Multilingual Large Language Models (LLMs) can process many languages, yet how they internally represent this diversity remains unclear. Do they form shared multilingual representations with language-specific decoding, and if so, why do multilingual performance gaps persist? To address this question, we train models on different multilingual mixtures and analyze their internal mechanisms using Cross-Layer Transcoders (CLTs) and Attribution Graphs. Our results reveal a consistent three-phase structure across models: early layers encode language-specific information, middle layers converge to highly shared multilingual representations, and final layers re-specialize for language-specific decoding.

Training models without English produces the same shared multilingual structures, suggesting that these representations emerge independently of English dominance. We further show that decoding relies on a small set of high-frequency late-layer features that linearly encode language identity from earlier layers, and that intervening on these features allows one language to be suppressed and another substituted. Finally, through a Model-Diffing experiment, we show that non-English failures arise from weak late-layer decoding features, weak middle-layer clusters, and tokenizer fragmentation that forces early layers to specialize in word reassembly. Finetuning strengthens these features and their connections, improving token assembly and language-specific decoding, providing a mechanistic explanation for multilingual performance gaps.

Mechanistic interpretability seeks to understand how Large Language Models (LLMs) represent and process information. Recent approaches based on dictionary learning and transcoders enable representing model computation in terms of sparse, interpretable features and their interactions, giving rise to feature attribution graphs. However, these graphs are often large and redundant, limiting their interpretability in practice. Cross-Layer Transcoders (CLTs) address this issue by sharing features across layers while preserving layer-specific decoding, yielding more compact representations, but remain difficult to train and analyze at scale. We introduce an open-source library for end-to-end training and interpretability of CLTs. Our framework integrates scalable distributed training with model sharding and compressed activation caching, a unified automated interpretability pipeline for feature analysis and explanation, attribution graph computation using Circuit-Tracer, and a flexible visualization interface. This provides a practical and unified solution for scaling CLT-based mechanistic interpretability. Our code is available at https://github.com/LLM-Interp/CLT-Forge.

The evaluation of Large Language Models (LLMs) remains challenging due to inconsistency, bias, and the absence of transparent decision criteria in automated judging. We present Debate, Deliberate, Decide (D3), a cost-aware, adversarial multi-agent framework that orchestrates structured debate among role-specialized agents (advocates, a judge, and an optional jury) to produce reliable and interpretable evaluations. D3 instantiates two complementary protocols: (1) Multi-Advocate One-Round Evaluation (MORE), which elicits k parallel defenses per answer to amplify signal via diverse advocacy, and (2) Single-Advocate Multi-Round Evaluation (SAMRE) with budgeted stopping, which iteratively refines arguments under an explicit token budget and convergence checks.We develop a probabilistic model of score gaps that (i) characterizes reliability and convergence under iterative debate and (ii) explains the separation gains from parallel advocacy. Under mild assumptions, the posterior distribution of the round-r gap concentrates around the true difference and the probability of mis-ranking vanishes; moreover, aggregating across k advocates provably increases expected score separation. We complement theory with a rigorous experimental suite across MT-Bench, AlignBench, and AUTO-J, showing state-of-the-art agreement with human judgments (accuracy and Cohen’s 𝜅), reduced positional and verbosity biases via anonymization and role diversification, and a favorable cost-accuracy frontier enabled by budgeted stopping. Ablations and qualitative analyses isolate the contributions of debate, aggregation, and anonymity.Together, these results establish D3 as a principled, practical recipe for reliable, interpretable, and cost-aware LLM evaluation.

Mechanistic interpretability research faces a gap between analyzing simple circuits in toy tasks and discovering features in large models. To bridge this gap, we propose text-to-SQL generation as an ideal task to study, as it combines the formal structure of toy tasks with real-world complexity. We introduce TinySQL, a synthetic dataset, progressing from basic to advanced SQL operations, and train models ranging from 33M to 1B parameters to establish a comprehensive testbed for interpretability. We apply multiple complementary interpretability techniques, including Edge Attribution Patching and Sparse Autoencoders, to identify minimal circuits and components supporting SQL generation. We compare circuits for different SQL subskills, evaluating their minimality, reliability, and identifiability. Finally, we conduct a layerwise logit lens analysis to reveal how models compose SQL queries across layers: from intent recognition to schema resolution to structured generation. Our work provides a robust framework for probing and comparing interpretability methods in a structured, progressively complex setting.