Foundation Model Systems Engineer - Llama Ecosystem

About the opportunity

Optimize distributed training, speculative decoding, and model quantization for open-weights Llama foundation models.

What you will do

• check_circle Implement speculative decoding and KV cache optimizations reducing LLM time-to-first-token by 40%.
• check_circle Quantize foundation models using AWQ, GPTQ, and FP8 formats for fast consumer GPU deployment.
• check_circle Manage massive petabyte synthetic data generation and filtering pipelines.
• check_circle Collaborate with the open-source community to maintain Llama GitHub reference implementations.

What we are looking for

• arrow_circle_right 4+ years experience in deep learning systems and large model training.
• arrow_circle_right Proficiency in PyTorch, C++, CUDA, Triton, and vLLM or TensorRT-LLM.
• arrow_circle_right Strong understanding of LLM architecture modifications (RoPE, GQA, MoE).

Skills & Tech Stack