About this role
Lead training for on-device model families including pretraining, post-training (SFT, RLHF, DPO, GRPO), distillation, quantization, and shipping models into production on-device runtimes. Own end-to-end model capabilities from data mixture and training objective through evaluation, checkpointing, and shipping into partner builds.
Skills for this role
PretrainingSFTRLHFDPOGRPODistillationQuantizationMixture of Experts (MoE)Speculative DecodingKV-cache CompressionData MixtureTraining ObjectivesExperiment DesignModel EvaluationModel DeploymentOn-device RuntimeCheckpointingAgentic Post-trainingOn-device ToolingResearch Writeups