Public benchmarks

The labs' evals as open leaderboards. Only verified sandbox runs count — client-reported results cannot be submitted.

Lab: ablation — which component does the work? — hittar-ratt-komponenterG
correct >= · lab: ablation-labb · 0 participants
Lab: build and validate a benchmark — benchmark-valideradG
ok >= · lab: benchmark-labb · 0 participants
Lab: find the circuit — activation patching and ablation in a small transformer — kretsen-hittad-och-verifieradG
ok >= · lab: aktiveringspatchning-labb · 0 participants
Project 6: Reproduce a paper — Dropout (Srivastava et al. 2014) — dropout-minskar-gapG
gap_reduction >= · lab: projekt-6-reproducera-dropout · 0 participants
Lab: LoRA on a small network — train only the adapters — lora-vs-fullF
lora_acc >= · lab: lora-tiny-labb · 0 participants
Project 4: Fine-tune a model with LoRA — lora-r4-forbattrarF
ppl_gain_r4 >= · lab: projekt-4-lora-finjustering · 0 participants
Project 5: Build a memory system for an agent — langtidF
accuracy_with_memory >= · lab: projekt-5-agentminne · 0 participants
Lab: BM25 and a minimal RAG pipeline — recall@1E
recall >= · lab: rag-mini-labb · 0 participants
Lab: build an eval harness — baseline-accE
accuracy >= · lab: evals-harness-labb · 0 participants
Lab: an agent loop with tools, a budget and a trace — task-successE
success >= · lab: agent-loop-labb · 0 participants
Lab: k-nearest neighbours from scratch — digits-accD
accuracy >= · lab: knn-labb · 0 participants
Lab: semantic search with embeddings — recall@3D
recall >= · lab: embeddings-sok-labb · 0 participants
Lab: train a digit classifier in PyTorch — digits-accD
accuracy >= · lab: trana-neuronnat-pytorch-labb · 0 participants
Project 2: Build a neural network from scratch — digits-accuracyD
accuracy >= · lab: projekt-2-neuronnat-fran-grunden · 0 participants
Project 1: Teach a computer to recognise patterns — traffsakerhetC
accuracy >= · lab: projekt-1-monster · 0 participants