Amal Zouaq

Reasoning models achieve strong performance on challenging tasks by generating explicit intermediate reasoning traces before producing a fin… (see more)al answer. Yet the internal structure of representation space when reasoning remains poorly understood: how do a model's hidden representations differ during thinking versus the embeddings of the input prompt, and can this structure be exploited to elicit stronger reasoning at inference time? We show that both input embeddings and thinking embeddings (mean-pooled last-layer hidden states over the prompt and reasoning trace, respectively) exhibit extremely high conicity, with all vectors clustering tightly around a single mean direction. Crucially, these mean input and thinking directions are non-collinear, with thinking embeddings occupying a geometrically distinct region of embedding space across many different models and benchmark tasks. This observation motivates casting the input-to-thinking transition as a rotation problem admitting a closed-form solution via orthogonal Procrustes analysis. We propose Rotate2Think, a training-free method that estimates this rotation from a small set of correctly solved examples and injects the resulting synthetic thinking vector between thinking delimiters at inference time, providing a geometric primer at the onset of the reasoning trace. Evaluated across multiple benchmarks and model families, Rotate2Think improves accuracy in 30 of 32 model-benchmark configurations across mathematics, science, and code tasks, and generalizes zero-shot to multimodal reasoning on MATH-Vision.

2026-06-01

arXiv (preprint)

DeSQ: Decomposition-based SPARQL Query Generation

Papa Abdou Karim Karou Diallo

Aditya Sharma

Neshat Elhami Fard

Dominant approaches to Knowledge Base Question Answering (KBQA) fall into two categories. First is the generation of a formal query that suf… (see more)fers from brittleness and limited explainability, and the second is direct answer retrieval through KB exploration that is computationally costly and prone to hallucination. To combine the strengths of both paradigms while mitigating their respective weaknesses, we introduce DeSQ (Decomposition-based SPARQL Query Generation), a KB-agnostic framework that operates in three stages. First, it decomposes complex questions into Atomic Constraints (ACs) that mirror the relational structure of the underlying KB. Second, it generates a two-part structured output: (a) Mapping of each AC to its corresponding SPARQL Fragment, using standardized variable and URIs placeholders, and (b) URIs Grounding block describing each placeholder. Third, it assembles these fragments into a complete SPARQL query. DeSQ surpasses state-of-the-art approaches on four out of five major benchmarks and demonstrates superior robustness to lexical variation. Beyond performance gains, our framework greatly simplifies evaluation by eliminating the need for a live KB endpoint, and its structured output enables fine-grained error analysis, allowing more targeted interventions for improvement.

2026-05-28

arXiv (preprint)

Probabilistic Calibration Is a Trainable Capability in Language Models

Sruthi Kuriakose

Language models are increasingly used in settings where outputs must satisfy user-specified randomness constraints, yet their generation pro… (see more)babilities are often poorly calibrated to those targets. We study whether this capability can be improved directly through fine-tuning. Concretely, we fine-tune language models on synthetic prompts that require sampling from mathematical distributions, and compare two Calibration Fine-Tuning variants: a soft-target method that converts the desired output distribution into trie-derived next-token targets, and a hard-target method that trains on sampled completions from the same target distribution. Across 12 models spanning four families, both methods substantially improve structured-sampling fidelity on held-out distribution families and unseen parameter settings, showing that probabilistic calibration is a trainable capability. Under our selected training configurations, the two methods exhibit different empirical profiles: hard-target fine-tuning is often strongest on structured numeric sampling, while soft-target fine-tuning performs better on broader stochastic generation benchmarks, including open-ended random generation, multiple-choice answer-position balancing, and NoveltyBench. The gains sometimes reduce downstream capability, especially arithmetic reasoning, with costs varying by model. Overall, our results show that probabilistic calibration can be improved through fine-tuning, with our hard-target configuration favoring exact numeric fidelity and our soft-target configuration favoring broader stochastic transfer. Code is available at https://github.com/chandar-lab/calibration-finetuning.

2026-05-11

arXiv (preprint)

FRASE: Frame-based Structured Representations for Generalizable SPARQL Query Generation

Papa Abdou Karim Karou Diallo

2026-05-10

Language Resources and Evaluation Conference (published)

LLMs Can't Play Hangman: On the Necessity of a Private Working Memory for Language Agents

Davide Baldelli

Ali Parviz

A. Chandar

2026-01-10

ArXiv (preprint)

Combining Domain and Alignment Vectors Provides Better Knowledge-Safety Trade-offs in LLMs

Megh Thakkar

Quentin Fournier

Matthew D Riemer

Pin-Yu Chen

Payel Das

A. Chandar

2025-06-30

Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) (published)

Combining Domain and Alignment Vectors to Achieve Better Knowledge-Safety Trade-offs in LLMs

Megh Thakkar

Yash More

Quentin Fournier

Matthew D Riemer

Pin-Yu Chen

Payel Das

A. Chandar

Chandar Research Lab

Mila - Québec

U. Montŕeal

AI Institute

Ibm Research

Polytechnique Montréal

There is a growing interest in training domain-expert LLMs that excel in specific technical fields compared to their general-purpose instruc… (see more)tion-tuned counterparts. However, these expert models often experience a loss in their safety abilities in the process, making them capable of generating harmful content. As a solution, we introduce an efficient and effective merging-based alignment method called \textsc{MergeAlign} that interpolates the domain and alignment vectors, creating safer domain-specific models while preserving their utility. We apply \textsc{MergeAlign} on Llama3 variants that are experts in medicine and finance, obtaining substantial alignment improvements with minimal to no degradation on domain-specific benchmarks. We study the impact of model merging through model similarity metrics and contributions of individual models being merged. We hope our findings open new research avenues and inspire more efficient development of safe expert LLMs.

2024-10-09

NeurIPS.cc/2024/Workshop/AFM (poster)

openreview.net

A Deep Dive into the Trade-Offs of Parameter-Efficient Preference Alignment Techniques

Megh Thakkar

Quentin Fournier

Matthew D Riemer

Pin-Yu Chen