Independent Paused
Learning with Verifiable Rewards
Does GRPO earn its cost over a properly built SFT baseline on GSM8K? The model and training foundations are built and tested; the study is paused before the first SFT run.
Investigations and builds
Projects
Implementations, experiments, and published collaborative work. Each page says who did the work, what is actually built and tested, and what is still planned.
How statuses are used
- Completed
- Evidence exists and supports the described scope.
- In progress
- Current, active work.
- Paused
- Intentionally not active; what remains is listed on the page.
- Planned
- Not implemented or tested yet.
Selected investigations
A few projects where the code, method, and lessons are documented.
Published collaboration Completed
Tailor at Farfetch
Size recommendation as sequence classification: LSTMs and attention over a customer's history to model evolving fit preferences. A co-authored paper at FashionXRecSys 2023.
Implementations
Smaller builds that work through a method or a system, with their limits stated.
Independent
LoRA and DoRA Implementation
LoRA and DoRA layers written from scratch in PyTorch and trained on an MNIST MLP, with an audit of where the code diverges from the papers.
Independent
RAG System with LlamaIndex, Elasticsearch & Llama3
A local-first RAG pipeline on Elasticsearch, LlamaIndex, and Llama3, and an account of what worked, what didn't, and what an evaluation would need.
Earlier experiments
Early builds kept for reference rather than as evidence.
Independent
Large Language Models with MLX
An early command-line chat that runs 7B models locally with Apple MLX. No benchmarks or evaluation; kept for reference.
No projects match. Try another word, or show all projects.