The advancement of large language models (LLMs) has significantly propelled the field of code generation. Previous work integrated reinforcement learning (RL) with compiler feedback for exploring the output space of LLMs to enhance code generation quality. However, the lengthy code generated by LLMs in response to complex human requirements makes RL exploration a challenge. Also, since the unit tests may not cover the complicated code, optimizing LLMs by using these unexecuted code snippets is ineffective. To tackle these challenges, we introduce StepCoder, a novel RL framework for code generation, consisting of two main components: CCCS addresses the exploration challenge by breaking the long sequences code generation task into a Curriculum of Code Completion Subtasks, while FGO only optimizes the model by masking the unexecuted code segments to provide Fine-Grained Optimization. In addition, we furthermore construct the APPS+ dataset for RL training, which is manually verified to ensure the correctness of unit tests. Experimental results show that our method improves the ability to explore the output space and outperforms state-of-the-art approaches in corresponding benchmarks. The code and dataset will be made available upon publication.

Large Vision-Language Models (LVLMs) have achieved impressive performance, yet research has pointed out a serious issue with object hallucinations within these models. However, there is no clear conclusion as to which part of the model these hallucinations originate from. In this paper, we present an in-depth investigation into the object hallucination problem specifically within the CLIP model, which serves as the backbone for many state-of-the-art vision-language systems. We unveil that even in isolation, the CLIP model is prone to object hallucinations, suggesting that the hallucination problem is not solely due to the interaction between vision and language modalities. To address this, we propose a counterfactual data augmentation method by creating negative samples with a variety of hallucination issues. We demonstrate that our method can effectively mitigate object hallucinations for CLIP model, and we show the the enhanced model can be employed as a visual encoder, effectively alleviating the object hallucination issue in LVLMs.

We introduce a new dynamic vocabulary for language models. It can involve arbitrary text spans during generation. These text spans act as basic generation bricks, akin to tokens in the traditional static vocabularies. We show that, the ability to generate multi-tokens atomically improve both generation quality and efficiency (compared to the standard language model, the MAUVE metric is increased by 25%, the latency is decreased by 20 %). The dynamic vocabulary can be deployed in a plug-and-play way, thus is attractive for various downstream applications. For example, we demonstrate that dynamic vocabulary can be applied to different domains in a training-free manner. It also helps to generate reliable citations in question answering tasks (substantially enhancing citation results without compromising answer accuracy).

Generalizing to longer sentences is important for recent Transformer-based language models. Besides algorithms manipulating explicit position features, the success of Transformers without position encodings (NoPE) provides a new way to overcome the challenge. In this paper, we study the length generalization property of NoPE. We find that although NoPE can extend to longer sequences than the commonly used explicit position encodings, it still has a limited context length. We identify a connection between the failure of NoPE’s generalization and the distraction of attention distributions. We propose a parameter-efficient tuning for searching attention heads’ best temperature hyper-parameters, which substantially expands NoPE’s context size. Experiments on long sequence language modeling, the synthetic passkey retrieval task and real-world long context tasks show that NoPE can achieve competitive performances with state-of-the-art length generalization algorithms. The source code is publicly accessible

Large language models (LLMs) have achieved impressive performance in numerous domains but often struggle to process lengthy inputs effectively and efficiently due to limited length generalization and attention’s quadratic computational demands. Many sought to mitigate this by restricting the attention window within the pre-trained length. However, these methods introduce new issues such as ignoring the middle context and requiring additional training. To address these problems, we propose LongHeads, a training-free framework that enhances LLM’s long context ability by unlocking multi-head attention’s untapped potential. Instead of allowing each head to attend to the full sentence, which struggles with generalizing to longer sequences due to out-of-distribution (OOD) issues, we allow each head to process in-distribution length by selecting and attending to important context chunks. To this end, we propose a chunk selection strategy that relies on the inherent correlation between the query and the key representations, efficiently distributing context chunks to different heads. In this way, each head ensures it can effectively process attended tokens within the trained length, while different heads in different layers can collectively process longer contexts. LongHeads works efficiently and fits seamlessly with many LLMs that use relative positional encoding. LongHeads achieves 100% accuracy at the 128k length on passkey retrieval task, verifying LongHeads’ efficacy in extending the usable context window for existing models.

We study the problem of defying catastrophic forgetting when learning a series of language processing tasks. Compared with previous methods, we emphasize the importance of not caching history tasks’ data, which makes the problem more challenging. Our proposed method applies the parameter isolation strategy. For each task, it allocates a small portion of private parameters and learns them with a shared pre-trained model. To load correct parameters at testing time, we introduce a simple yet effective non-parametric method. Experiments on continual language learning benchmarks show that our method is significantly better than all existing no-data-cache methods, and is comparable (or even better) than those using historical data.

Recent multilingual models benefit from strong unified semantic representation models. However, due to conflict linguistic regularities, ignoring language-specific features during multilingual learning may suffer from negative transfer. In this work, we analyze the relationbetween a language’s position space and its typological characterization, and suggest deploying different position spaces for different languages. We develop a position generation network which combines prior knowledge from typology features and existing position vectors. Experiments on the multilingual dependency parsing task show that the learned position vectors exhibit meaningful hidden structures, and they can help achieving the best multilingual parsing results.

Event detection is a classic natural language processing task. However, the constantly emerging new events make supervised methods not applicable to unseen types. Previous zero-shot event detection methods either require predefined event types as heuristic rules or resort to external semantic analyzing tools. To overcome this weakness, we propose an end-to-end framework named Zero-Shot Event Detection Based on Ordered Contrastive Learning and Prompt-Based Prediction (ZEOP). By creatively introducing multiple contrastive samples with ordered similarities, the encoder can learn event representations from both instance-level and class-level, which makes the distinctions between different unseen types more significant. Meanwhile, we utilize the prompt-based prediction to identify trigger words without relying on external resources. Experiments demonstrate that our model detects events more effectively and accurately than state-of-the-art methods.

Syntactic trees have been widely applied in relation extraction (RE). However, since parsing qualities are not stable on different text domains and a pre-defined grammar may not well fit the target relation schema, the introduction of syntactic structures sometimes fails to improve RE performances consistently. In this work, we study RE models with various unsupervised structures mined from pre-trained language models (e.g., BERT). We show that, similar to syntactic trees, unsupervised structures are quite informative for RE task: they are able to obtain competitive (even the best) performance scores on benchmark RE datasets (ACE05, WebNLG, SciERC). We also conduct detailed analyses on their abilities of adapting new RE domains and influence of noise links in those structures. The results suggest that unsupervised structures are reasonable alternatives of commonly used syntactic structures in relation extraction models.

Adapting word order from one language to another is a key problem in cross-lingual structured prediction. Current sentence encoders (e.g., RNN, Transformer with position embeddings) are usually word order sensitive. Even with uniform word form representations (MUSE, mBERT), word order discrepancies may hurt the adaptation of models. In this paper, we build structured prediction models with bag-of-words inputs, and introduce a new reordering module to organizing words following the source language order, which learns task-specific reordering strategies from a general-purpose order predictor model. Experiments on zero-shot cross-lingual dependency parsing, POS tagging, and morphological tagging show that our model can significantly improve target language performances, especially for languages that are distant from the source language.

Transition systems usually contain various dynamic structures (e.g., stacks, buffers). An ideal transition-based model should encode these structures completely and efficiently. Previous works relying on templates or neural network structures either only encode partial structure information or suffer from computation efficiency. In this paper, we propose a novel attention-based encoder unifying representation of all structures in a transition system. Specifically, we separate two views of items on structures, namely structure-invariant view and structure-dependent view. With the help of parallel-friendly attention network, we are able to encoding transition states with O(1) additional complexity (with respect to basic feature extractors). Experiments on the PTB and UD show that our proposed method significantly improves the test speed and achieves the best transition-based model, and is comparable to state-of-the-art methods.

We investigate the problem of efficiently incorporating high-order features into neural graph-based dependency parsing. Instead of explicitly extracting high-order features from intermediate parse trees, we develop a more powerful dependency tree node representation which captures high-order information concisely and efficiently. We use graph neural networks (GNNs) to learn the representations and discuss several new configurations of GNN’s updating and aggregation functions. Experiments on PTB show that our parser achieves the best UAS and LAS on PTB (96.0%, 94.3%) among systems without using any external resources.

We describe the graph-based dependency parser in our system (AntNLP) submitted to the CoNLL 2018 UD Shared Task. We use bidirectional lstm to get the word representation, then a bi-affine pointer networks to compute scores of candidate dependency edges and the MST algorithm to get the final dependency tree. From the official testing results, our system gets 70.90 LAS F1 score (rank 9/26), 55.92 MLAS (10/26) and 60.91 BLEX (8/26).

We present a multilingual dependency parser with a bidirectional-LSTM (BiLSTM) feature extractor and a multi-layer perceptron (MLP) classifier. We trained our transition-based projective parser in UD version 2.0 datasets without any additional data. The parser is fast, lightweight and effective on big treebanks. In the CoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies, the official results show that the macro-averaged LAS F1 score of our system Mengest is 61.33%.