LLMs acquire vast amounts of knowledge during pre-training, but often lack the specialized knowledge needed to answer questions from niche sources such as manuals or technical documents unseen during pre-training. Continued pre-training (CPT) is widely used to inject such knowledge into model parameters. However, niche documents seldom repeat facts, making it difficult for CPT to robustly acquire such knowledge. Recent works address this by generating multiple paraphrases of the new knowledge, but paraphrasing is computationally expensive and typically requires powerful LLMs. In this work, we introduce KItCAT: Knowledge Injection via Corrupted Auto-regressive Training, a lightweight training strategy that reduces the need for paraphrasing in decoder-only LLMs. KItCAT augments standard next-token prediction by stochastically corrupting the input sequence. During training, a random subset of input tokens is replaced with other vocabulary tokens while the original next-token labels are kept unchanged. This simple intervention generates diverse training inputs from each sample, enabling large-scale data augmentation at negligible cost. We show that KItCAT consistently improves over CPT across multiple datasets and model families. Code is available at https://github.com/meghanadhpulivarthi/KItCAT.

Deploying large language models in production often requires constraining outputs to structured formats such as JSON or XML, and prior work treats the resulting accuracy loss as an inherent "structure tax". We re-examine this claim by evaluating a battery of models, datasets and schemas, measuring task accuracy, confidence calibration, and hidden-state geometry. The tax turns out to depend on schema design rather than on structure per se: reasoning-first field ordering matches or exceeds free-form accuracy, while answer-first ordering causes steep drops, particularly in smaller models. Format sensitivity scales inversely with a task’s own structural constraints, and schemas that preserve reasoning order also improve calibration with CKA showing greater separability between correct and incorrect representations in middle transformer layers. Our findings indicate that properly designed structured formats can match or exceed free-form performance, reframing the critical question from ’whether to structure’ to ’how to structure’ for optimal reasoning preservation.

LLM-as-a-Judge is the standard paradigm for NLP evaluation, yet its systemic reliability remains poorly understood despite being widely treated as a deterministic ground truth. We present a comprehensive reliability audit, stress-testing six frontier models across four benchmarks, five prompt formats, two presentation orders, three sampling temperatures, and ten repetitions per condition. Our empirical analysis reveals severe vulnerabilities: verdicts change across identical replications at temperature zero, position-order swaps flip the majority of verdicts on challenging tasks, and the most deterministic judge achieves perfect consistency by trivially repeating incorrect verdicts, agreeing with ground truth only 51% of the time. To formalize these multi-faceted failure modes, we introduce the trustworthy verdict rate (T), a unified metric capturing the joint probability that an evaluation is reproducible, order-invariant, and accurate. Using T , we derive a theoretical upper bound on accuracy imposed by position bias and show that reliability is item-specific rather than model-level. Finally, we demonstrate that shifting from pairwise win-rate to holistic rubric scoring improves trustworthiness more than any single-format prompting intervention, offering an actionable framework for robust NLP evaluation.

Large language models are increasingly adapted to downstream tasks through fine-tuning. Full supervised fine-tuning (SFT) and parameter-efficient fine-tuning (PEFT) methods, such as Low-Rank Adaptation (LoRA), are two dominant approaches. While PEFT methods are widely used for their computational efficiency, the implications of their configurations (e.g., rank) remain under-explored in downstream Q&A tasks and generalisation. In this work, we perform a comprehensive evaluation across multiple reasoning and recall datasets, conducting a rank sweep to quantify the trade-off between SFT and PEFT. We also compare the accuracy of PEFT and SFT models across in-domain and out-of-domain adaptation, highlighting distinct generalisation behaviour and task-specific forgetting. We demonstrate that LoRA achieves competitive and in some cases superior performance compared to SFT, particularly on reasoning tasks at specific rank values. Additionally, we analyze the internal representations via spectral features and layer-wise attention structures, offering insights into representational drift and structural changes in attention patterns.

An emergent research trend explores the use of Large Language Models (LLMs) as the backbone of agentic systems (e.g., SWE-Bench, Agent-Bench). To fulfill LLMs’ potential as autonomous agents, they must be able to identify, call, and interact with a variety of external tools and application program interfaces (APIs). This capability of LLMs, commonly termed function calling, leads to a myriad of advantages such as access to current and domain-specific information in databases and the outsourcing of tasks that can be reliably performed by tools. In this work, we introduce Granite-20B-FunctionCalling, a model trained using a multi-task training approach on seven fundamental tasks encompassed in function calling. Our comprehensive evaluation on multiple out-of-domain datasets, which compares Granite-20B-FunctionCalling to more than 15 other best proprietary and open models, shows that Granite-20B-FunctionCalling has better generalizability on multiple tasks across seven different evaluation benchmarks. Moreover, Granite-20B-FunctionCalling shows the best performance among all open models and ranks among the top on the Berkeley Function Calling Leaderboard (BFCL).

A major concern in using deep learning based generative models for document-grounded dialogs is the potential generation of responses that are not faithful to the underlying document. Existing automated metrics used for evaluating the faithfulness of response with respect to the grounding document measure the degree of similarity between the generated response and the document’s content. However, these automated metrics are far from being well aligned with human judgments. Therefore, to improve the measurement of faithfulness, we propose a new metric that utilizes (Conditional) Point-wise Mutual Information (PMI) between the generated response and the source document, conditioned on the dialogue. PMI quantifies the extent to which the document influences the generated response – with a higher PMI indicating a more faithful response. We build upon this idea to create a new decoding technique that incorporates PMI into the response generation process to predict more faithful responses. Our experiments on the BEGIN benchmark demonstrate an improved correlation of our metric with human evaluation. We also show that our decoding technique is effective in generating more faithful responses when compared to standard decoding techniques on a set of publicly available document-grounded dialog datasets.

The rapidly growing market demand for automatic dialogue agents capable of goal-oriented behavior has caused many tech-industry leaders to invest considerable efforts into task-oriented dialog systems. The success of these systems is highly dependent on the accuracy of their intent identification – the process of deducing the goal or meaning of the user’s request and mapping it to one of the known intents for further processing. Gaining insights into unrecognized utterances – user requests the systems fails to attribute to a known intent – is therefore a key process in continuous improvement of goal-oriented dialog systems. We present an end-to-end pipeline for processing unrecognized user utterances, deployed in a real-world, commercial task-oriented dialog system, including a specifically-tailored clustering algorithm, a novel approach to cluster representative extraction, and cluster naming. We evaluated the proposed components, demonstrating their benefits in the analysis of unrecognized user requests.

In this paper we present the Exemplar Encoder-Decoder network (EED), a novel conversation model that learns to utilize similar examples from training data to generate responses. Similar conversation examples (context-response pairs) from training data are retrieved using a traditional TF-IDF based retrieval model and the corresponding responses are used by our decoder to generate the ground truth response. The contribution of each retrieved response is weighed by the similarity of corresponding context with the input context. As a result, our model learns to assign higher similarity scores to those retrieved contexts whose responses are crucial for generating the final response. We present detailed experiments on two large data sets and we find that our method out-performs state of the art sequence to sequence generative models on several recently proposed evaluation metrics.

An interactive Question Answering (QA) system frequently encounters non-sentential (incomplete) questions. These non-sentential questions may not make sense to the system when a user asks them without the context of conversation. The system thus needs to take into account the conversation context to process the question. In this work, we present a recurrent neural network (RNN) based encoder decoder network that can generate a complete (intended) question, given an incomplete question and conversation context. RNN encoder decoder networks have been show to work well when trained on a parallel corpus with millions of sentences, however it is extremely hard to obtain conversation data of this magnitude. We therefore propose to decompose the original problem into two separate simplified problems where each problem focuses on an abstraction. Specifically, we train a semantic sequence model to learn semantic patterns, and a syntactic sequence model to learn linguistic patterns. We further combine syntactic and semantic sequence models to generate an ensemble model. Our model achieves a BLEU score of 30.15 as compared to 18.54 using a standard RNN encoder decoder model.