Eliciting In-context Retrieval and Reasoning for Long-context Large Language Models

Recent advancements in long-context language models (LCLMs) promise to transform Retrieval-Augmented Generation (RAG) by simplifying pipelines. With their expanded context windows, LCLMs can process entire knowledge bases and perform retrieval and reasoning directly -- a capability we define as In-Context Retrieval and Reasoning (ICR^2). However, existing benchmarks like LOFT often overestimate LCLM performance by providing overly simplified contexts. To address this, we introduce ICR^2, a benchmark that evaluates LCLMs in more realistic scenarios by including confounding passages retrieved with strong retrievers. We then propose three methods to enhance LCLM performance: (1) retrieve-then-generate fine-tuning, (2) retrieval-attention-probing, which uses attention heads to filter and de-noise long contexts during decoding, and (3) joint retrieval head training alongside the generation head. Our evaluation of five well-known LCLMs on LOFT and ICR^2 demonstrates significant gains with our best approach applied to Mistral-7B: +17 and +15 points by Exact Match on LOFT, and +13 and +2 points on ICR^2, compared to vanilla RAG and supervised fine-tuning, respectively. It even outperforms GPT-4-Turbo on most tasks despite being a much smaller model.

Training Software Engineering Agents and Verifiers with SWE-Gym

We present SWE-Gym, the first environment for training real-world software engineering (SWE) agents. SWE-Gym contains 2,438 real-world Python task instances, each comprising a codebase with an executable runtime environment, unit tests, and a task specified in natural language. We use SWE-Gym to train language model based SWE agents , achieving up to 19% absolute gains in resolve rate on the popular SWE-Bench Verified and Lite test sets. We also experiment with inference-time scaling through verifiers trained on agent trajectories sampled from SWE-Gym. When combined with our fine-tuned SWE agents, we achieve 32.0% and 26.0% on SWE-Bench Verified and Lite, respectively, reflecting a new state-of-the-art for open-weight SWE agents. To facilitate further research, we publicly release SWE-Gym, models, and agent trajectories.

Learnable Prompting SAM-induced Knowledge Distillation for Semi-supervised Medical Image Segmentation

The limited availability of labeled data has driven advancements in semi-supervised learning for medical image segmentation. Modern large-scale models tailored for general segmentation, such as the Segment Anything Model (SAM), have revealed robust generalization capabilities. However, applying these models directly to medical image segmentation still exposes performance degradation. In this paper, we propose a learnable prompting SAM-induced Knowledge distillation framework (KnowSAM) for semi-supervised medical image segmentation. Firstly, we propose a Multi-view Co-training (MC) strategy that employs two distinct sub-networks to employ a co-teaching paradigm, resulting in more robust outcomes. Secondly, we present a Learnable Prompt Strategy (LPS) to dynamically produce dense prompts and integrate an adapter to fine-tune SAM specifically for medical image segmentation tasks. Moreover, we propose SAM-induced Knowledge Distillation (SKD) to transfer useful knowledge from SAM to two sub-networks, enabling them to learn from SAM's predictions and alleviate the effects of incorrect pseudo-labels during training. Notably, the predictions generated by our subnets are used to produce mask prompts for SAM, facilitating effective inter-module information exchange. Extensive experimental results on various medical segmentation tasks demonstrate that our model outperforms the state-of-the-art semi-supervised segmentation approaches. Crucially, our SAM distillation framework can be seamlessly integrated into other semi-supervised segmentation methods to enhance performance. The code will be released upon acceptance of this manuscript at: https://github.com/taozh2017/KnowSAM

On the Way to LLM Personalization: Learning to Remember User Conversations

Large Language Models (LLMs) have quickly become an invaluable assistant for a variety of tasks. However, their effectiveness is constrained by their ability to tailor responses to human preferences and behaviors via personalization. Prior work in LLM personalization has largely focused on style transfer or incorporating small factoids about the user, as knowledge injection remains an open challenge. In this paper, we explore injecting knowledge of prior conversations into LLMs to enable future work on less redundant, personalized conversations. We identify two real-world constraints: (1) conversations are sequential in time and must be treated as such during training, and (2) per-user personalization is only viable in parameter-efficient settings. To this aim, we propose PLUM, a pipeline performing data augmentation for up-sampling conversations as question-answer pairs, that are then used to finetune a low-rank adaptation adapter with a weighted cross entropy loss. Even in this first exploration of the problem, we perform competitively with baselines such as RAG, attaining an accuracy of 81.5% across 100 conversations.

TypeScore: A Text Fidelity Metric for Text-to-Image Generative Models

Evaluating text-to-image generative models remains a challenge, despite the remarkable progress being made in their overall performances. While existing metrics like CLIPScore work for coarse evaluations, they lack the sensitivity to distinguish finer differences as model performance rapidly improves. In this work, we focus on the text rendering aspect of these models, which provides a lens for evaluating a generative model's fine-grained instruction-following capabilities. To this end, we introduce a new evaluation framework called TypeScore to sensitively assess a model's ability to generate images with high-fidelity embedded text by following precise instructions. We argue that this text generation capability serves as a proxy for general instruction-following ability in image synthesis. TypeScore uses an additional image description model and leverages an ensemble dissimilarity measure between the original and extracted text to evaluate the fidelity of the rendered text. Our proposed metric demonstrates greater resolution than CLIPScore to differentiate popular image generation models across a range of instructions with diverse text styles. Our study also evaluates how well these vision-language models (VLMs) adhere to stylistic instructions, disentangling style evaluation from embedded-text fidelity. Through human evaluation studies, we quantitatively meta-evaluate the effectiveness of the metric. Comprehensive analysis is conducted to explore factors such as text length, captioning models, and current progress towards human parity on this task. The framework provides insights into remaining gaps in instruction-following for image generation with embedded text.

Scaling Diffusion Language Models via Adaptation from Autoregressive Models

Diffusion Language Models (DLMs) have emerged as a promising new paradigm for text generative modeling, potentially addressing limitations of autoregressive (AR) models. However, current DLMs have been studied at a smaller scale compared to their AR counterparts and lack fair comparison on language modeling benchmarks. Additionally, training diffusion models from scratch at scale remains challenging. Given the prevalence of open-source AR language models, we propose adapting these models to build text diffusion models. We demonstrate connections between AR and diffusion modeling objectives and introduce a simple continual pre-training approach for training diffusion models. Through systematic evaluation on language modeling, reasoning, and commonsense benchmarks, we show that we can convert AR models ranging from 127M to 7B parameters (GPT2 and LLaMA) into diffusion models DiffuGPT and DiffuLLaMA, using less than 200B tokens for training. Our experimental results reveal that these models outperform earlier DLMs and are competitive with their AR counterparts. We release a suite of DLMs (with 127M, 355M, and 7B parameters) capable of generating fluent text, performing in-context learning, filling in the middle without prompt re-ordering, and following instructions \url{https://github.com/HKUNLP/DiffuLLaMA}.

DART: Denoising Autoregressive Transformer for Scalable Text-to-Image Generation

Diffusion models have become the dominant approach for visual generation. They are trained by denoising a Markovian process that gradually adds noise to the input. We argue that the Markovian property limits the models ability to fully utilize the generation trajectory, leading to inefficiencies during training and inference. In this paper, we propose DART, a transformer-based model that unifies autoregressive (AR) and diffusion within a non-Markovian framework. DART iteratively denoises image patches spatially and spectrally using an AR model with the same architecture as standard language models. DART does not rely on image quantization, enabling more effective image modeling while maintaining flexibility. Furthermore, DART seamlessly trains with both text and image data in a unified model. Our approach demonstrates competitive performance on class-conditioned and text-to-image generation tasks, offering a scalable, efficient alternative to traditional diffusion models. Through this unified framework, DART sets a new benchmark for scalable, high-quality image synthesis.

TSBP: Improving Object Detection in Histology Images via Test-time Self-guided Bounding-box Propagation

A global threshold (e.g., 0.5) is often applied to determine which bounding boxes should be included in the final results for an object detection task. A higher threshold reduces false positives but may result in missing a significant portion of true positives. A lower threshold can increase detection recall but may also result in more false positives. Because of this, using a preset global threshold (e.g., 0.5) applied to all the bounding box candidates may lead to suboptimal solutions. In this paper, we propose a Test-time Self-guided Bounding-box Propagation (TSBP) method, leveraging Earth Mover's Distance (EMD) to enhance object detection in histology images. TSBP utilizes bounding boxes with high confidence to influence those with low confidence, leveraging visual similarities between them. This propagation mechanism enables bounding boxes to be selected in a controllable, explainable, and robust manner, which surpasses the effectiveness of using simple thresholds and uncertainty calibration methods. Importantly, TSBP does not necessitate additional labeled samples for model training or parameter estimation, unlike calibration methods. We conduct experiments on gland detection and cell detection tasks in histology images. The results show that our proposed TSBP significantly improves detection outcomes when working in conjunction with state-of-the-art deep learning-based detection networks. Compared to other methods such as uncertainty calibration, TSBP yields more robust and accurate object detection predictions while using no additional labeled samples. The code is available at https://github.com/jwhgdeu/TSBP.

On the Limited Generalization Capability of the Implicit Reward Model Induced by Direct Preference Optimization

Reinforcement Learning from Human Feedback (RLHF) is an effective approach for aligning language models to human preferences. Central to RLHF is learning a reward function for scoring human preferences. Two main approaches for learning a reward model are 1) training an EXplicit Reward Model (EXRM) as in RLHF, and 2) using an implicit reward learned from preference data through methods such as Direct Preference Optimization (DPO). Prior work has shown that the implicit reward model of DPO (denoted as DPORM) can approximate an EXRM in the limit. DPORM's effectiveness directly implies the optimality of the learned policy, and also has practical implication for LLM alignment methods including iterative DPO. However, it is unclear how well DPORM empirically matches the performance of EXRM. This work studies the accuracy at distinguishing preferred and rejected answers for both DPORM and EXRM. Our findings indicate that even though DPORM fits the training dataset comparably, it generalizes less effectively than EXRM, especially when the validation datasets contain distribution shifts. Across five out-of-distribution settings, DPORM has a mean drop in accuracy of 3% and a maximum drop of 7%. These findings highlight that DPORM has limited generalization ability and substantiates the integration of an explicit reward model in iterative DPO approaches.

Adaptify: A Refined Adaptation Scheme for Frame Classification in Atrophic Gastritis Videos

Atrophic gastritis is a significant risk factor for developing gastric cancer. The incorporation of machine learning algorithms can efficiently elevate the possibility of accurately detecting atrophic gastritis. Nevertheless, when the trained model is applied in real-life circumstances, its output is often not consistently reliable. In this paper, we propose Adaptify, an adaptation scheme in which the model assimilates knowledge from its own classification decisions. Our proposed approach includes keeping the primary model constant, while simultaneously running and updating the auxiliary model. By integrating the knowledge gleaned by the auxiliary model into the primary model and merging their outputs, we have observed a notable improvement in output stability and consistency compared to relying solely on either the main model or the auxiliary model.