Draft Model Knows When to Stop: A Self-Verification Length Policy for Speculative Decoding

Speculative Decoding (SD) has become an important technique in accelerating the inference speed of large language models. Conventional SD methods employ a fixed draft length, which ignores the token generation difficulty across tasks. Consequently, in this paper, we address such an issue and introduce SVIP - a difficulty-aware dynamic draft length policy for speculative decoding systems. Based on a theoretical lower bound of draft token acceptance rate and its inference-time approximation, SVIP adaptively determines the lengths of draft sequences based on the entropy of each draft token distribution. Experimental results on mainstream SD benchmarks and frameworks demonstrate the superior performance of SVIP, achieving up to 20\% walltime speedup on SpecBench over baseline SD methods and 60\% speedup on MT-Bench for long-form generation of up to 8K tokens. Moreover, SVIP is totally training-free and compatible with any existing SD methods that generate draft tokens autoregressively. Experimental results also show that SVIP yields consistent walltime improvement on top of GliDe & CaPE and EAGLE-2.

GALLa: Graph Aligned Large Language Models for Improved Source Code Understanding

Programming languages possess rich semantic information such as data flow that is represented by graphs and not available from the surface form of source code. Recent code language models have scaled to billions of parameters, but model source code solely as text tokens while ignoring any other structural information. Conversely, models that do encode structural information of code make modifications to the Transformer architecture, limiting their scale and compatibility with pretrained LLMs. In this work, we take the best of both worlds with GALLa - Graph Aligned Large Language Model. GALLa utilizes graph neural networks and cross-modal alignment technologies to inject the structural information of code into LLMs as an auxiliary task during finetuning. This framework is both model-agnostic and task-agnostic, as it can be applied to any code LLM for any code downstream task, and requires the structural graph data only at training time from a corpus unrelated to the finetuning data, while incurring no cost at inference time over the baseline LLM. Experiments on five code tasks with four different baseline LLMs ranging in size from 350M to 8B validate the effectiveness of GALLa, demonstrating consistent improvement over the baseline, even for powerful models such as LLaMA3.

Multiple-Choice Questions are Efficient and Robust LLM Evaluators

We present GSM-MC and MATH-MC, two multiple-choice (MC) datasets constructed by collecting answers and incorrect predictions on GSM8K and MATH from over 50 open-source models. Through extensive experiments, we show that LLMs' performance on the MC versions of these two popular benchmarks is strongly correlated with their performance on the original versions, and is quite robust to distractor choices and option orders, while the evaluation time is reduced by a factor of up to 30. Following a similar procedure, we also introduce PythonIO, a new program output prediction MC dataset constructed from two other popular LLM evaluation benchmarks HumanEval and MBPP. Our data and code are available at https://github.com/Geralt-Targaryen/MC-Evaluation.

Is Cognition and Action Consistent or Not: Investigating Large Language Model's Personality

In this study, we investigate the reliability of Large Language Models (LLMs) in professing human-like personality traits through responses to personality questionnaires. Our goal is to evaluate the consistency between LLMs' professed personality inclinations and their actual "behavior", examining the extent to which these models can emulate human-like personality patterns. Through a comprehensive analysis of LLM outputs against established human benchmarks, we seek to understand the cognition-action divergence in LLMs and propose hypotheses for the observed results based on psychological theories and metrics.

Can ChatGPT Rival Neural Machine Translation? A Comparative Study

Inspired by the increasing interest in leveraging large language models for translation, this paper evaluates the capabilities of large language models (LLMs) represented by ChatGPT in comparison to the mainstream neural machine translation (NMT) engines in translating Chinese diplomatic texts into English. Specifically, we examine the translation quality of ChatGPT and NMT engines as measured by four automated metrics and human evaluation based on an error-typology and six analytic rubrics. Our findings show that automated metrics yield similar results for ChatGPT under different prompts and NMT systems, while human annotators tend to assign noticeably higher scores to ChatGPT when it is provided an example or contextual information about the translation task. Pairwise correlation between automated metrics and dimensions of human evaluation produces weak and non-significant results, suggesting the divergence between the two methods of translation quality assessment. These findings provide valuable insights into the potential of ChatGPT as a capable machine translator, and the influence of prompt engineering on its performance.

Distinguishing Translations by Human, NMT, and ChatGPT: A Linguistic and Statistical Approach

The growing popularity of neural machine translation (NMT) and LLMs represented by ChatGPT underscores the need for a deeper understanding of their distinct characteristics and relationships. Such understanding is crucial for language professionals and researchers to make informed decisions and tactful use of these cutting-edge translation technology, but remains underexplored. This study aims to fill this gap by investigating three key questions: (1) the distinguishability of ChatGPT-generated translations from NMT and human translation (HT), (2) the linguistic characteristics of each translation type, and (3) the degree of resemblance between ChatGPT-produced translations and HT or NMT. To achieve these objectives, we employ statistical testing, machine learning algorithms, and multidimensional analysis (MDA) to analyze Spokesperson's Remarks and their translations. After extracting a wide range of linguistic features, supervised classifiers demonstrate high accuracy in distinguishing the three translation types, whereas unsupervised clustering techniques do not yield satisfactory results. Another major finding is that ChatGPT-produced translations exhibit greater similarity with NMT than HT in most MDA dimensions, which is further corroborated by distance computing and visualization. These novel insights shed light on the interrelationships among the three translation types and have implications for the future advancements of NMT and generative AI.

A Survey on Language Models for Code

In this work we systematically review the recent advancements in code processing with language models, covering 50+ models, 30+ evaluation tasks, 150+ datasets, and 550 related works. We break down code processing models into general language models represented by the GPT family and specialized models that are specifically pretrained on code, often with tailored objectives. We discuss the relations and differences between these models, and highlight the historical transition of code modeling from statistical models and RNNs to pretrained Transformers and LLMs, which is exactly the same course that had been taken by NLP. We also discuss code-specific features such as AST, CFG, and unit tests, along with their application in training code language models, and identify key challenges and potential future directions in this domain. We keep the survey open and updated on GitHub repository at https://github.com/codefuse-ai/Awesome-Code-LLM.

MELA: Multilingual Evaluation of Linguistic Acceptability

Recent benchmarks for Large Language Models (LLMs) have mostly focused on application-driven tasks such as complex reasoning and code generation, and this has led to a scarcity in purely linguistic evaluation of LLMs. Against this background, we introduce Multilingual Evaluation of Linguistic Acceptability -- MELA, the first multilingual benchmark on linguistic acceptability with 48K samples covering 10 languages from a diverse set of language families. We establish baselines of commonly used LLMs along with supervised models, and conduct cross-lingual transfer and multi-task learning experiments with XLM-R. In pursuit of multilingual interpretability, we analyze the weights of fine-tuned XLM-R to explore the possibility of identifying transfer difficulty between languages. Our results show that ChatGPT benefits much from in-context examples but still lags behind fine-tuned XLM-R, while the performance of GPT-4 is on par with fine-tuned XLM-R even in zero-shot setting. Cross-lingual and multi-task learning experiments show that unlike semantic tasks, in-language training data is crucial in acceptability judgements. Results in layerwise probing indicate that the upper layers of XLM-R become a task-specific but language-agnostic region for multilingual acceptability judgment. We also introduce the concept of conflicting weight, which could be a potential indicator for the difficulty of cross-lingual transfer between languages. Our data will be available at https://github.com/sjtu-compling/MELA.

Self-distillation Regularized Connectionist Temporal Classification Loss for Text Recognition: A Simple Yet Effective Approach

Text recognition methods are gaining rapid development. Some advanced techniques, e.g., powerful modules, language models, and un- and semi-supervised learning schemes, consecutively push the performance on public benchmarks forward. However, the problem of how to better optimize a text recognition model from the perspective of loss functions is largely overlooked. CTC-based methods, widely used in practice due to their good balance between performance and inference speed, still grapple with accuracy degradation. This is because CTC loss emphasizes the optimization of the entire sequence target while neglecting to learn individual characters. We propose a self-distillation scheme for CTC-based model to address this issue. It incorporates a framewise regularization term in CTC loss to emphasize individual supervision, and leverages the maximizing-a-posteriori of latent alignment to solve the inconsistency problem that arises in distillation between CTC-based models. We refer to the regularized CTC loss as Distillation Connectionist Temporal Classification (DCTC) loss. DCTC loss is module-free, requiring no extra parameters, longer inference lag, or additional training data or phases. Extensive experiments on public benchmarks demonstrate that DCTC can boost text recognition model accuracy by up to 2.6%, without any of these drawbacks.

Revisiting Acceptability Judgements

Years have passed since the NLP community has last focused on linguistic acceptability. In this work, we revisit this topic in the context of large language models. We introduce CoLAC - Corpus of Linguistic Acceptability in Chinese, the first large-scale non-English acceptability dataset that is verified by native speakers and comes with two sets of labels. Our experiments show that even the largest InstructGPT model performs only at chance level on CoLAC, while ChatGPT's performance (48.30 MCC) is also way below supervised models (59.03 MCC) and human (65.11 MCC). Through cross-lingual transfer experiments and fine-grained linguistic analysis, we demonstrate for the first time that knowledge of linguistic acceptability can be transferred across typologically distinct languages, as well as be traced back to pre-training.