SLM & VLM Engineer - Post-Training, Tool Calling & Agents

About the role

Jabil is looking for an engineer/researcher to lead R&D on Small Language Models (SLMs) and Vision-Language Models (VLMs) for edge and low-latency, cost-efficient production scenarios. You will own the continuous pretraining, supervised instruction tuning (SFT) and compression/distillation pipelines, working with platform teams to improve inference efficiency, tool-use success rate and overall model quality.

Key responsibilities

• Continuous pretraining and SFT for SLMs and VLMs, with reproducible PyTorch workflows covering data processing, training, evaluation and model versioning.
• Compression and distillation strategies (knowledge distillation, pruning, quantization-oriented training or post-training optimization).
• Inference optimization for low-latency and edge scenarios: quantization, caching/KV optimizations, batching and decoding-time optimizations to improve throughput and cost per token.
• A production-grade tool calling framework: cataloging and metadata design, routing and argument construction, parameter validation, safe fallback and retry, and call-chain tracing and observability.
• Post-training methods such as PPO, DPO or GRPO-like optimization and reward modeling to align models on semantic understanding, tool-use success and content quality, with offline and online iteration loops.
• Automated data pipelines for collection, cleaning, de-duplication, labeling and weak supervision, supporting both SFT and preference/RL-style post-training.
• Evaluation mechanisms: offline benchmarks, task suites for tool-use, regression tests and reliability metrics, with A/B comparisons, ablations and failure analysis.

Required profile

• Strong software engineering skills in Python and C++, including ML training and evaluation pipelines in PyTorch.
• Hands-on experience in model efficiency and inference optimization (distillation, quantization, pruning, serving optimization).
• Experience with high-performance computing and acceleration: CUDA and/or SIMD, profiling and performance tuning.
• Ability to read recent research papers and implement algorithms with strong experimental discipline.
• Bilingual in Mandarin Chinese and English, to liaise with colleagues in China.