Mikail Okyay Top-Executive
๐๐๐ง๐ข๐จ๐ซ ๐๐๐๐ก๐ข๐ง๐ ๐๐๐๐ซ๐ง๐ข๐ง๐ ๐๐๐ข๐๐ง๐ญ๐ข๐ฌ๐ญ (๐๐ฎ๐ฅ๐ญ๐ข๐ฏ๐๐ซ๐ฌ๐ ๐๐จ๐ฆ๐ฉ๐ฎ๐ญ๐ข๐ง๐ , 12.2024-Proceed)
- Working on LLM and multimodal compression.
- Using multiple training techniques such as SFT, DPO, RL and GRPO.
- Applying knowledge distillation techniques.
- Leveraging Deepspeed, FSDP, NeMo, and Megatron.
- Optimizing models for efficient inference and deployment.
- LLMs, VLMs and multimodals evaluation by using lm-eval, lmms-eval and VLMEvalKit
- High-performance serving for LLMs, VLMs and multimodals by using SGLang and vLLM.
- Training on Leonardo HPCs.
๐๐ ๐๐ง๐ ๐ข๐ง๐๐๐ซ (๐๐ข๐ฅ๐จ ๐๐, 09.2023-12.2024)
- Led projects focused on actionable sustainability powered by AI.
- Developed and optimized Boolean Search Assistant and Coding Assistant.
- Developed scrapper based on google search and sitemap crawling.
- Utilized advanced LLMs including Mixtral-8x7b, Starcoder2-7b, Codegemma-7b-it, Deepseek-coder-33b-instruct.
- Applied SFT, DPO trainers, and fast fine-tuning methods like PEFT, QLoRa.
- Deployed LLMs using frameworks like vLLM, Text Generation Inference, Deepspeed-mii, OpenLLM.
- Integrated APIs from OpenAI, Anthropic, and Gemini.
- Worked with RAG tools LangChain and LlamaIndex, and vector DBs including Chroma DB, pgvector, Qdrant, Weaviate.
- Managed cloud services across GCP, AWS, and Azure.
- Deployed applications using RestAPI (FastAPI, Flask), Docker, and Kubernetes.
๐๐๐ ๐๐๐ข๐๐ง๐ญ๐ข๐ฌ๐ญ (๐๐ฎ๐๐ฐ๐๐ข, 04.2021-09.2023)
- Developed spell error correction for map search using probabilistic methods.
- Led NER Project using CRF, BERT, XLM-Roberta with ONNX quantization.
-Applied knowledge distillation and active learning for NER tasks.
- Built question-answering systems with T5, Bert, and Llama-2-7B.
๐๐จ๐๐ญ๐ฐ๐๐ซ๐ ๐๐ง๐ ๐ข๐ง๐๐๐ซ (๐๐ง๐ข๐ฉ๐, 11.2019-03.2021)
- Developed and maintained web APIs for university IT systems.
- Conducted data analysis with machine learning tools.
๐๐ค๐ข๐ฅ๐ฅ๐ฌ
LLMs, Multimodals, VLMs, NLP, Machine Learning, Deep Learning, Image Processing, Classification, Regression, Clustering, Tensorflow, Pytorch, Scikit-Learn, SciPy, OpenCV, Transformers, SFT, DPO, RL, GRPO, PEFT, QLoRa, Accelerator, FlashAttention, Deepspeed, FSDP, NeMo, Megatron, BitsAndBytes, Llama.cpp (GGUF), GPTQ, AWQ, Half Quadratic Quantization (HQQ), ONNX Runtime, LangChain, LangGraph, LlamaIndex, vLLM, SGLang, HPCs, Text Generation Inference, OpenLLM, AWS, Google Cloud, Microsoft Azure, Python, Java, C#, SQL, HTML, CSS, Linux, MySQL, PostgreSQL, MsSQL, FastAPI, Flask, Docker, Kubernetes, GitHub, GitLab
