Unveiling the Secret Recipe: A Guide For Supervised Fine-Tuning Small LLMs

The rise of large language models (LLMs) has created a significant disparity: industrial research labs with their computational resources, expert teams, and advanced infrastructures, can effectively fine-tune LLMs, while individual developers and small organizations face barriers due to limited resources. In this paper, we aim to bridge this gap by presenting a comprehensive study on supervised fine-tuning of LLMs using instruction-tuning datasets spanning diverse knowledge domains and skills. We focus on small-sized LLMs (3B to 7B parameters) for their cost-efficiency and accessibility. We explore various training configurations and strategies across four open-source pre-trained models. We provide detailed documentation of these configurations, revealing findings that challenge several common training practices, including hyperparameter recommendations from TULU and phased training recommended by Orca. Key insights from our work include: (i) larger batch sizes paired with lower learning rates lead to improved model performance on benchmarks such as MMLU, MTBench, and Open LLM Leaderboard; (ii) early-stage training dynamics, such as lower gradient norms and higher loss values, are strong indicators of better final model performance, enabling early termination of sub-optimal runs and significant computational savings; (iii) through a thorough exploration of hyperparameters like warmup steps and learning rate schedules, we provide guidance for practitioners and find that certain simplifications do not compromise performance; and (iv) we observed no significant difference in performance between phased and stacked training strategies, but stacked training is simpler and more sample efficient. With these findings holding robustly across datasets and models, we hope this study serves as a guide for practitioners fine-tuning small LLMs and promotes a more inclusive environment for LLM research.

Quantifying reliance on external information over parametric knowledge during Retrieval Augmented Generation (RAG) using mechanistic analysis

Retrieval Augmented Generation (RAG) is a widely used approach for leveraging external context in several natural language applications such as question answering and information retrieval. Yet, the exact nature in which a Language Model (LM) leverages this non-parametric memory or retrieved context isn't clearly understood. This paper mechanistically examines the RAG pipeline to highlight that LMs demonstrate a "shortcut'' effect and have a strong bias towards utilizing the retrieved context to answer questions, while relying minimally on model priors. We propose (a) Causal Mediation Analysis; for proving that parametric memory is minimally utilized when answering a question and (b) Attention Contributions and Knockouts for showing the last token residual stream do not get enriched from the subject token in the question, but gets enriched from tokens of RAG-context. We find this pronounced "shortcut'' behaviour to be true across both LLMs (e.g.,LlaMa) and SLMs (e.g., Phi)

From RAGs to rich parameters: Probing how language models utilize external knowledge over parametric information for factual queries

Retrieval Augmented Generation (RAG) enriches the ability of language models to reason using external context to augment responses for a given user prompt. This approach has risen in popularity due to practical applications in various applications of language models in search, question/answering, and chat-bots. However, the exact nature of how this approach works isn't clearly understood. In this paper, we mechanistically examine the RAG pipeline to highlight that language models take shortcut and have a strong bias towards utilizing only the context information to answer the question, while relying minimally on their parametric memory. We probe this mechanistic behavior in language models with: (i) Causal Mediation Analysis to show that the parametric memory is minimally utilized when answering a question and (ii) Attention Contributions and Knockouts to show that the last token residual stream do not get enriched from the subject token in the question, but gets enriched from other informative tokens in the context. We find this pronounced shortcut behaviour true across both LLaMa and Phi family of models.

Comparing Neighbors Together Makes it Easy: Jointly Comparing Multiple Candidates for Efficient and Effective Retrieval

A common retrieve-and-rerank paradigm involves retrieving a broad set of relevant candidates using a scalable bi-encoder, followed by expensive but more accurate cross-encoders to a limited candidate set. However, this small subset often leads to error propagation from the bi-encoders, thereby restricting the performance of the overall pipeline. To address these issues, we propose the Comparing Multiple Candidates (CMC) framework, which compares a query and multiple candidate embeddings jointly through shallow self-attention layers. While providing contextualized representations, CMC is scalable enough to handle multiple comparisons simultaneously, where comparing 2K candidates takes only twice as long as comparing 100. Practitioners can use CMC as a lightweight and effective reranker to improve top-1 accuracy. Moreover, when integrated with another retriever, CMC reranking can function as a virtually enhanced retriever. This configuration adds only negligible latency compared to using a single retriever (virtual), while significantly improving recall at K (enhanced).} Through experiments, we demonstrate that CMC, as a virtually enhanced retriever, significantly improves Recall@k (+6.7, +3.5%-p for R@16, R@64) compared to the initial retrieval stage on the ZeSHEL dataset. Meanwhile, we conduct experiments for direct reranking on entity, passage, and dialogue ranking. The results indicate that CMC is not only faster (11x) than cross-encoders but also often more effective, with improved prediction performance in Wikipedia entity linking (+0.7%-p) and DSTC7 dialogue ranking (+3.3%-p). The code and link to datasets are available at https://github.com/yc-song/cmc

WorldValuesBench: A Large-Scale Benchmark Dataset for Multi-Cultural Value Awareness of Language Models

The awareness of multi-cultural human values is critical to the ability of language models (LMs) to generate safe and personalized responses. However, this awareness of LMs has been insufficiently studied, since the computer science community lacks access to the large-scale real-world data about multi-cultural values. In this paper, we present WorldValuesBench, a globally diverse, large-scale benchmark dataset for the multi-cultural value prediction task, which requires a model to generate a rating response to a value question based on demographic contexts. Our dataset is derived from an influential social science project, World Values Survey (WVS), that has collected answers to hundreds of value questions (e.g., social, economic, ethical) from 94,728 participants worldwide. We have constructed more than 20 million examples of the type "(demographic attributes, value question) $\rightarrow$ answer" from the WVS responses. We perform a case study using our dataset and show that the task is challenging for strong open and closed-source models. On merely $11.1\%$, $25.0\%$, $72.2\%$, and $75.0\%$ of the questions, Alpaca-7B, Vicuna-7B-v1.5, Mixtral-8x7B-Instruct-v0.1, and GPT-3.5 Turbo can respectively achieve $<0.2$ Wasserstein 1-distance from the human normalized answer distributions. WorldValuesBench opens up new research avenues in studying limitations and opportunities in multi-cultural value awareness of LMs.

Multistage Collaborative Knowledge Distillation from Large Language Models

We study semi-supervised sequence prediction tasks where labeled data are too scarce to effectively finetune a model and at the same time few-shot prompting of a large language model (LLM) has suboptimal performance. This happens when a task, such as parsing, is expensive to annotate and also unfamiliar to a pretrained LLM. In this paper, we present a discovery that student models distilled from a prompted LLM can often generalize better than their teacher on such tasks. Leveraging this finding, we propose a new distillation method, multistage collaborative knowledge distillation from an LLM (MCKD), for such tasks. MCKD first prompts an LLM using few-shot in-context learning to produce pseudolabels for unlabeled data. Then, at each stage of distillation, a pair of students are trained on disjoint partitions of the pseudolabeled data. Each student subsequently produces new and improved pseudolabels for the unseen partition to supervise the next round of student(s) with. We show the benefit of multistage cross-partition labeling on two constituency parsing tasks. On CRAFT biomedical parsing, 3-stage MCKD with 50 labeled examples matches the performance of supervised finetuning with 500 examples and outperforms the prompted LLM and vanilla KD by 7.5% and 3.7% parsing F1, respectively.

Editing Commonsense Knowledge in GPT

Memory editing methods for updating encyclopedic knowledge in transformers have received increasing attention for their efficacy, specificity, and generalization advantages. However, it remains unclear if such methods can be adapted for the more nuanced domain of commonsense knowledge. We propose $MEMIT_{CSK}$, an adaptation of MEMIT to edit commonsense mistakes in GPT-2 Large and XL. We extend editing to various token locations and employ a robust layer selection strategy. Models edited by $MEMIT_{CSK}$ outperforms the fine-tuning baselines by 10.97% and 10.73% F1 scores on subsets of PEP3k and 20Q. We further propose a novel evaluation dataset, MEMIT-CSK-PROBE, that contains unaffected neighborhood, affected neighborhood, affected paraphrase, and affected reasoning challenges. $MEMIT_{CSK}$ demonstrates favorable semantic generalization, outperforming fine-tuning baselines by 13.72% and 5.57% overall scores on MEMIT-CSK-PROBE. These results suggest a compelling future direction of incorporating context-specific user feedback concerning commonsense in GPT by direct model editing, rectifying and customizing model behaviors via human-in-the-loop systems.

SAD: Semi-Supervised Anomaly Detection on Dynamic Graphs

Anomaly detection aims to distinguish abnormal instances that deviate significantly from the majority of benign ones. As instances that appear in the real world are naturally connected and can be represented with graphs, graph neural networks become increasingly popular in tackling the anomaly detection problem. Despite the promising results, research on anomaly detection has almost exclusively focused on static graphs while the mining of anomalous patterns from dynamic graphs is rarely studied but has significant application value. In addition, anomaly detection is typically tackled from semi-supervised perspectives due to the lack of sufficient labeled data. However, most proposed methods are limited to merely exploiting labeled data, leaving a large number of unlabeled samples unexplored. In this work, we present semi-supervised anomaly detection (SAD), an end-to-end framework for anomaly detection on dynamic graphs. By a combination of a time-equipped memory bank and a pseudo-label contrastive learning module, SAD is able to fully exploit the potential of large unlabeled samples and uncover underlying anomalies on evolving graph streams. Extensive experiments on four real-world datasets demonstrate that SAD efficiently discovers anomalies from dynamic graphs and outperforms existing advanced methods even when provided with only little labeled data.

GRANDE: a neural model over directed multigraphs with application to anti-money laundering

The application of graph representation learning techniques to the area of financial risk management (FRM) has attracted significant attention recently. However, directly modeling transaction networks using graph neural models remains challenging: Firstly, transaction networks are directed multigraphs by nature, which could not be properly handled with most of the current off-the-shelf graph neural networks (GNN). Secondly, a crucial problem in FRM scenarios like anti-money laundering (AML) is to identify risky transactions and is most naturally cast into an edge classification problem with rich edge-level features, which are not fully exploited by the prevailing GNN design that follows node-centric message passing protocols. In this paper, we present a systematic investigation of design aspects of neural models over directed multigraphs and develop a novel GNN protocol that overcomes the above challenges via efficiently incorporating directional information, as well as proposing an enhancement that targets edge-related tasks using a novel message passing scheme over an extension of edge-to-node dual graph. A concrete GNN architecture called GRANDE is derived using the proposed protocol, with several further improvements and generalizations to temporal dynamic graphs. We apply the GRANDE model to both a real-world anti-money laundering task and public datasets. Experimental evaluations show the superiority of the proposed GRANDE architecture over recent state-of-the-art models on dynamic graph modeling and directed graph modeling.

ConReader: Exploring Implicit Relations in Contracts for Contract Clause Extraction

We study automatic Contract Clause Extraction (CCE) by modeling implicit relations in legal contracts. Existing CCE methods mostly treat contracts as plain text, creating a substantial barrier to understanding contracts of high complexity. In this work, we first comprehensively analyze the complexity issues of contracts and distill out three implicit relations commonly found in contracts, namely, 1) Long-range Context Relation that captures the correlations of distant clauses; 2) Term-Definition Relation that captures the relation between important terms with their corresponding definitions; and 3) Similar Clause Relation that captures the similarities between clauses of the same type. Then we propose a novel framework ConReader to exploit the above three relations for better contract understanding and improving CCE. Experimental results show that ConReader makes the prediction more interpretable and achieves new state-of-the-art on two CCE tasks in both conventional and zero-shot settings.