SearcharxivSearch

arXiv subjects

Abir Harrasse

Publications and source records attributed to Abir Harrasse.

8 recordsLinked to original sources

CLT-Forge: A Scalable Library for Cross-Layer Transcoders and Attribution Graphs

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.

cs.LG

Reasoning Fine-Tuning Induces Persistent Latent Policy States

Reasoning-specialized language models show large performance gains over base models, yet the internal changes responsible for improved multi-step reasoning remain poorly understood. It is unclear whether reasoning fine-tuning improves local token-level competence or globally reorganizes how models structure inference over time. We address this question by modeling Chain-of-Thought reasoning as a switching dynamical system (SDS), in which internal representations evolve under discrete latent policy states. Our framework combines time-aware contrastive representation learning with discrete regime discovery to recover latent policies from activation trajectories. Across four benchmarks and model scales from 1.5B to 32B parameters, reasoning-fine-tuned models exhibit richer latent-policy organization than their base counterparts, characterized by more differentiated transition structure and model-dependent changes in state utilization, persistence, and mixing. The recovered regimes exhibit functional specialization aligned with distinct reasoning stages, and extensive controls confirm that their structure is not explained by correctness, representation learning, or modeling priors, but depends on the coherent temporal organization of reasoning trajectories. Causal interventions further show that the regimes are functionally meaningful: state-swap ablations reduce one-step predictive fit, while transplanting reasoning dynamics into base models improves performance on challenging reasoning problems. Finally, SDS-guided pruning of failure-prone reasoning prefixes outperforms self-consistency in 11 of 12 model-dataset settings, with gains of up to 12.5 percentage points. Together, our results suggest that reasoning fine-tuning globally reorganizes latent dynamics, offering a new lens for mechanistic analysis and process-level control of reasoning models.

cs.CL

STRIDE: Training Data Attribution via Sparse Recovery from Subset Perturbations

Training Data Attribution (TDA) seeks to trace a model's predictions back to its training data. The gold standard for TDA relies on causal interventions, observing how a model changes when data is added or removed, but repeated retraining is computationally challenging for Large Language Models (LLMs). Consequently, most approaches approximate this effect in the parameter space using gradients. However, tracking gradients across billions of parameters is not only prohibitively expensive but relies on local approximations. In this work, we propose a shift: rather than estimating parameter changes, we model the functional effect of training data in the activation space. We introduce STRIDE (Steering-based Training Data Influence Decomposition), a framework that formulates TDA as a sparse recovery problem in the spirit of compressive sensing. STRIDE learns lightweight "steering operators" that mimic the behavioral shift caused by training on data subsets. By measuring how these operators perturb test predictions, we recover individual training example influences via sparse linear decomposition. STRIDE achieves state-of-the-art for LLM pre-training attribution while being an order of magnitude ($13\times$) faster than previous art. We further validate its practical utility through downstream applications including data selection, data contamination, and qualitative analysis.

cs.LG

Curveball Steering: The Right Direction To Steer Isn't Always Linear

Activation steering is a widely used approach for controlling large language model (LLM) behavior by intervening on internal representations. Existing methods largely rely on the Linear Representation Hypothesis, assuming behavioral attributes can be manipulated using global linear directions. In practice, however, such linear interventions often behave inconsistently. We question this assumption by analyzing the intrinsic geometry of LLM activation spaces. Measuring geometric distortion via the ratio of geodesic to Euclidean distances, we observe substantial and concept-dependent distortions, indicating that activation spaces are not well-approximated by a globally linear geometry. Motivated by this, we propose "Curveball steering", a nonlinear steering method based on polynomial kernel PCA that performs interventions in a feature space, better respecting the learned activation geometry. Curveball steering consistently outperforms linear PCA-based steering, particularly in regimes exhibiting strong geometric distortion, suggesting that geometry-aware, nonlinear steering provides a principled alternative to global, linear interventions.

cs.AI

Tracing Multilingual Representations in LLMs with Cross-Layer Transcoders

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 does performance favor the dominant training language? To address this, we train models on different multilingual mixtures and analyze their internal mechanisms using Cross-Layer Transcoders (CLTs) and Attribution Graphs. Our results reveal multilingual shared representations: the model employs highly similar features across languages, while language-specific decoding emerges in later layers. Training models without English shows identical multilingual shared space structures. Decoding relies partly on a small set of high-frequency features in the final layers, which linearly encode language identity from early layers. Intervening on these features allows one language to be suppressed and another substituted. Finally, to explain non-English failures, we perform a Model-Diffing experiment: underperformance arises from dim late-layer features, weak middle-layer clusters, and tokenizer bias toward English that forces early layers to specialize in word reassembly. Finetuning strengthens these features and their links, improving token assembly and language-specific decoding, providing a mechanistic explanation for multilingual gaps. Our models and CLTs are available at https://huggingface.co/collections/CausalNLP/multilingual-clts and https://huggingface.co/collections/CausalNLP/multilingual-gpt2-models. Our code is available at: https://github.com/abirharrasse/MultilingualCLTs

cs.CL

Debate, Deliberate, Decide (D3): A Cost-Aware Adversarial Framework for Reliable and Interpretable LLM Evaluation

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 kappa), 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.

cs.CL

TinySQL: A Progressive Text-to-SQL Dataset for Mechanistic Interpretability Research

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.

cs.LG

Activation Space Interventions Can Be Transferred Between Large Language Models

The study of representation universality in AI models reveals growing convergence across domains, modalities, and architectures. However, the practical applications of representation universality remain largely unexplored. We bridge this gap by demonstrating that safety interventions can be transferred between models through learned mappings of their shared activation spaces. We demonstrate this approach on two well-established AI safety tasks: backdoor removal and refusal of harmful prompts, showing successful transfer of steering vectors that alter the models' outputs in a predictable way. Additionally, we propose a new task, \textit{corrupted capabilities}, where models are fine-tuned to embed knowledge tied to a backdoor. This tests their ability to separate useful skills from backdoors, reflecting real-world challenges. Extensive experiments across Llama, Qwen and Gemma model families show that our method enables using smaller models to efficiently align larger ones. Furthermore, we demonstrate that autoencoder mappings between base and fine-tuned models can serve as reliable ``lightweight safety switches", allowing dynamic toggling between model behaviors.

cs.AI