MamayLM, передова мовна модель для української мови
• 44
AI & ML interests
None defined yet.
Recent Activity
Papers
Aletheia: What Makes RLVR For Code Verifiers Tick?
AR-VLA: True Autoregressive Action Expert for Vision-Language-Action Models
Articles
AttackViz dataset + robust chart QA model from ChartAttack: Testing the Vulnerability of LLMs to Malicious Prompting in Chart Generation
An Automated Reranking Benchmark for Mathematical Information Retrieval
Generalist VLAs with same VLM backbone and insulated action heads for on FAST, Flow-matching, or Autoregressive output.
Recovered in Translation: Efficient Pipeline for Automated Translation of Benchmarks and Datasets (ACL 2026)
-
Recovered in Translation: Efficient Pipeline for Automated Translation of Benchmarks and Datasets
Paper • 2602.22207 • Published • 45 -
INSAIT-Institute/rit-mmlu
Viewer • Updated • 111k • 38 -
INSAIT-Institute/rit-winogrande
Viewer • Updated • 8.87k • 40 -
INSAIT-Institute/rit-hellaswag
Viewer • Updated • 70.3k • 64
Our ReVLA model finetuned at different datasets
The first benchmark for evaluating LLM sycophancy in mathematical reasoning.
A collection of the Open Proof Corpus dataset and the finetuned judging model.
BgGPT-Gemma-2 collection of models.
-
INSAIT-Institute/BgGPT-Gemma-2-27B-IT-v1.0
Text Generation • 27B • Updated • 61 • • 17 -
INSAIT-Institute/BgGPT-Gemma-2-2.6B-IT-v1.0
Text Generation • 3B • Updated • 200 • • 7 -
INSAIT-Institute/BgGPT-Gemma-2-9B-IT-v1.0
Text Generation • 9B • Updated • 52 • • 11 -
INSAIT-Institute/BgGPT-Gemma-2-2.6B-IT-v1.0-GGUF
3B • Updated • 259 • 4
Aletheia: What Makes RLVR For Code Verifiers Tick? Benchmark datasets and fine-tuned verifier models (1.5B/7B/14B). Mirror of Aletheia-Bench.
Biggest release of Ukrainian-adapted MamayLM model series
-
INSAIT-Institute/MamayLM-Gemma-3-27B-IT-v2.0
Image-Text-to-Text • 27B • Updated • 225 • 9 -
INSAIT-Institute/MamayLM-Gemma-3-12B-IT-v2.0
Image-Text-to-Text • 12B • Updated • 1.57k • 5 -
INSAIT-Institute/MamayLM-Gemma-3-27B-IT-v2.0-FP8-dynamic
Image-Text-to-Text • 29B • Updated • 70 • 2 -
INSAIT-Institute/MamayLM-Gemma-3-27B-IT-v2.0-GGUF
Image-Text-to-Text • 27B • Updated • 594 • 6
-
INSAIT-Institute/BgGPT-Gemma-3-27B-IT
Image-Text-to-Text • 27B • Updated • 3.9k • 11 -
INSAIT-Institute/BgGPT-Gemma-3-12B-IT
Image-Text-to-Text • 12B • Updated • 95 • • 4 -
INSAIT-Institute/BgGPT-Gemma-3-4B-IT
Image-Text-to-Text • 4B • Updated • 109 • 4 -
INSAIT-Institute/BgGPT-Gemma-3-27B-IT-GGUF
Image-Text-to-Text • 27B • Updated • 377 • 7
Models of the SPEAR series.
First Open and Multimodal Ukrainian-focused LLM
-
INSAIT-Institute/MamayLM-Gemma-3-12B-IT-v1.0
Image-Text-to-Text • 12B • Updated • 398 • • 22 -
MamayLM v1.0 Release Blog
🌻27MamayLM v1.0 - now multimodal and even stronger!
-
INSAIT-Institute/MamayLM-Gemma-3-4B-IT-v1.0
Image-Text-to-Text • 4B • Updated • 1.06k • • 16 -
INSAIT-Institute/MamayLM-Gemma-3-12B-IT-v1.0-GGUF
12B • Updated • 201 • 16
A model collection consisting of different model adapters trained with our adversarial training method MixAT
Ukrainian-focused MamayLM model collection, based on Gemma2
AttackViz dataset + robust chart QA model from ChartAttack: Testing the Vulnerability of LLMs to Malicious Prompting in Chart Generation
Aletheia: What Makes RLVR For Code Verifiers Tick? Benchmark datasets and fine-tuned verifier models (1.5B/7B/14B). Mirror of Aletheia-Bench.
An Automated Reranking Benchmark for Mathematical Information Retrieval
Biggest release of Ukrainian-adapted MamayLM model series
-
INSAIT-Institute/MamayLM-Gemma-3-27B-IT-v2.0
Image-Text-to-Text • 27B • Updated • 225 • 9 -
INSAIT-Institute/MamayLM-Gemma-3-12B-IT-v2.0
Image-Text-to-Text • 12B • Updated • 1.57k • 5 -
INSAIT-Institute/MamayLM-Gemma-3-27B-IT-v2.0-FP8-dynamic
Image-Text-to-Text • 29B • Updated • 70 • 2 -
INSAIT-Institute/MamayLM-Gemma-3-27B-IT-v2.0-GGUF
Image-Text-to-Text • 27B • Updated • 594 • 6
Generalist VLAs with same VLM backbone and insulated action heads for on FAST, Flow-matching, or Autoregressive output.
-
INSAIT-Institute/BgGPT-Gemma-3-27B-IT
Image-Text-to-Text • 27B • Updated • 3.9k • 11 -
INSAIT-Institute/BgGPT-Gemma-3-12B-IT
Image-Text-to-Text • 12B • Updated • 95 • • 4 -
INSAIT-Institute/BgGPT-Gemma-3-4B-IT
Image-Text-to-Text • 4B • Updated • 109 • 4 -
INSAIT-Institute/BgGPT-Gemma-3-27B-IT-GGUF
Image-Text-to-Text • 27B • Updated • 377 • 7
Recovered in Translation: Efficient Pipeline for Automated Translation of Benchmarks and Datasets (ACL 2026)
-
Recovered in Translation: Efficient Pipeline for Automated Translation of Benchmarks and Datasets
Paper • 2602.22207 • Published • 45 -
INSAIT-Institute/rit-mmlu
Viewer • Updated • 111k • 38 -
INSAIT-Institute/rit-winogrande
Viewer • Updated • 8.87k • 40 -
INSAIT-Institute/rit-hellaswag
Viewer • Updated • 70.3k • 64
Models of the SPEAR series.
Our ReVLA model finetuned at different datasets
First Open and Multimodal Ukrainian-focused LLM
-
INSAIT-Institute/MamayLM-Gemma-3-12B-IT-v1.0
Image-Text-to-Text • 12B • Updated • 398 • • 22 -
MamayLM v1.0 Release Blog
🌻27MamayLM v1.0 - now multimodal and even stronger!
-
INSAIT-Institute/MamayLM-Gemma-3-4B-IT-v1.0
Image-Text-to-Text • 4B • Updated • 1.06k • • 16 -
INSAIT-Institute/MamayLM-Gemma-3-12B-IT-v1.0-GGUF
12B • Updated • 201 • 16
The first benchmark for evaluating LLM sycophancy in mathematical reasoning.
A model collection consisting of different model adapters trained with our adversarial training method MixAT
A collection of the Open Proof Corpus dataset and the finetuned judging model.
Ukrainian-focused MamayLM model collection, based on Gemma2
BgGPT-Gemma-2 collection of models.
-
INSAIT-Institute/BgGPT-Gemma-2-27B-IT-v1.0
Text Generation • 27B • Updated • 61 • • 17 -
INSAIT-Institute/BgGPT-Gemma-2-2.6B-IT-v1.0
Text Generation • 3B • Updated • 200 • • 7 -
INSAIT-Institute/BgGPT-Gemma-2-9B-IT-v1.0
Text Generation • 9B • Updated • 52 • • 11 -
INSAIT-Institute/BgGPT-Gemma-2-2.6B-IT-v1.0-GGUF
3B • Updated • 259 • 4