Provides 999,847 persona records—599,847 grounded from real sources and 400,000 synthetic—each encoded as 1,290 categorical attributes packed into 645-byte Parquet blobs. Includes a codebook, postings index, and calibration/audit artifacts; decode with pyarrow and persona_codes.schema.json.
Measures whether models produce valid JSON/YAML that strictly follow a requested schema across diverse, naturally phrased prompts. Contains 2,000 frozen test prompts with binary structural validation (no constrained decoding), focusing on schema compliance and edge cases like escaping, wrapper keys, and fenced code blocks.
Provides ~1,467 single-speaker Sanskrit chant audio clips (≈5.3 hours) with aligned transcripts and prosodic metadata for meter-aware TTS training. Two recording/config styles (style_a/style_b), 24 kHz mono WAVs, metadata includes Devanagari, SLP1, Kannada text, meter, duration, session/take. CC-BY-4.0.
Detects and redacts personally identifiable information (PII) in user-typed text on-device, replacing sensitive values with stable placeholders before any data leaves the browser. Uses a small quantized ONNX token-classification model plus deterministic recognizers for structured identifiers, and applies a policy-driven keep-set for coarse geography.
Provides 300 adversarial "metacognitive-trap" problems to measure whether LLMs notice and recover from their own reasoning errors. Combines multiple-choice vulnerability tests with free-form adapter-gain evaluation and ships per-model metacognition adapters for frozen-base probing.
Treats hybrid layer selection as a budget-constrained subset optimization and introduces FlashMorph: a pipeline that equips each transformer layer with a linear-attention branch, jointly optimizes layerwise gates on synthetic long-context retrieval data, then discretizes, distills, and finetunes—achieving strong long-context recall using only 20M selection tokens.
Trains a transformer-based graph encoder with RL-guided adaptive masking so retrieved subgraphs embed relationships that better align with frozen LLM text encoders, improving GraphRAG performance with non-parametric retrievers on GraphQA benchmarks.
Evaluates how long-term memory in LLM agents amplifies sycophantic behavior and when memory should or should not influence decisions. Provides five targeted tasks, 1,550 standardized samples, an evaluation pipeline, and baseline adapters to test memory use, conflicts, scope, updates, and personalization.
Predicts per-request MoE expert footprints from prefill activations and routes decode requests to workers that maximize expert-locality, lowering decode latency by combining offline K-means partitioning with online locality-band routing and a KV-block–coindexed signature cache.
Compiles natural-language function specifications into compact, locally-executable neural programs (PAW) that run on a small frozen interpreter; a 4B compiler emits LoRA adapters for a 0.6B runtime to provide offline, low-memory fuzzy text functions.
Introduces a bounded-memory, typed-retrieval contract for long-horizon LLM agents and evaluates it in Slay the Spire 2 — assembling per-decision prompts from five typed slots rather than appending raw transcripts. Key outputs include ablationable memory layers, 298 labeled trajectories, and reproducible analysis scripts.
A 1B-parameter 'Thinking' language model fine-tuned on Fable 5 to improve coding and instruction-following; supports chain-of-thought style outputs, XML tool-call format, and up to 128K-token context, with GGUF builds for single-GPU local deployment.