Large Language Model Intern

Location: Singapore

Company

Razer develops gamer-focused software and products for a global gaming audience.

What you will do

• Construct pretraining corpora through data cleaning, filtering, deduplication, and mixture design.
• Run fine-tuning experiments using SFT, LoRA, and preference optimization.
• Evaluate model quality, latency, and memory usage against production constraints.
• Optimize deployment for on-device targets through compression and quantization experiments.
• Own defined workstreams end-to-end, from experiment design and execution through debugging, analysis, and iteration.

Requirements

• Currently pursuing a Bachelor's, Master's, or PhD in Computer Science, AI, Data Science, or a related field.
• Knowledge of LLM training paradigms, including pretraining, SFT, LoRA, and preference optimization.
• Strong understanding of Transformer and deep learning fundamentals.
• Hands-on Python and PyTorch model training experience, including training-corpus data engineering and experiment analysis.
• Experience benchmarking model quality, latency, and memory usage.
• Demonstrable hands-on model training or fine-tuning experience is required; API-calling or prompt-engineering-only experience does not qualify.

Nice to have

• Experience with model compression, quantization, distillation, or on-device deployment.
• Distributed training experience, including multi-GPU or parallel training.
• Experience with Hugging Face, Accelerate, DeepSpeed, or Linux environments.
• Training models from scratch or publishing at major machine learning conferences.

Culture & Benefits

• Work with senior data scientists and engineers on a production AI feature.
• Receive mentorship and visibility into production roadmap decisions.
• Gain practical experience across the full LLM training lifecycle at production scale.
• Join a global team operating across five continents.
• Work in an inclusive environment committed to equal opportunity and reasonable accommodations.