Dnotitia Inc.

Team

company

https://dnotitia.com

dnotitia

Activity Feed

AI & ML interests

Pioneering the Next Era of AI with Vector Intelligence

Recent Activity

minseob00 published a dataset about 23 hours ago

dnotitia/dna_style

minseob00 published a dataset about 23 hours ago

dnotitia/dna_hallucination

minseob00 published a dataset about 23 hours ago

dnotitia/dna_confusion

View all activity

Papers

N-Gram Memories at Two Bits: Quantizing Lookup Tables in Large Language Models

Making Qwen3 Think in Korean with Reinforcement Learning

View all Papers

dnotitia 's collections 13

LRL Calibrated Models

Where Do Multilingual Vision-Language Encoders Fail on Low-Resource Languages? (EMNLP2026)

dnotitia/geometric-bottleneck-metaclip2

Zero-Shot Image Classification • 2B • Updated 26 days ago • 6
dnotitia/geometric-bottleneck-siglip2

Zero-Shot Image Classification • 1B • Updated 26 days ago • 12
dnotitia/geometric-bottleneck-altclip

Zero-Shot Image Classification • 0.9B • Updated 26 days ago • 10
dnotitia/geometric-bottleneck-msiglip

Zero-Shot Image Classification • 0.4B • Updated 26 days ago • 7

DNA 3.1

dnotitia/DNA3.1-27B

Image-Text-to-Text • 27B • Updated Sep 22 • 68 • 1

DNA 3.0

DNA 3.0 preserves Qwen 3.5/3.6 strengths while removing censorship on China-related topics.

dnotitia/DNA3.0-0.8B

Image-Text-to-Text • 1B • Updated Jul 2 • 47
dnotitia/DNA3.0-2B

Image-Text-to-Text • 3B • Updated Jul 2 • 39 • 1
dnotitia/DNA3.0-4B

Image-Text-to-Text • 5B • Updated Jul 2 • 56 • 3
dnotitia/DNA3.0-9B

Image-Text-to-Text • 9B • Updated Jul 2 • 109 • 2

Qwen3-ChatTemplate

Patched versions of Qwen3 with identical weights, featuring a separated chat_template.jinja and {% generation %} tags for seamless trl assistant_only.

dnotitia/Qwen3-4B-Thinking-2507

Text Generation • 4B • Updated Feb 9 • 10
dnotitia/Qwen3-4B-Instruct-2507

Text Generation • 4B • Updated Feb 9 • 41
dnotitia/Qwen3-4B

Text Generation • 4B • Updated Feb 9 • 169 •
dnotitia/Qwen3-4B-Base

Text Generation • 4B • Updated Feb 9 • 16

DNA 2.0

High-performance LLM developed by Dnotitia Inc., incorporating cutting-edge techniques for superior reasoning tasks.

dnotitia/DNA-2.0-0.6B

Text Generation • 0.6B • Updated Jul 24, 2025
dnotitia/DNA-2.0-1.7B

Text Generation • 2B • Updated Jul 24, 2025
dnotitia/DNA-2.0-4B

Text Generation • 4B • Updated Jul 24, 2025 • 1
dnotitia/DNA-2.0-8B

Text Generation • 8B • Updated Jul 24, 2025 • 2

DNA 1.0

8B Korean SoTA model, which is instruction-tuned by Dnotitia Inc.

DNA 1.0 Technical Report

Paper • 2501.10648 • Published Jan 18, 2025 • 1
dnotitia/Llama-DNA-1.0-8B-Instruct

Text Generation • 8B • Updated Jun 27, 2025 • 97 • • 61
dnotitia/Llama-DNA-1.0-8B-Instruct-GGUF

Text Generation • 8B • Updated Jan 22, 2025 • 549 • 1

Smoothie Qwen2.5

For more details, please visit https://github.com/dnotitia/smoothie-qwen

Smoothie-Qwen: Post-Hoc Smoothing to Reduce Language Bias in Multilingual LLMs

Paper • 2507.05686 • Published Jul 8, 2025 • 1
dnotitia/Smoothie-Qwen2.5-0.5B-Instruct

Text Generation • 0.5B • Updated May 4, 2025 • 13
dnotitia/Smoothie-Qwen2.5-1.5B-Instruct

Text Generation • 2B • Updated May 4, 2025 • 18 • 1
dnotitia/Smoothie-Qwen2.5-3B-Instruct

Text Generation • 3B • Updated May 4, 2025 • 17 • 2

DNA Evaluation Suite

Released datasets for dna-eval

dnotitia/dna_performance

Viewer • Updated Oct 7 • 30 • 165
dnotitia/dna_persona

Viewer • Updated Oct 7 • 50 • 192
dnotitia/dna_confusion

Viewer • Updated Oct 7 • 45 • 74
dnotitia/dna_hallucination

Viewer • Updated Oct 7 • 100 • 120

DNA-VL-STEER

dnotitia/DNA-VL-STEER-2B

Sentence Similarity • 2B • Updated Sep 10 • 276 • 14

4B Experiments

Systematic SFT, GRPO for Qwen3-4B. We explore diverse dataset compositions and training recipes to benchmark and improve performance across tasks.

dnotitia/Qwen3-4B-Instruct-2507

Text Generation • 4B • Updated Feb 9 • 41
dnotitia/Qwen3-4B-Thinking-2507

Text Generation • 4B • Updated Feb 9 • 10
dnotitia/Qwen3-4B

Text Generation • 4B • Updated Feb 9 • 169 •
dnotitia/Qwen3-4B-Base

Text Generation • 4B • Updated Feb 9 • 16

DNA 2.1

Making Qwen3 Think in Korean with Reinforcement Learning https://arxiv.org/abs/2508.10355

Making Qwen3 Think in Korean with Reinforcement Learning

Paper • 2508.10355 • Published Aug 14, 2025 • 2
dnotitia/DNA-2.1-14B

Text Generation • 15B • Updated Oct 20, 2025 • 32 • 3

DNA-R1

Reasoning model distilled from DeepSeek-R1, enhanced with GRPO using supplementary reasoning datasets.

dnotitia/DNA-R1

Text Generation • 15B • Updated Mar 11, 2025 • 54 • 41

Smoothie Qwen3

For more details, please visit https://github.com/dnotitia/smoothie-qwen

Smoothie-Qwen: Post-Hoc Smoothing to Reduce Language Bias in Multilingual LLMs

Paper • 2507.05686 • Published Jul 8, 2025 • 1
dnotitia/Smoothie-Qwen3-0.6B

Text Generation • 0.6B • Updated May 4, 2025 • 37 • 1
dnotitia/Smoothie-Qwen3-1.7B

Text Generation • 2B • Updated May 4, 2025 • 34 • 2
dnotitia/Smoothie-Qwen3-4B

Text Generation • 4B • Updated May 4, 2025 • 19 • 4

LRL Calibrated Models

Where Do Multilingual Vision-Language Encoders Fail on Low-Resource Languages? (EMNLP2026)

dnotitia/geometric-bottleneck-metaclip2

Zero-Shot Image Classification • 2B • Updated 26 days ago • 6
dnotitia/geometric-bottleneck-siglip2

Zero-Shot Image Classification • 1B • Updated 26 days ago • 12
dnotitia/geometric-bottleneck-altclip

Zero-Shot Image Classification • 0.9B • Updated 26 days ago • 10
dnotitia/geometric-bottleneck-msiglip

Zero-Shot Image Classification • 0.4B • Updated 26 days ago • 7

DNA Evaluation Suite

Released datasets for dna-eval

dnotitia/dna_performance

Viewer • Updated Oct 7 • 30 • 165
dnotitia/dna_persona

Viewer • Updated Oct 7 • 50 • 192
dnotitia/dna_confusion

Viewer • Updated Oct 7 • 45 • 74
dnotitia/dna_hallucination

Viewer • Updated Oct 7 • 100 • 120

DNA 3.1

dnotitia/DNA3.1-27B

Image-Text-to-Text • 27B • Updated Sep 22 • 68 • 1

DNA-VL-STEER

dnotitia/DNA-VL-STEER-2B

Sentence Similarity • 2B • Updated Sep 10 • 276 • 14

DNA 3.0

DNA 3.0 preserves Qwen 3.5/3.6 strengths while removing censorship on China-related topics.

dnotitia/DNA3.0-0.8B

Image-Text-to-Text • 1B • Updated Jul 2 • 47
dnotitia/DNA3.0-2B

Image-Text-to-Text • 3B • Updated Jul 2 • 39 • 1
dnotitia/DNA3.0-4B

Image-Text-to-Text • 5B • Updated Jul 2 • 56 • 3
dnotitia/DNA3.0-9B

Image-Text-to-Text • 9B • Updated Jul 2 • 109 • 2

4B Experiments

Systematic SFT, GRPO for Qwen3-4B. We explore diverse dataset compositions and training recipes to benchmark and improve performance across tasks.

dnotitia/Qwen3-4B-Instruct-2507

Text Generation • 4B • Updated Feb 9 • 41
dnotitia/Qwen3-4B-Thinking-2507

Text Generation • 4B • Updated Feb 9 • 10
dnotitia/Qwen3-4B

Text Generation • 4B • Updated Feb 9 • 169 •
dnotitia/Qwen3-4B-Base

Text Generation • 4B • Updated Feb 9 • 16

Qwen3-ChatTemplate

Patched versions of Qwen3 with identical weights, featuring a separated chat_template.jinja and {% generation %} tags for seamless trl assistant_only.

dnotitia/Qwen3-4B-Thinking-2507

Text Generation • 4B • Updated Feb 9 • 10
dnotitia/Qwen3-4B-Instruct-2507

Text Generation • 4B • Updated Feb 9 • 41
dnotitia/Qwen3-4B

Text Generation • 4B • Updated Feb 9 • 169 •
dnotitia/Qwen3-4B-Base

Text Generation • 4B • Updated Feb 9 • 16

DNA 2.1

Making Qwen3 Think in Korean with Reinforcement Learning https://arxiv.org/abs/2508.10355

Making Qwen3 Think in Korean with Reinforcement Learning

Paper • 2508.10355 • Published Aug 14, 2025 • 2
dnotitia/DNA-2.1-14B

Text Generation • 15B • Updated Oct 20, 2025 • 32 • 3

DNA 2.0

High-performance LLM developed by Dnotitia Inc., incorporating cutting-edge techniques for superior reasoning tasks.

dnotitia/DNA-2.0-0.6B

Text Generation • 0.6B • Updated Jul 24, 2025
dnotitia/DNA-2.0-1.7B

Text Generation • 2B • Updated Jul 24, 2025
dnotitia/DNA-2.0-4B

Text Generation • 4B • Updated Jul 24, 2025 • 1
dnotitia/DNA-2.0-8B

Text Generation • 8B • Updated Jul 24, 2025 • 2

DNA-R1

Reasoning model distilled from DeepSeek-R1, enhanced with GRPO using supplementary reasoning datasets.

dnotitia/DNA-R1

Text Generation • 15B • Updated Mar 11, 2025 • 54 • 41

DNA 1.0

8B Korean SoTA model, which is instruction-tuned by Dnotitia Inc.

DNA 1.0 Technical Report

Paper • 2501.10648 • Published Jan 18, 2025 • 1
dnotitia/Llama-DNA-1.0-8B-Instruct

Text Generation • 8B • Updated Jun 27, 2025 • 97 • • 61
dnotitia/Llama-DNA-1.0-8B-Instruct-GGUF

Text Generation • 8B • Updated Jan 22, 2025 • 549 • 1

Smoothie Qwen3

For more details, please visit https://github.com/dnotitia/smoothie-qwen

Smoothie-Qwen: Post-Hoc Smoothing to Reduce Language Bias in Multilingual LLMs

Paper • 2507.05686 • Published Jul 8, 2025 • 1
dnotitia/Smoothie-Qwen3-0.6B

Text Generation • 0.6B • Updated May 4, 2025 • 37 • 1
dnotitia/Smoothie-Qwen3-1.7B

Text Generation • 2B • Updated May 4, 2025 • 34 • 2
dnotitia/Smoothie-Qwen3-4B

Text Generation • 4B • Updated May 4, 2025 • 19 • 4

Smoothie Qwen2.5

For more details, please visit https://github.com/dnotitia/smoothie-qwen

Smoothie-Qwen: Post-Hoc Smoothing to Reduce Language Bias in Multilingual LLMs

Paper • 2507.05686 • Published Jul 8, 2025 • 1
dnotitia/Smoothie-Qwen2.5-0.5B-Instruct

Text Generation • 0.5B • Updated May 4, 2025 • 13
dnotitia/Smoothie-Qwen2.5-1.5B-Instruct

Text Generation • 2B • Updated May 4, 2025 • 18 • 1
dnotitia/Smoothie-Qwen2.5-3B-Instruct

Text Generation • 3B • Updated May 4, 2025 • 17 • 2

AI & ML interests

Recent Activity

Papers

Team members 32

dnotitia 's collections 13