# Yusuf Erdem Erdem

> Yusuf Erdem Erdem, AI research engineer in Istanbul: Turkish LLMs, vision-language models, efficient inference. ModernBERT-TR, ChainFlow, Cosmos-LLaVA. Also published as "Yusuf Erdem". Open to research opportunities and collaborations.

Main language: English. A Turkish version of every page lives under https://erdemerdem.com/tr/.
Contact: https://erdemerdem.com/contact/

## Pages
- [Home](https://erdemerdem.com/): overview, bio, selected work
- [Work experience](https://erdemerdem.com/experience/): COSMOS, Baykar, ASELSAN
- [Research](https://erdemerdem.com/research/): themes, selected work and all publications
- [Projects, models and datasets](https://erdemerdem.com/projects/)
- [Leadership, community and honors](https://erdemerdem.com/community/)
- [News](https://erdemerdem.com/news/)

## Publications


## Models, datasets and code
- ModernBERT-TR: A long-context (8,192 tokens) Turkish encoder with a family of specialized downstream models. https://huggingface.co/collections/ytu-ce-cosmos/modernbert-tr
- ChainFlow: A flow-matching drafter for lossless speculative decoding in vLLM: up to 2.5x faster on Qwen3.5 (4B to 27B), with Turkish-adapted checkpoints. https://huggingface.co/collections/ytu-ce-cosmos/chain-flow https://github.com/Cosmos-YTU/chain-flow
- Cosmos-LLaVA: Turkish's first large-scale multimodal datasets along with a vision-language instruction-following model. https://huggingface.co/ytu-ce-cosmos/Turkish-LLaVA-v0.1
- ABSA-TR: Turkish Aspect-based sentiment analysis dataset. Built from native sources in the e-commerce, supplements, and movie domains. https://huggingface.co/datasets/ytu-ce-cosmos/absa-tr
- Guardrail-TR: About 405K Turkish prompts for safe/unsafe classification. Multi-label hazard categories covering 11 labels. https://huggingface.co/datasets/ytu-ce-cosmos/guardrail-tr
- Turkish-LLaVA-Pretrain: Turkish multimodal pretraining dataset behind Cosmos-LLaVA. https://huggingface.co/datasets/ytu-ce-cosmos/Turkish-LLaVA-Pretrain
- Turkish-LLaVA-Finetune: Turkish visual instruction fine-tuning dataset. One of the datasets behind Cosmos-LLaVA. https://huggingface.co/datasets/ytu-ce-cosmos/Turkish-LLaVA-Finetune
- turkce-kitap: Turkish conversations about book covers, for vision-language training. https://huggingface.co/datasets/ytu-ce-cosmos/turkce-kitap
- Android intent and slot demo: ModernBERT-TR intent and slot filling running natively on Android. https://github.com/yusuf740/Modernbert_v2_intent-slot_demo
- Turkish language models: A range of Turkish language models built on Llama, Gemma, E5, BERT, GPT and other architectures. More in the COSMOS Hugging Face organization. https://huggingface.co/ytu-ce-cosmos

## Profiles
- linkedin: https://www.linkedin.com/in/yusuf-erdem-erdem
- scholar: https://scholar.google.com/citations?user=izoF-24AAAAJ
- github: https://github.com/erdem-erdem
- huggingface: https://huggingface.co/erdem-erdem
- orcid: https://orcid.org/0009-0003-6598-4434

## More
- Full text version: https://erdemerdem.com/llms-full.txt