← Back
Home › Systems › Frontier AI research

interpretability-platform

features and circuits at model scale

SAEs, attribution graphs, Gemma Scope

SystemTechniques · 5Primitives · 36 interpretability-platfo… Data loading pipelinesmachine-learning · medium: Turn files on disk into shuffled, sharded, resumable GPU batches fast enough that the accelerator never waits.›LLM-as-judgemachine-learning · medium: Grade open-ended outputs with a language model under a rubric, then correct its known biases by order swapping, voting and calibration against human labels.›Experiment trackingml-training · medium: Logging params, metrics, artifacts and code versions per run with an append-only metric store and comparisonHNSWsearch-sort · medium: Hierarchical Navigable Small World: a layered proximity graph searched greedily from a sparse top layer; the default in-memory approximate nearest-neighbour index.›Chunked array storagestorage · medium: Cut an N-dimensional array into a regular grid of independently compressed blocks, so a slice read touches only the overlapping blocks and writers to different blocks run in parallel.›Object storagecloud · low: S3-compatible blob storage at cents per GB-month with 11 nines durability; pick by egress price, request price and latency, not by storage price alone.›List virtualization (windowing)graphics · low: Create and lay out only the items inside the viewport plus a little overscan, and fake the full scroll extent, so frame cost depends on the screen, not on N.›Embeddingsmachine-learning · low: A learned table mapping discrete ids to dense vectors: gather forward, sparse scatter-add backward, row-wise optimizers to keep huge tables cheap.›PCA and SVDmachine-learning · low: Find orthogonal directions of maximum variance by an SVD of the centred data; randomized SVD gets the top k components in a few passes.›AdamWml-training · low: Adam's per-parameter adaptive steps with weight decay applied directly to the weights; the default optimizer for transformers and LLM pretraining.›Mixed-precision trainingml-training · low: Run matmuls in bf16/fp16/fp8 on tensor cores while keeping master weights, optimizer state and reductions in fp32; about 2x speed, half the activation memory.›Reverse-mode automatic differentiat…ml-training · low: Record the forward computation on a tape, then sweep it backwards once to get the gradient of a scalar loss for all inputs.›Transformer blockml-training · low: The decoder-only pre-norm block (attention plus MLP on a residual stream) behind every modern LLM, with exact shapes, parameter, FLOP and KV-cache counts.›ZeRO and FSDP shardingml-training · low: Data parallelism that shards optimizer state, gradients and weights across ranks and gathers weights just in time; per-GPU model memory drops from 16Ψ to 16Ψ/N.›Top-k selectionsearch-sort · low: Keep the k best of n items without sorting all n - bounded min-heap for streams, quickselect for arrays, threshold pruning to skip most items.›Columnar file formatstorage · low: Row groups, column chunks, pages, encodings and footer stats (Parquet/ORC style) (listed in Moved to the wiki)mmap and the page cachehardware-os · low: How read() and mmap reach file data through the kernel page cache, what readahead and madvise do, and why databases use their own buffer pool instead.›Sequence packingml-training · low: Packing variable-length sequences into fixed-size batches with attention masks to avoid padding wastePRNGs (PCG, xoshiro)numerics · low: Small, fast, statistically strong non-cryptographic generators: PCG32 and xoshiro256**, with seeding, parallel streams and unbiased ranges.›Token samplingml-training · low: Turn a vector of scores into one chosen item with temperature, top-k, top-p, min-p, repetition penalties and a seed.›Annotator agreementstatistics · low: Inter-annotator agreement measures: Cohen and Fleiss kappa, Krippendorff alpha, gold-question accuracy+ 20 more primitives

Similar systems

Share the most building blocks with this one.

surrogate-modeling 25 %ml-weather-model 20 %robot-policy 20 %