AiLogic Neural Network Pvt Ltd
Website:
ailogic.com
Job details:
About the Role
We are building custom transformer-based models for multilingual translation and document understanding using our own architecture.
We are looking for engineers who have hands-on experience training and fine-tuning transformer models, not just using LLM APIs or building RAG pipelines.
Key Responsibilities
- Design, build, and train transformer-based models from scratch or via fine-tuning
- Work on sequence-to-sequence (Seq2Seq) models for translation and text generation
- Develop and optimize training pipelines using PyTorch
- Work on tokenization, vocabulary design, and data preprocessing pipelines
- Fine-tune large models using LoRA / QLoRA / PEFT techniques
- Evaluate models using metrics such as BLEU, perplexity, accuracy
- Optimize models for inference (latency, quantization, efficiency)
- Guide junior engineers and define best practices for model training
MUST HAVE (STRICT REQUIREMENTS)
Candidates must have hands-on experience in at least 3 of the following:
- Training transformer models using PyTorch / TensorFlow
- Working with Seq2Seq models (T5, BART, encoder-decoder)
- Fine-tuning LLMs using LoRA / QLoRA / PEFT
- Experience with Hugging Face Transformers (training, not just inference)
- Understanding of attention mechanisms, multi-head attention, transformer architecture
- Experience with custom datasets, training loops, and hyperparameter tuning
STRONG PLUS (HIGH PRIORITY)
- Experience in machine translation / multilingual models
- Tokenization techniques:
- BPE
- SentencePiece
- Distributed training:
- DeepSpeed
- FSDP
- Model optimization:
- quantization
- pruning
Click on Apply to know more.