Daniel Cheng

Haoxin Li, Phillip Keung, Daniel Cheng et al. · allen-ai, uw

Large-scale language model pretraining is a very successful form of self-supervised learning in natural language processing, but it is increasingly expensive to perform as the models and pretraining corpora have become larger over time. We propose NarrowBERT, a modified transformer encoder that increases the throughput for masked language model pretraining by more than $2\times$. NarrowBERT sparsifies the transformer model such that the self-attention queries and feedforward layers only operate on the masked tokens of each sentence during pretraining, rather than all of the tokens as with the usual transformer encoder. We also show that NarrowBERT increases the throughput at inference time by as much as $3.5\times$ with minimal (or no) performance degradation on sentence encoding tasks like MNLI. Finally, we examine the performance of NarrowBERT on the IMDB and Amazon reviews classification and CoNLL NER tasks and show that it is also comparable to standard BERT performance.

Haoxin Li, Daniel Cheng, Phillip Keung et al.

Generative retrieval (Wang et al., 2022; Tay et al., 2022) is a popular approach for end-to-end document retrieval that directly generates document identifiers given an input query. We introduce summarization-based document IDs, in which each document's ID is composed of an extractive summary or abstractive keyphrases generated by a language model, rather than an integer ID sequence or bags of n-grams as proposed in past work. We find that abstractive, content-based IDs (ACID) and an ID based on the first 30 tokens are very effective in direct comparisons with previous approaches to ID creation. We show that using ACID improves top-10 and top-20 recall by 15.6% and 14.4% (relative) respectively versus the cluster-based integer ID baseline on the MSMARCO 100k retrieval task, and 9.8% and 9.9% respectively on the Wikipedia-based NQ 100k retrieval task. Our results demonstrate the effectiveness of human-readable, natural-language IDs created through summarization for generative retrieval. We also observed that extractive IDs outperformed abstractive IDs on Wikipedia articles in NQ but not the snippets in MSMARCO, which suggests that document characteristics affect generative retrieval performance.

Daniel Cheng, Oscar Heath, Daniyaal Farooqi et al.

Academic procrastination is a persistent challenge in computing education, yet evidence on the effectiveness of course-level interventions remains fragmented across diverse designs and contexts. We present a systematic literature review of studies published in the past decade that empirically examine interventions to reduce academic procrastination among post-secondary computing students. Evidence from 19 articles examines interventions that target procrastination through structural, feedback-based, motivational, and self-regulatory mechanisms. Our findings suggest that interventions introducing clear temporal structure consistently promote earlier starts and more distributed work, which act as key mediators of performance gains. The magnitude of these gains depends strongly on task structure, with greater benefits for long-horizon, multi-step assignments than for short, routine tasks. Moreover, supportive designs reliably outperform punitive or restrictive schemes, while uniform interventions yield uneven benefits across students. This review highlights the importance of designing structured, supportive, and personalized interventions to address procrastination in computing education.

Omer Dunay, Daniel Cheng, Adam Tait et al.

CodeCompose is an AI-assisted code authoring tool powered by large language models (LLMs) that provides inline suggestions to 10's of thousands of developers at Meta. In this paper, we present how we scaled the product from displaying single-line suggestions to multi-line suggestions. This evolution required us to overcome several unique challenges in improving the usability of these suggestions for developers. First, we discuss how multi-line suggestions can have a 'jarring' effect, as the LLM's suggestions constantly move around the developer's existing code, which would otherwise result in decreased productivity and satisfaction. Second, multi-line suggestions take significantly longer to generate; hence we present several innovative investments we made to reduce the perceived latency for users. These model-hosting optimizations sped up multi-line suggestion latency by 2.5x. Finally, we conduct experiments on 10's of thousands of engineers to understand how multi-line suggestions impact the user experience and contrast this with single-line suggestions. Our experiments reveal that (i) multi-line suggestions account for 42% of total characters accepted (despite only accounting for 16% for displayed suggestions) (ii) multi-line suggestions almost doubled the percentage of keystrokes saved for users from 9% to 17%. Multi-line CodeCompose has been rolled out to all engineers at Meta, and less than 1% of engineers have opted out of multi-line suggestions.

Jake Hesford, Daniel Cheng, Alan Wan et al.

Our paper provides empirical comparisons between recent IDSs to provide an objective comparison between them to help users choose the most appropriate solution based on their requirements. Our results show that no one solution is the best, but is dependent on external variables such as the types of attacks, complexity, and network environment in the dataset. For example, BoT_IoT and Stratosphere IoT datasets both capture IoT-related attacks, but the deep neural network performed the best when tested using the BoT_IoT dataset while HELAD performed the best when tested using the Stratosphere IoT dataset. So although we found that a deep neural network solution had the highest average F1 scores on tested datasets, it is not always the best-performing one. We further discuss difficulties in using IDS from literature and project repositories, which complicated drawing definitive conclusions regarding IDS selection.

Chandra Maddila, Adam Tait, Claire Chang et al.

Aim: With the advent of LLMs, sophisticated agentic program repair has become viable at large organizations with large codebases. In this work, we develop an Engineering Agent that fixes the source code based on test failures at scale across diverse software offerings internally. Method: Using Llama as the base, we employ the ReAct harness to develop an agent. We start with a test failure that was triaged by a rule-based test failure bot. We then set up an agentic harness and allow the agent to reason and run a set of 15 actions from reading a file to generating a patch. We provide feedback to the agent through static analysis and test failures so it can refine its solution. We leverage an LLM-as-a-Judge to ensure that the patch conforms to the standards followed by a human review to land fixes. Benchmark Findings: We curated offline benchmarks for our patch generator, the Engineering Agent loop, and the LLM-as-a-Judge. In offline evaluations we found that a specialized 70B model is highly competitive with the much larger but vanilla Llama-405B. In an ablation study, we found that the ReAct harness (neural model) benefited from the symbolic information from static analysis tools and test execution traces. A model that strikes a balance between the solve rate and error rate vs the cost and latency has a benchmark solve rate of 42.3% using an average 11.8 feedback iterations. Production Findings: In a three month period, 80% of the generated fixes were reviewed, of which 31.5% were landed (25.5% of the total number of generated fixes). Feedback from Engineers: We used open coding to extract qualitative themes from engineers' feedback. We saw positive feedback in the form of quick approvals, gratitude, and surprise. We also found mixed feedback when the Engineering Agent's solution was partially correct and it served as a good starting point.

Nora Gourmelon, Konrad Heidler, Erik Loebel et al.

Calving front position variation of marine-terminating glaciers is an indicator of ice mass loss and a crucial parameter in numerical glacier models. Deep Learning (DL) systems can automatically extract this position from Synthetic Aperture Radar (SAR) imagery, enabling continuous, weather- and illumination-independent, large-scale monitoring. This study presents the first comparison of DL systems on a common calving front benchmark dataset. A multi-annotator study with ten annotators is performed to contrast the best-performing DL system against human performance. The best DL model's outputs deviate 221 m on average, while the average deviation of the human annotators is 38 m. This significant difference shows that current DL systems do not yet match human performance and that further research is needed to enable fully automated monitoring of glacier calving fronts. The study of Vision Transformers, foundation models, and the inclusion and processing strategy of more information are identified as avenues for future research.

Vijayaraghavan Murali, Chandra Maddila, Imad Ahmad et al.

Generative LLMs have been shown to effectively power AI-based code authoring tools that can suggest entire statements or blocks of code during code authoring. In this paper we present CodeCompose, an AI-assisted code authoring tool developed and deployed at Meta internally. CodeCompose is based on the InCoder LLM that merges generative capabilities with bi-directionality. We have scaled up CodeCompose to serve tens of thousands of developers at Meta, across 9 programming languages and several coding surfaces. We present our experience in making design decisions about the model and system architecture for CodeCompose that addresses these challenges. To release a LLM model at this scale, we needed to first ensure that it is sufficiently accurate. In a random sample of 20K source code files, depending on the language, we are able to reproduce hidden lines between 40% and 58% of the time, an improvement of 1.4x and 4.1x over a model trained only on public data. We gradually rolled CodeCompose out to developers. At the time of this writing, 16K developers have used it with 8% of their code coming directly from CodeCompose. To triangulate our numerical findings, we conduct a thematic analysis on the feedback from 70 developers. We find that 91.5% of the feedback is positive, with the most common themes being discovering APIs, dealing with boilerplate code, and accelerating coding. Meta continues to integrate this feedback into CodeCompose.