Company: Avito
Employment Type: FULL_TIME
Data Scientist in Avito's LLM Core team. Training and development of own foundational language models: from research hypotheses on pre-training, post-training, and RL to production implementation and inference optimization.
Tasks:
- Formulate and test hypotheses in the areas of pre-training, continual pre-training, post-training, RL, synthetic data, and model architectures
- Independently lead complex research and engineering tasks: from setting goals, choosing metrics, and designing experiments to implementation, results analysis, and deployment
- Scale training and inference: profile bottlenecks, improve the efficiency of distributed training and inference, work with quantization, speculative decoding, and modern inference engines
- Build robust quality evaluation procedures, conduct error analysis, identify regressions, and design datasets and metrics for comparing models by quality, latency, and cost
- Write production code in Python and PyTorch, conduct code reviews, improve experiment reproducibility, reliability, and maintainability of ML solutions
- Share knowledge, mentor less experienced colleagues, participate in architectural discussions and internal LLM seminars
- Scale the foundational A-vibe model: explore new architectures and training approaches
- Optimize the speed and cost of inference for high-load models in production
- Design and assist in developing internal LLM platforms and multi-agent systems
- Build robust infrastructure for model quality evaluation
Requirements:
- At least 3 years of experience in implementing and operating ML solutions
- Deep understanding of modern ML algorithms, Transformers, and LLM training principles
- Practical experience in training, adapting, or deploying NLP models and the ability to turn ideas into testable experiments
- Ability to work independently with uncertainty: decompose problems, select metrics, analyze results, and make technical decisions
- Confident command of Python and PyTorch, good knowledge of software engineering practices
- Understanding of the trade-offs between quality, latency, throughput, and cost, and consideration of these factors when choosing a solution
Conditions:
- Opportunity to influence the product: decisions will improve the experience of millions of users
- Complex technological tasks on a large scale
- Access to high-performance GPU clusters for experiments
- Strong community of experts and regular LLM seminars
- Budget for training, courses, conferences, and professional literature
- Voluntary medical insurance with dental coverage from day one, on-site therapists and massage therapists
- Flexibility: possibility to work remotely or from offices in 4 cities in Russia
Skills:
- Python
- PyTorch
- LLM
- Transformer
- RL
- LoRA
- SFT
- tokenizers
- DeepSpeed
- vLLM
- SGLang
- TensorRT-LLM
- speculative decoding
- Weights & Biases
- MLflow
- DVC
- NLP
- MLOps
