Location : India (Remote / Hybrid)
Experience : 3-6 years in ML / NLP
Role Overview :
You will lead the fine-tuning and evaluation of open-source LLMs (e.g., Llama, Mistral, Qwen). This role requires hands-on skills in Python, PyTorch, Hugging Face, and distributed training frameworks.
Responsibilities :
- Fine-tune and adapt open-source LLMs using LoRA / QLoRA.
- Build pipelines for supervised fine-tuning and preference optimization (DPO / RLHF).
- Evaluate model performance using standard and domain-specific benchmarks.
- Optimize inference with quantization and serving frameworks (vLLM, TGI).
- Collaborate with data engineers on preprocessing and tokenization.
Requirements :
Strong coding in Python and PyTorch.Experience with Hugging Face, DeepSpeed, or similar frameworks.Familiarity with distributed training and GPU optimization.Knowledge of transformers, attention mechanisms, and NLP workflows.Bonus : Experience in domain-specific fine-tuning (finance, legal, etc.)(ref : hirist.tech)