Learning reward functions for physical skills are challenging due to the vast spectrum of skills, the high-dimensionality of state and action space, and nuanced sensory feedback. The complexity of these tasks makes acquiring expert demonstration data both costly and time-consuming. Large Language Models (LLMs) contain valuable task-related knowledge that can aid in learning these reward functions. However, the direct application of LLMs for proposing reward functions has its limitations such as numerical instability and inability to incorporate the environment feedback. We aim to extract task knowledge from LLMs using environment feedback to create efficient reward functions for physical skills. Our approach consists of two components. We first use the LLM to propose features and parameterization of the reward function. Next, we update the parameters of this proposed reward function through an iterative self-alignment process. In particular, this process minimizes the ranking inconsistency between the LLM and our learned reward functions based on the new observations. We validated our method by testing it on three simulated physical skill learning tasks, demonstrating effective support for our design choices.
Sequential tabular data is one of the most commonly used data types in real-world applications. Different from conventional tabular data, where rows in a table are independent, sequential tabular data contains rich contextual and sequential information, where some fields are dynamically changing over time and others are static. Existing transformer-based approaches analyzing sequential tabular data overlook the differences between dynamic and static fields by replicating and filling static fields into each transformer, and ignore temporal information between rows, which leads to three major disadvantages: (1) computational overhead, (2) artificially simplified data for masked language modeling pre-training task that may yield less meaningful representations, and (3) disregarding the temporal behavioral patterns implied by time intervals. In this work, we propose FATA-Trans, a model with two field transformers for modeling sequential tabular data, where each processes static and dynamic field information separately. FATA-Trans is field- and time-aware for sequential tabular data. The field-type embedding in the method enables FATA-Trans to capture differences between static and dynamic fields. The time-aware position embedding exploits both order and time interval information between rows, which helps the model detect underlying temporal behavior in a sequence. Our experiments on three benchmark datasets demonstrate that the learned representations from FATA-Trans consistently outperform state-of-the-art solutions in the downstream tasks. We also present visualization studies to highlight the insights captured by the learned representations, enhancing our understanding of the underlying data. Our codes are available at https://github.com/zdy93/FATA-Trans.
It has been shown that Large Language Model (LLM) alignments can be circumvented by appending specially crafted attack suffixes with harmful queries to elicit harmful responses. To conduct attacks against private target models whose characterization is unknown, public models can be used as proxies to fashion the attack, with successful attacks being transferred from public proxies to private target models. The success rate of attack depends on how closely the proxy model approximates the private model. We hypothesize that for attacks to be transferrable, it is sufficient if the proxy can approximate the target model in the neighborhood of the harmful query. Therefore, in this paper, we propose \emph{Local Fine-Tuning (LoFT)}, \textit{i.e.}, fine-tuning proxy models on similar queries that lie in the lexico-semantic neighborhood of harmful queries to decrease the divergence between the proxy and target models. First, we demonstrate three approaches to prompt private target models to obtain similar queries given harmful queries. Next, we obtain data for local fine-tuning by eliciting responses from target models for the generated similar queries. Then, we optimize attack suffixes to generate attack prompts and evaluate the impact of our local fine-tuning on the attack's success rate. Experiments show that local fine-tuning of proxy models improves attack transferability and increases attack success rate by $39\%$, $7\%$, and $0.5\%$ (absolute) on target models ChatGPT, GPT-4, and Claude respectively.
Large Language Models (LLMs) are frequently used for multi-faceted language generation and evaluation tasks that involve satisfying intricate user constraints or taking into account multiple aspects and criteria. However, their performance can fall short, due to the model's lack of coherence and inability to plan and decompose the problem. We propose Branch-Solve-Merge (BSM), a Large Language Model program (Schlag et al., 2023) for tackling such challenging natural language tasks. It consists of branch, solve, and merge modules that are parameterized with specific prompts to the base LLM. These three modules plan a decomposition of the task into multiple parallel sub-tasks, independently solve them, and fuse the solutions to the sub-tasks. We apply our method to the tasks of LLM response evaluation and constrained text generation and evaluate its effectiveness with multiple LLMs, including Vicuna, LLaMA-2-chat, and GPT-4. BSM improves the evaluation correctness and consistency for each LLM by enhancing human-LLM agreement by up to 26%, reducing length and pairwise position biases by up to 50%, and allowing LLaMA2-chat to match or outperform GPT-4 on most domains. On a constraint story generation task, BSM improves the coherence of stories while also improving constraint satisfaction by 12%.
As language models are applied to an increasing number of real-world applications, understanding their inner workings has become an important issue in model trust, interpretability, and transparency. In this work we show that representation dissimilarity measures, which are functions that measure the extent to which two model's internal representations differ, can be a valuable tool for gaining insight into the mechanics of language models. Among our insights are: (i) an apparent asymmetry in the internal representations of model using SoLU and GeLU activation functions, (ii) evidence that dissimilarity measures can identify and locate generalization properties of models that are invisible via in-distribution test set performance, and (iii) new evaluations of how language model features vary as width and depth are increased. Our results suggest that dissimilarity measures are a promising set of tools for shedding light on the inner workings of language models.
We introduce a novel data generation method for contradiction detection, which leverages the generative power of large language models as well as linguistic rules. Our vision is to provide a condensed corpus of prototypical contradictions, allowing for in-depth linguistic analysis as well as efficient language model fine-tuning. To this end, we instruct the generative models to create contradicting statements with respect to descriptions of specific contradiction types. In addition, the model is also instructed to come up with completely new contradiction typologies. As an auxiliary approach, we use linguistic rules to construct simple contradictions such as those arising from negation, antonymy and numeric mismatch. We find that our methods yield promising results in terms of coherence and variety of the data. Further studies, as well as manual refinement are necessary to make use of this data in a machine learning setup.
The sequence-to-sequence (Seq2Seq) approach has recently been widely used in grammatical error correction (GEC) and shows promising performance. However, the Seq2Seq GEC approach still suffers from two issues. First, a Seq2Seq GEC model can only be trained on parallel data, which, in GEC task, is often noisy and limited in quantity. Second, the decoder of a Seq2Seq GEC model lacks an explicit awareness of the correctness of the token being generated. In this paper, we propose a unified decoding intervention framework that employs an external critic to assess the appropriateness of the token to be generated incrementally, and then dynamically influence the choice of the next token. We discover and investigate two types of critics: a pre-trained left-to-right language model critic and an incremental target-side grammatical error detector critic. Through extensive experiments on English and Chinese datasets, our framework consistently outperforms strong baselines and achieves results competitive with state-of-the-art methods.
Self-supervised pre-trained models such as HuBERT and WavLM leverage unlabeled speech data for representation learning and offer significantly improve for numerous downstream tasks. Despite the success of these methods, their large memory and strong computational requirements hinder their application on resource restricted devices. Therefore, this paper introduces GenDistiller, a novel knowledge distillation framework to distill hidden representations from teacher network based on generative language model. The generative structure enables the proposed model to generate the target teacher hidden layers autoregressively, considering the interactions between hidden layers without instroducing additional inputs. A two-dimensional attention mechanism is implemented to ensure the causality of hidden layers, while preserving bidirectional attention in the time dimension. Experiments reveal the advantage of the generative distiller over the baseline system that predicts the hidden layers of teacher network directly without a generatvie model.
We introduce DavIR, a model-based data selection method for post-training Large Language Models. DavIR generalizes Reducible Holdout Loss to core-set selection problem of causal language modeling, and quantifies the learnability of a given datum with respect to a pre-trained LLM based on relative reduction in loss during fine-tuning, a metric we show to be closely related to the implicit reward model described in Direct Preference Optimization (DPO). We show that 6% of Alpaca dataset selected with DavIR can steer both the LLaMA and Gemma model family to produce superior performance compared to the same models trained on the full 52K dataset. We also show that Alpaca dataset compressed with DavIR can be combined with GSM8K dataset to effectively balance open-domain freeform QA and mathematical reasoning capabilities. Finally, we apply the DavIR objective to DPO and develop a normalized DavIR-DPO objective which improves alignment performance of Zephyr-7B-SFT model by 8% (relative) on AlpacaEval, compared against training on vanilla DPO objective.
In this study, we investigated the ability of the large language model (LLM) to enhance healthcare data interoperability. We leveraged the LLM to convert clinical texts into their corresponding FHIR resources. Our experiments, conducted on 3,671 snippets of clinical text, demonstrated that the LLM not only streamlines the multi-step natural language processing and human calibration processes but also achieves an exceptional accuracy rate of over 90% in exact matches when compared to human annotations.
For a language model (LM) to faithfully model human language, it must compress vast, potentially infinite information into relatively few dimensions. We propose analyzing compression in (pre-trained) LMs from two points of view: geometric and information-theoretic. We demonstrate that the two views are highly correlated, such that the intrinsic geometric dimension of linguistic data predicts their coding length under the LM. We then show that, in turn, high compression of a linguistic dataset predicts rapid adaptation to that dataset, confirming that being able to compress linguistic information is an important part of successful LM performance. As a practical byproduct of our analysis, we evaluate a battery of intrinsic dimension estimators for the first time on linguistic data, showing that only some encapsulate the relationship between information-theoretic compression, geometric compression, and ease-of-adaptation.
With the thriving of pre-trained language model (PLM) widely verified in various of NLP tasks, pioneer efforts attempt to explore the possible cooperation of the general textual information in PLM with the personalized behavioral information in user historical behavior sequences to enhance sequential recommendation (SR). However, despite the commonalities of input format and task goal, there are huge gaps between the behavioral and textual information, which obstruct thoroughly modeling SR as language modeling via PLM. To bridge the gap, we propose a novel Unified pre-trained language model enhanced sequential recommendation (UPSR), aiming to build a unified pre-trained recommendation model for multi-domain recommendation tasks. We formally design five key indicators, namely naturalness, domain consistency, informativeness, noise & ambiguity, and text length, to guide the text-item adaptation and behavior sequence-text sequence adaptation differently for pre-training and fine-tuning stages, which are essential but under-explored by previous works. In experiments, we conduct extensive evaluations on seven datasets with both tuning and zero-shot settings and achieve the overall best performance. Comprehensive model analyses also provide valuable insights for behavior modeling via PLM, shedding light on large pre-trained recommendation models. The source codes will be released in the future.
Typological information has the potential to be beneficial in the development of NLP models, particularly for low-resource languages. Unfortunately, current large-scale typological databases, notably WALS and Grambank, are inconsistent both with each other and with other sources of typological information, such as linguistic grammars. Some of these inconsistencies stem from coding errors or linguistic variation, but many of the disagreements are due to the discrete categorical nature of these databases. We shed light on this issue by systematically exploring disagreements across typological databases and resources, and their uses in NLP, covering the past and present. We next investigate the future of such work, offering an argument that a continuous view of typological features is clearly beneficial, echoing recommendations from linguistics. We propose that such a view of typology has significant potential in the future, including in language modeling in low-resource scenarios.
In the field of natural language processing, open-domain chatbots have emerged as an important research topic. However, a major limitation of existing open-domain chatbot research is its singular focus on short single-session dialogue, neglecting the potential need for understanding contextual information in multiple consecutive sessions that precede an ongoing dialogue. Among the elements that compose the context in multi-session conversation settings, the time intervals between sessions and the relationships between speakers would be particularly important. Despite their importance, current research efforts have not sufficiently addressed these dialogical components. In this paper, we introduce a new 1M multi-session dialogue dataset, called Conversation Chronicles, for implementing a long-term conversation setup in which time intervals and fine-grained speaker relationships are incorporated. Following recent works, we exploit a large language model to produce the data. The extensive human evaluation shows that dialogue episodes in Conversation Chronicles reflect those properties while maintaining coherent and consistent interactions across all the sessions. We also propose a dialogue model, called ReBot, which consists of chronological summarization and dialogue generation modules using only around 630M parameters. When trained on Conversation Chronicles, ReBot demonstrates long-term context understanding with a high human engagement score.
Quantization is a promising approach for reducing memory overhead and accelerating inference, especially in large pre-trained language model (PLM) scenarios. While having no access to original training data due to security and privacy concerns has emerged the demand for zero-shot quantization. Most of the cutting-edge zero-shot quantization methods primarily 1) apply to computer vision tasks, and 2) neglect of overfitting problem in the generative adversarial learning process, leading to sub-optimal performance. Motivated by this, we propose a novel zero-shot sharpness-aware quantization (ZSAQ) framework for the zero-shot quantization of various PLMs. The key algorithm in solving ZSAQ is the SAM-SGA optimization, which aims to improve the quantization accuracy and model generalization via optimizing a minimax problem. We theoretically prove the convergence rate for the minimax optimization problem and this result can be applied to other nonconvex-PL minimax optimization frameworks. Extensive experiments on 11 tasks demonstrate that our method brings consistent and significant performance gains on both discriminative and generative PLMs, i.e., up to +6.98 average score. Furthermore, we empirically validate that our method can effectively improve the model generalization.
Over the past few years, various domain-specific pretrained language models (PLMs) have been proposed and have outperformed general-domain PLMs in specialized areas such as biomedical, scientific, and clinical domains. In addition, financial PLMs have been studied because of the high economic impact of financial data analysis. However, we found that financial PLMs were not pretrained on sufficiently diverse financial data. This lack of diverse training data leads to a subpar generalization performance, resulting in general-purpose PLMs, including BERT, often outperforming financial PLMs on many downstream tasks. To address this issue, we collected a broad range of financial corpus and trained the Financial Language Model (FiLM) on these diverse datasets. Our experimental results confirm that FiLM outperforms not only existing financial PLMs but also general domain PLMs. Furthermore, we provide empirical evidence that this improvement can be achieved even for unseen corpus groups.
Hearing is arguably an essential ability of artificial intelligence (AI) agents in the physical world, which refers to the perception and understanding of general auditory information consisting of at least three types of sounds: speech, audio events, and music. In this paper, we propose SALMONN, a speech audio language music open neural network, built by integrating a pre-trained text-based large language model (LLM) with speech and audio encoders into a single multimodal model. SALMONN enables the LLM to directly process and understand general audio inputs and achieve competitive performances on a number of speech and audio tasks used in training, such as automatic speech recognition and translation, auditory-information-based question answering, emotion recognition, speaker verification, and music and audio captioning etc. SALMONN also has a diverse set of emergent abilities unseen in the training, which includes but is not limited to speech translation to untrained languages, speech-based slot filling, spoken-query-based question answering, audio-based storytelling, and speech audio co-reasoning etc. The presence of cross-modal emergent abilities is studied, and a novel few-shot activation tuning approach is proposed to activate such abilities. To our knowledge, SALMONN is the first model of its type and can be regarded as a step towards AI with generic hearing abilities. The source code, model checkpoints and data are available at https://github.com/bytedance/SALMONN.
Multi-modal open-domain question answering typically requires evidence retrieval from databases across diverse modalities, such as images, tables, passages, etc. Even Large Language Models (LLMs) like GPT-4 fall short in this task. To enable LLMs to tackle the task in a zero-shot manner, we introduce MoqaGPT, a straightforward and flexible framework. Using a divide-and-conquer strategy that bypasses intricate multi-modality ranking, our framework can accommodate new modalities and seamlessly transition to new models for the task. Built upon LLMs, MoqaGPT retrieves and extracts answers from each modality separately, then fuses this multi-modal information using LLMs to produce a final answer. Our methodology boosts performance on the MMCoQA dataset, improving F1 by +37.91 points and EM by +34.07 points over the supervised baseline. On the MultiModalQA dataset, MoqaGPT surpasses the zero-shot baseline, improving F1 by 9.5 points and EM by 10.1 points, and significantly closes the gap with supervised methods. Our codebase is available at https://github.com/lezhang7/MOQAGPT.
Blockchain technology has revolutionized the financial landscape, with cryptocurrencies gaining widespread adoption for their decentralized and transparent nature. As the sentiment expressed on social media platforms can significantly influence cryptocurrency discussions and market movements, sentiment analysis has emerged as a crucial tool for understanding public opinion and predicting market trends. Motivated by the aim to enhance sentiment analysis accuracy in the cryptocurrency domain, this paper investigates fine-tuning techniques on large language models. This paper also investigates the efficacy of supervised fine-tuning and instruction-based fine-tuning on large language models for unseen tasks. Experimental results demonstrate a significant average zero-shot performance gain of 40% after fine-tuning, highlighting the potential of this technique in optimizing pre-trained language model efficiency. Additionally, the impact of instruction tuning on models of varying scales is examined, revealing that larger models benefit from instruction tuning, achieving the highest average accuracy score of 75.16%. In contrast, smaller-scale models may experience reduced generalization due to the complete utilization of model capacity. To gain deeper insight about how instruction works with these language models, this paper presents an experimental investigation into the response of an instruction-based model under different instruction tuning setups. The investigation demonstrates that the model achieves an average accuracy score of 72.38% for short and simple instructions. This performance significantly outperforms its accuracy under long and complex instructions by over 12%, thereby effectively highlighting the profound significance of instruction characteristics in maximizing model performance.
This paper presents a novel approach to enhance the performance of binary code comment quality classification models through the application of Generative Artificial Intelligence (AI). By leveraging the OpenAI API, a dataset comprising 1239 newly generated code-comment pairs, extracted from various GitHub repositories and open-source projects, has been labelled as "Useful" or "Not Useful", and integrated into the existing corpus of 9048 pairs in the C programming language. Employing a cutting-edge Large Language Model Architecture, the generated dataset demonstrates notable improvements in model accuracy. Specifically, when incorporated into the Support Vector Machine (SVM) model, a 6% increase in precision is observed, rising from 0.79 to 0.85. Additionally, the Artificial Neural Network (ANN) model exhibits a 1.5% increase in recall, climbing from 0.731 to 0.746. This paper sheds light on the potential of Generative AI in augmenting code comment quality classification models. The results affirm the effectiveness of this methodology, indicating its applicability in broader contexts within software development and quality assurance domains. The findings underscore the significance of integrating generative techniques to advance the accuracy and efficacy of machine learning models in practical software engineering scenarios.