ„์šฉ ์ œ์•ˆ ๋ฐ ํˆฌ์ž ๋ฌธ์˜)

This project demonstrates practical competency in designing custom micro-scale architectures, end-to-end multi-language pre-training, and ultra-low-latency deployment under extreme compute constraints.

I am actively seeking AI Engineering / Research opportunities, recruitment offers, and technical partnerships.

  • ๊ณต์‹ ์›น์‚ฌ์ดํŠธ (Official Website): https://baar.uk/model/
  • ๊ธฐ์—… ๋„์ž… ๋ฐ ๋ผ์ด์„ ์Šค ๋ฌธ์˜ (Commercial Adoption): admin@099.kr
  • ์ฑ„์šฉ ๋ฐ ํฌ์ง€์…˜ ์ œ์•ˆ (Recruitment & Hiring): admin@099.kr
  • ํˆฌ์ž ๋ฐ ๊ธฐ์ˆ  ํ˜‘๋ ฅ ๋ฌธ์˜ (Investment & Partnerships): admin@099.kr

๊ทน๋‹จ์ ์ธ ์ €์‚ฌ์–‘ ๋ฆฌ์†Œ์Šค ํ™˜๊ฒฝ์—์„œ๋„ ์ฆ‰๊ฐ ์ž‘๋™ํ•˜๋Š” ๊ณ ํšจ์œจ ๋‹ค๊ตญ์–ด ๋ชจ๋ธ ์•„ํ‚คํ…์ฒ˜๋ฅผ ์„ค๊ณ„ํ•˜๊ณ , ๋ฐ์ดํ„ฐ ์ •์ œ๋ถ€ํ„ฐ ํ•™์Šตยท์„œ๋น™๊นŒ์ง€ ์ „ ๊ณผ์ •์„ ์ง์ ‘ ๊ตฌํ˜„ํ•˜๋Š” AI ์—”์ง€๋‹ˆ์–ด์ž…๋‹ˆ๋‹ค. ๊ธฐ์ˆ  ๋„์ž…, ์ฑ„์šฉ ์ œ์•ˆ ๋ฐ ํ˜‘์—…์— ๊ด€์‹ฌ ์žˆ๋Š” ๊ธฐ์—…๊ณผ ํŒ€์˜ ๋ฌธ์˜๋ฅผ ํ™˜์˜ํ•ฉ๋‹ˆ๋‹ค.

Downloads last month
479
Safetensors
Model size
4.38M params
Tensor type
F16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Collection including aixk/BAAR2-3M