AI Jobs Map

Cliead · South Africa

Machine Learning Research Engineer- SOUTH AFRICA (REMOTE)

Remoteentry_levelfull timePosted 20 days ago
Apply on LinkedInOpens the original posting. AI Jobs Map never asks for your details.

Stack mentioned

etlpytorchtransformerspythonmachine-learningreinforcement-learninghugging-facefine-tuning

Our client is looking for a highly capable Machine Learning Research Engineer to join the AI R&D team and work on the development and optimization of next-generation language models for specialized AI applications.

You will work across model training, fine-tuning, post-training, data development, evaluation, and inference optimization. This is a hands-on research engineering role for someone who enjoys running experiments, understanding why models behave the way they do, and turning research ideas into working models.

What You will Do

- Fine-tune and post-train open-source language models.

- Experiment with supervised fine-tuning, continued pretraining, distillation, preference optimization, and reinforcement learning.

- Develop high-quality training datasets and synthetic data pipelines.

- Generate and curate high-quality reasoning and instruction datasets.

- Experiment with different model architectures, training strategies, and hyperparameters.

- Build rigorous evaluation benchmarks to measure model capabilities.

- Analyze model failures and identify opportunities for improvement.

- Optimize models for inference latency, memory usage, throughput, and cost.

- Experiment with models ranging from hundreds of millions to several billion parameters.

- Research techniques for transferring capabilities from larger models into smaller models.

- Work closely with AI systems engineers to integrate models into production AI systems.

- Reproduce relevant academic research and translate promising ideas into experiments.

- Maintain clear experiment tracking, documentation, and reproducible training pipelines.

Required Experience

- Strong experience with PyTorch and modern ML frameworks.

- Hands-on experience fine-tuning or training language models.

- Strong understanding of transformer architectures.

- Experience with Hugging Face Transformers and related tooling.

- Experience with SFT, LoRA/QLoRA, knowledge distillation, or other post-training methods.

- Strong Python programming skills.

- Experience preparing and processing large-scale datasets.

- Strong understanding of model evaluation and benchmarking.

- Ability to design controlled experiments and interpret results.

Nice to Have

- Experience with models under 7B parameters.

- Experience with continued pretraining.

- Experience with DPO, GRPO, RLHF or related methods.

- Experience generating synthetic training data.

- Experience with coding models or developer-focused AI.

- Experience with distributed training, FSDP, DeepSpeed or similar.

- Experience with model quantization and inference optimization.

- Publications, research projects, or meaningful open-source contributions.

What We are Looking For

We care less about titles and more about what you have actually built.

If you have taken an open model, trained it on a new dataset, improved its capabilities, diagnosed why it failed, and iterated until you got a measurable improvement, we want to hear from you.

More jobs at Cliead