Senior / Staff SLM & VLM Engineer — Post-Training, Tool Calling & Agents

Job Summary 

We are looking for a highly capable engineer/researcher to lead the R&D of Small Language Models (SLMs) and Vision-Language Models (VLMs) for edge / low-latency and cost-efficient production scenarios. You will own the continuous pretraining, supervised instruction tuning (SFT), and compression/distillation pipelines, and work closely with platform teams to deliver reliable, measurable improvements in inference efficiency, tool-use success rate, and overall model quality.  

Key Responsibilities 

1) SLM/VLM Training: Continuous Pretraining & Instruction Tuning (SFT) 

• Conduct continuous pretraining and SFT for SLMs and VLMs to improve task performance and domain adaptation. 

• Build reproducible training workflows in PyTorch, including data processing, training, evaluation, and model versioning.  

2) Compression, Distillation & Edge/Low-Latency Inference Optimization 

• Design and implement efficient compression strategies for SLM/VLM, including knowledge distillation, pruning, and quantization-oriented training or post-training optimization. 

• Optimize model serving and inference for low-latency / edge scenarios by improving throughput and cost-per-token via techniques such as quantization, caching/KV optimizations, batching strategies, and decoding-time optimizations.  

3) Tool Calling System: Catalog, Routing, Validation, Fallback & Observability 

• Architect and implement a production-grade tool calling (function/tool calling) framework:  

• Tool cataloging and metadata/schema design 

• Tool selection/routing and argument construction 

• Parameter validation, result verification, and safe fallback/retry strategies 

• Call-chain tracing, monitoring, and observability to improve success rate and ROI 

4) RL & Reward Modeling for Alignment and Tool-Use Reliability 

• Apply post-training methods such as PPO / DPO / GRPO-like optimization and reward modeling to align the model toward objectives including:  

• semantic understanding 

• tool-use success rate 

• content generation quality and consistency 

• Support both offline and online iteration loops, including policy evaluation, regression checks, and safe deployment gating.  

5) Data Pipeline Automation (Collection, Cleaning, Curation) 

• Design automated pipelines for data collection, filtering, cleaning, de-duplication, labeling/weak supervision, and dataset version management to continuously improve training quality. 

• Ensure datasets support both SFT and preference/RL style post-training.  

6) Rigorous Evaluation, Testing & Iteration 

• Build robust evaluation mechanisms: offline benchmarks, task suites for tool-use, regression tests, and reliability metrics. 

• Drive rapid iteration through A/B comparisons, ablations, and failure analysis, improving both quality and efficiency over time.  

Required Qualifications 

• Strong software engineering skills in Python and C++, including experience building ML training/evaluation pipelines in PyTorch. 

• Hands-on experience in model efficiency and inference optimization (e.g., distillation, quantization, pruning, serving optimization).  

• Experience with high-performance computing and acceleration: CUDA and/or SIMD, profiling and performance tuning.  

• Ability to read and reproduce key ideas from recent papers and implement algorithms with strong experimental discipline. 

• Ability to communicate effectively in both Chinese (Mandarin) and English as the successful person will have to liaise with the our counterparts in China.