ML Research Intern: Speculative Decoding for Fast Inference

Together Computer Inc; San Francisco (CA); In loco; USD 80.000 - 96.000; Tempo pieno

Vantaggi offerti da questo lavoro

Housing stipend

Descrizione del lavoro

Together AI is seeking a research intern to advance speculative decoding, training and evaluating draft models at scale on dedicated GPU clusters in San Francisco. You will explore new drafters, verification schemes, and distillation recipes, testing ideas against real workloads and reporting results to the team.

Join the Speculative Decoding group to push frontier models, work on efficient inference, and contribute to robust, reproducible evaluation across production-like traffic during a 12–14

Competenze

• Currently pursuing a Master's or Ph.D. in Computer Science, Electrical Engineering, or related field.
• Strong knowledge of Machine Learning and Deep Learning fundamentals.
• Hands-on experience training or fine-tuning large models with distributed training frameworks (e.g., Megatron-LM, FSDP).
• Familiarity with modern model architectures, such as Mixture-of-Experts and hybrid attention.

Mansioni

• Train, fine-tune and distill draft models at scale on dedicated GPU clusters.
• Test ideas against realistic serving workloads including long-context traffic.
• Communicate plans, progress and results to the broader team.

Conoscenze

Machine Learning fundamentals

Distributed training

Megatron-LM

FSDP

Mixture-of-Experts

Hybrid attention

Formazione

Master's or Ph.D. in CS/EE

Strumenti

Megatron-LM

FSDP

vLLM

TensorRT-LLM

Descrizione del lavoro

Together AI is seeking a research intern to advance speculative decoding, training and evaluating draft models at scale on dedicated GPU clusters in San Francisco. You will explore new drafters, verification schemes, and distillation recipes, testing ideas against real workloads and reporting results to the team.

Join the Speculative Decoding group to push frontier models, work on efficient inference, and contribute to robust, reproducible evaluation across production-like traffic during a 12–14