PsiBorg is a universally promptable foundation model trained on visual and robot experience. We study policy inference, simulation, and progress tracking through one probabilistic autoregressive model...

Quantum circuit optimization for fault-tolerant computing requires exact functional equivalence while minimizing expensive non-Clifford resources such as T gates. We study this problem using a compact 44.8M-parameter encoder-decoder transformer with structured circuit tokenization, evaluating on par

In autoregressive large language models (LLMs), temporal straightening offers an account of how the next-token prediction objective shapes representations. Models learn to progressively straighten the...

Integer-valued autoregressive (INAR) models are a standard framework for count time series, but their innovation law is usually fixed in advance and their th...

Laya answers typed multiple-choice questions in one forward pass by scoring a hidden state at one [MASK] marker per option. We re-engineer the checkpoint (421M parameters; a 2-layer option head with no positional tensor; a documented but unused parallel layout) and test whether the answer depends o…

Speculative decoding has become a standard way to accelerate LLM inference: a lightweight drafter proposes tokens that the target verifies in one parallel pass. Existing drafters face trade-offs that limits their speedups: autoregressive drafters generate tokens one at a time, so drafting cost grow…

Introduction: Large language model inference is frequently bottlenecked by sequential autoregressive token generation and high GPU memory bandwidth demands. Speculative decoding addresses this latency challenge by coupling a lightweight draft model with a large target model to verify multiple token predictions in parallel. Learning Objectives & Secrets: Objective 1 Master the architectural trade-offs between standard autoregressive decoding and speculative verification pipelines.

The open-source System One decision model. Sub-15ms, non-autoregressive, local drop-in alternative to TypeSafe Jev.

Speculative decoding has become a standard way to accelerate LLM inference: a lightweight drafter proposes tokens that the target verifies in one parallel pass. Existing drafters face trade-offs that limits their speedups: autoregressive drafters generate tokens one at a time, so drafting cost grow…

Provides tools for modelling, forecasting and visualising long-memory functional time series with the functional autoregressive fractionally integrated moving-average model. The model is fitted in three steps, each of which can also be used on its own. First, the memory parameter is estimated from the leading principal component scores, for example by local Whittle or by bias-corrected local polynomial Whittle with noise. Second, the autoregressive and moving-average orders of the fractionally differenced curves are selected by a functional information criterion, functional Box-Jenkins or cross-validation. Third, the operators are estimated. The point forecasts come with conformal prediction bands that quantify their uncertainty. Visualisation tools include improved rainbow plots and plots of the fitted model and its forecasts.