ISAX heterogeneous processors integrate cores that share a common base ISA, with certain cores offering extension ISAs (e.g., vector extension) to accelerate computation. ISAX balances performance and energy efficiency while facilitating the reuse of existing software ecosystems. RISC-V, which adopts the ISAX architecture, has gained extensive attention in both industry and academia. Binary translation via binary rewriting enables transparent ISAX heterogeneous computing by translating extension instructions when migrating a program to cores without extension support. However, current binary rewriting methods still struggle to achieve both high performance and correctness.

Large Language Models (LLMs) have demonstrated a remarkable understanding of language nuances through instruction tuning, enabling them to effectively tackle various natural language processing tasks.Recent research has focused on the quality of instruction data rather than the quantity of instructions.However, existing high-quality instruction selection methods rely on external models or rules, overlooking the intrinsic association between pretrained model and instruction data, making it difficult to select data that align with the preferences of pre-trained model.To address this challenge, we propose a strategy that utilizes noise injection to identify the quality of instruction data, without relying on external model.We also implement the strategy of combining inter-class diversity and intra-class diversity to improve model performance.The experimental results demonstrate that our method significantly outperforms the model trained on the entire dataset and established baselines.Our study provides a new perspective on noise injection in the field of instruction tuning, and also illustrates that the pre-trained model itself should be considered in defining high-quality.Additionally, we publish our selected highquality instruction data at https://github. com/HUSTNLP-codes/Alpaca-selectd.

Augmenting Language Models (LMs) with structured knowledge graphs (KGs) aims to leverage structured world knowledge to enhance the capability of LMs to complete knowledge-intensive tasks. However, existing methods are unable to effectively utilize the structured knowledge in a KG due to their inability to capture the rich relational semantics of knowledge triplets. Moreover, the modality gap between natural language text and KGs has become a challenging obstacle when aligning and fusing cross-modal information. To address these challenges, we propose a novel knowledge-augmented question answering (QA) model, namely, Graph Reasoning Transformers (GRT). Different from conventional node-level methods, the GRT serves knowledge triplets as atomic knowledge and utilize a triplet-level graph encoder to capture triplet-level graph features. Furthermore, to alleviate the negative effect of the modality gap on joint reasoning, we propose a representation alignment pretraining to align the cross-modal representations and introduce a cross-modal information fusion module with attention bias to enable fine-grained information fusion. Extensive experiments conducted on three knowledge-intensive QA benchmarks show that the GRT outperforms the state-of-the-art KG-augmented QA systems, demonstrating the effectiveness and adaptation of our proposed model.

Shared libraries are widely used in software development to execute third-party functions. However, the size and complexity of shared libraries tend to increase with the need to support more features, resulting in bloated shared libraries. This leads to resource waste and security issues as a significant amount of generic functionality is included unnecessarily in most scenarios, especially in embedded systems. To address this issue, previous works attempt to debloat shared libraries through binary rewriting or recompilation. However, these works face a tradeoff between flexibility in usage (needs recompilation and runtime support) and the effectiveness of debloating (binary rewriting achieves insufficient file size reduction). We propose D-Linker, a tool that debloats shared libraries by reducing both code and data sections in link-time at the object level without recompilation. Our key insight is that object-level shared library debloating is especially suitable for embedded systems because it strikes a balance of flexibility and efficiency. D-Linker identifies the required ELF object files of the shared libraries in an application and relinks them to produce a debloated shared library with better-debloating effectiveness by avoiding the data reference analysis. Our approach achieves over 70% of gadgets reduction as a security benefit and an average size reduction of 49.6% for a stripped libc of coreutils. The results also indicate that D-Linker improves debloating effectiveness by approximately 30% compared to binary-level shared library debloating and incurs a 5% decrease in code gadgets reduction compared to source-code-level shared library debloating.