Large language models (LLMs) for code have become indispensable in various domains, including code generation, reasoning tasks and agent systems. While open-access code LLMs are increasingly approaching the performance levels of proprietary models, high-quality code LLMs suitable for rigorous scientific investigation, particularly those with reproducible data processing pipelines and transparent training protocols, remain limited. The scarcity is due to various challenges, including resource constraints, ethical considerations, and the competitive advantages of keeping models advanced. To address the gap, we introduce OpenCoder, a top-tier code LLM that not only achieves performance comparable to leading models but also serves as an "open cookbook" for the research community. Unlike most prior efforts, we release not only model weights and inference code, but also the reproducible training data, complete data processing pipeline, rigorous experimental ablation results, and detailed training protocols for open scientific research. Through this comprehensive release, we identify the key ingredients for building a top-tier code LLM: (1) code optimized heuristic rules for data cleaning and methods for data deduplication, (2) recall of text corpus related to code and (3) high-quality synthetic data in both annealing and supervised fine-tuning stages. By offering this level of openness, we aim to broaden access to all aspects of a top-tier code LLM, with OpenCoder serving as both a powerful model and an open foundation to accelerate research, and enable reproducible advancements in code AI.
Large Language Models (LLMs) specializing in code generation (which are also often referred to as code LLMs), e.g., StarCoder and Code Llama, play increasingly critical roles in various software development scenarios. It is also crucial for code LLMs to possess both code generation and natural language abilities for many specific applications, such as code snippet retrieval using natural language or code explanations. The intricate interaction between acquiring language and coding skills complicates the development of strong code LLMs. Furthermore, there is a lack of thorough prior studies on the LLM pretraining strategy that mixes code and natural language. In this work, we propose a pretraining strategy to enhance the integration of natural language and coding capabilities within a single LLM. Specifically, it includes two phases of training with appropriately adjusted code/language ratios. The resulting model, Crystal, demonstrates remarkable capabilities in both domains. Specifically, it has natural language and coding performance comparable to that of Llama 2 and Code Llama, respectively. Crystal exhibits better data efficiency, using 1.4 trillion tokens compared to the more than 2 trillion tokens used by Llama 2 and Code Llama. We verify our pretraining strategy by analyzing the training process and observe consistent improvements in most benchmarks. We also adopted a typical application adaptation phase with a code-centric data mixture, only to find that it did not lead to enhanced performance or training efficiency, underlining the importance of a carefully designed data recipe. To foster research within the community, we commit to open-sourcing every detail of the pretraining, including our training datasets, code, loggings and 136 checkpoints throughout the training.
Speculative decoding is widely adopted to reduce latency in large language model (LLM) inference by leveraging smaller draft models capable of handling diverse user tasks. However, emerging AI applications, such as LLM-based agents, present unique workload characteristics: instead of diverse independent requests, agentic frameworks typically submit repetitive inference requests, such as multi-agent pipelines performing similar subtasks or self-refinement loops iteratively enhancing outputs. These workloads result in long and highly predictable sequences, which current speculative decoding methods do not effectively exploit. To address this gap, we introduce \emph{SuffixDecoding}, a novel method that utilizes efficient suffix trees to cache long token sequences from prompts and previous outputs. By adaptively speculating more tokens when acceptance likelihood is high and fewer when it is low, SuffixDecoding effectively exploits opportunities for longer speculations while conserving computation when those opportunities are limited. Evaluations on agentic benchmarks, including SWE-Bench and Text-to-SQL, demonstrate that SuffixDecoding achieves speedups of up to 5.3$\times$, outperforming state-of-the-art methods -- 2.8$\times$ faster than model-based approaches like EAGLE-2/3 and 1.9$\times$ faster than model-free approaches such as Token Recycling. SuffixDecoding is open-sourced at https://github.com/snowflakedb/ArcticInference.
Empirical evidence suggests that LLMs exhibit spontaneous cross-lingual alignment. Our findings suggest that although LLMs also demonstrate promising cross-lingual alignment in Information Extraction, there remains significant imbalance across languages, revealing an underlying deficiency in the IE alignment. To address this issue, we propose AlignXIE, a powerful code-based LLM that significantly enhances cross-lingual IE alignment through two strategies. Firstly, AlignXIE formulates IE across different languages, especially non-English ones, as code generation tasks, standardizing the representation of various schemas using Python classes to ensure consistency of the same ontology in different languages and align the schema. Secondly, it incorporates an IE cross-lingual alignment phase through a translated instance prediction task proposed in this paper to align the extraction process, utilizing ParallelNER, an IE bilingual parallel dataset with 257,190 samples, generated by our proposed LLM-based automatic pipeline for IE parallel data construction, with manual annotation to ensure quality. Ultimately, we obtain AlignXIE through multilingual IE instruction tuning. Although without training in 9 unseen languages, AlignXIE surpasses ChatGPT by $30.17\%$ and SoTA by $20.03\%$, thereby demonstrating superior cross-lingual IE capabilities. Comprehensive evaluations on 63 IE benchmarks in Chinese and English under various settings, demonstrate that AlignXIE significantly enhances cross-lingual and multilingual IE through boosting the IE alignment.
Code generation, the automatic creation of source code from natural language descriptions, has garnered significant attention due to its potential to streamline software development. Inspired by research that links task-personality alignment with improved development outcomes, we conduct an empirical study on personality-guided code generation using large language models (LLMs). Specifically, we investigate how emulating personality traits appropriate to the coding tasks affects LLM performance. We extensively evaluate this approach using seven widely adopted LLMs across four representative datasets. Our results show that personality guidance significantly enhances code generation accuracy, with improved pass rates in 23 out of 28 LLM-dataset combinations. Notably, in 11 cases, the improvement exceeds 5%, and in 5 instances, it surpasses 10%, with the highest gain reaching 12.9%. Additionally, personality guidance can be easily integrated with other prompting strategies to further boost performance. We open-source our code and data at https://github.com/IanWalls/Persona-Code.
Large Language Models (LLMs) have been applied to various hardware design tasks, including Verilog code generation, EDA tool scripting, and RTL bug fixing. Despite this extensive exploration, LLMs are yet to be used for the task of post-synthesis metric reasoning and estimation of HDL designs. In this paper, we assess the ability of LLMs to reason about post-synthesis metrics of Verilog designs. We introduce MetRex, a large-scale dataset comprising 25,868 Verilog HDL designs and their corresponding post-synthesis metrics, namely area, delay, and static power. MetRex incorporates a Chain of Thought (CoT) template to enhance LLMs' reasoning about these metrics. Extensive experiments show that Supervised Fine-Tuning (SFT) boosts the LLM's reasoning capabilities on average by 37.0\%, 25.3\%, and 25.7\% on the area, delay, and static power, respectively. While SFT improves performance on our benchmark, it remains far from achieving optimal results, especially on complex problems. Comparing to state-of-the-art regression models, our approach delivers accurate post-synthesis predictions for 17.4\% more designs (within a 5\% error margin), in addition to offering a 1.7x speedup by eliminating the need for pre-processing. This work lays the groundwork for advancing LLM-based Verilog code metric reasoning.
In the era of data-driven decision-making, accurate table-level representations and efficient table recommendation systems are becoming increasingly crucial for improving table management, discovery, and analysis. However, existing approaches to tabular data representation often face limitations, primarily due to their focus on cell-level tasks and the lack of high-quality training data. To address these challenges, we first formulate a clear definition of table similarity in the context of data transformation activities within data-driven enterprises. This definition serves as the foundation for synthetic data generation, which require a well-defined data generation process. Building on this, we propose a novel synthetic data generation pipeline that harnesses the code generation and data manipulation capabilities of Large Language Models (LLMs) to create a large-scale synthetic dataset tailored for table-level representation learning. Through manual validation and performance comparisons on the table recommendation task, we demonstrate that the synthetic data generated by our pipeline aligns with our proposed definition of table similarity and significantly enhances table representations, leading to improved recommendation performance.
The advent of large language models (LLMs) like GitHub Copilot has significantly enhanced programmers' productivity, particularly in code generation. However, these models often struggle with real-world tasks without fine-tuning. As LLMs grow larger and more performant, fine-tuning for specialized tasks becomes increasingly expensive. Parameter-efficient fine-tuning (PEFT) methods, which fine-tune only a subset of model parameters, offer a promising solution by reducing the computational costs of tuning LLMs while maintaining their performance. Existing studies have explored using PEFT and LLMs for various code-related tasks and found that the effectiveness of PEFT techniques is task-dependent. The application of PEFT techniques in unit test generation remains underexplored. The state-of-the-art is limited to using LLMs with full fine-tuning to generate unit tests. This paper investigates both full fine-tuning and various PEFT methods, including LoRA, (IA)^3, and prompt tuning, across different model architectures and sizes. We use well-established benchmark datasets to evaluate their effectiveness in unit test generation. Our findings show that PEFT methods can deliver performance comparable to full fine-tuning for unit test generation, making specialized fine-tuning more accessible and cost-effective. Notably, prompt tuning is the most effective in terms of cost and resource utilization, while LoRA approaches the effectiveness of full fine-tuning in several cases.
Multimodal Large Language Models (MLLMs) have demonstrated remarkable performance on the design-to-code task, i.e., generating UI code from UI mock-ups. However, existing benchmarks only contain static web pages for evaluation and ignore the dynamic interaction, limiting the practicality, usability and user engagement of the generated webpages. To bridge these gaps, we present the first systematic investigation of MLLMs in generating interactive webpages. Specifically, we formulate the Interaction-to-Code task and establish the Interaction2Code benchmark, encompassing 127 unique webpages and 374 distinct interactions across 15 webpage types and 31 interaction categories. Through comprehensive experiments utilizing state-of-the-art (SOTA) MLLMs, evaluated via both automatic metrics and human assessments, we identify four critical limitations of MLLM on Interaction-to-Code task: (1) inadequate generation of interaction compared with full page, (2) prone to ten types of failure, (3) poor performance on visually subtle interactions, and (4) insufficient undestanding on interaction when limited to single-modality visual descriptions. To address these limitations, we propose four enhancement strategies: Interactive Element Highlighting, Failureaware Prompting (FAP), Visual Saliency Enhancement, and Visual-Textual Descriptions Combination, all aiming at improving MLLMs' performance on the Interaction-toCode task. The Interaction2Code benchmark and code are available in https://github. com/WebPAI/Interaction2Code.
Recent advancements in Generative AI offer promising capabilities for spatial analysis. Despite their potential, the integration of generative AI with established GIS platforms remains underexplored. In this study, we propose a framework for integrating LLMs directly into existing GIS platforms, using QGIS as an example. Our approach leverages the reasoning and programming capabilities of LLMs to autonomously generate spatial analysis workflows and code through an informed agent that has comprehensive documentation of key GIS tools and parameters. The implementation of this framework resulted in the development of a "GIS Copilot" that allows GIS users to interact with QGIS using natural language commands for spatial analysis. The GIS Copilot was evaluated with over 100 spatial analysis tasks with three complexity levels: basic tasks that require one GIS tool and typically involve one data layer to perform simple operations; intermediate tasks involving multi-step processes with multiple tools, guided by user instructions; and advanced tasks which involve multi-step processes that require multiple tools but not guided by user instructions, necessitating the agent to independently decide on and executes the necessary steps. The evaluation reveals that the GIS Copilot demonstrates strong potential in automating foundational GIS operations, with a high success rate in tool selection and code generation for basic and intermediate tasks, while challenges remain in achieving full autonomy for more complex tasks. This study contributes to the emerging vision of Autonomous GIS, providing a pathway for non-experts to engage with geospatial analysis with minimal prior expertise. While full autonomy is yet to be achieved, the GIS Copilot demonstrates significant potential for simplifying GIS workflows and enhancing decision-making processes.
Generating executable code from natural language instructions using Large Language Models (LLMs) poses challenges such as semantic ambiguity and understanding taskspecific contexts. To address these issues, we propose a system called DemoCraft, which enhances code generation by leveraging in-context learning and demonstration selection, combined with latent concept learning. Latent concept learning introduces additional concept tokens, which are trainable embeddings that capture task-specific knowledge. We then test our system on two major datasets: MBPP and Humaneval. Our experimental results demonstrate that the proposed system achieves an approximate 2x increase in the pass@k metric compared to baseline models. Furthermore, we introduce two novel evaluation metrics: correctness@k and similarity@k. Our empirical studies indicate that our system attains nearly a 3x improvement in these metrics as well.
Humans have the ability to reason about geometric patterns in images and scenes from a young age. However, developing large multimodal models (LMMs) capable of similar reasoning remains a challenge, highlighting the need for robust evaluation methods to assess these capabilities. We introduce TurtleBench, a benchmark designed to evaluate LMMs' capacity to interpret geometric patterns -- given visual examples, textual instructions, or both -- and generate precise code outputs. Inspired by turtle geometry, a notion used to teach children foundational coding and geometric concepts, TurtleBench features tasks with patterned shapes that have underlying algorithmic logic. Our evaluation reveals that leading LMMs struggle significantly with these tasks, with GPT-4o achieving only 19\% accuracy on the simplest tasks and few-shot prompting only marginally improves their performance ($<2\%$). TurtleBench highlights the gap between human and AI performance in intuitive and visual geometrical understanding, setting the stage for future research in this area. TurtleBench stands as one of the few benchmarks to evaluate the integration of visual understanding and code generation capabilities in LMMs, setting the stage for future research. Code and Dataset for this paper is provided here: https://github.com/sinaris76/TurtleBench
Modern visual effects (VFX) software has made it possible for skilled artists to create imagery of virtually anything. However, the creation process remains laborious, complex, and largely inaccessible to everyday users. In this work, we present AutoVFX, a framework that automatically creates realistic and dynamic VFX videos from a single video and natural language instructions. By carefully integrating neural scene modeling, LLM-based code generation, and physical simulation, AutoVFX is able to provide physically-grounded, photorealistic editing effects that can be controlled directly using natural language instructions. We conduct extensive experiments to validate AutoVFX's efficacy across a diverse spectrum of videos and instructions. Quantitative and qualitative results suggest that AutoVFX outperforms all competing methods by a large margin in generative quality, instruction alignment, editing versatility, and physical plausibility.
Recent advancements in Large Language Models (LLMs) have led to their widespread application in automated code generation. However, these models can still generate defective code that deviates from the specification. Previous research has mainly focused on the mistakes in LLM-generated standalone functions, overlooking real-world software development situations where the successful generation of the code requires software contexts such as external dependencies. In this paper, we considered both of these code generation situations and identified a range of \textit{non-syntactic mistakes} arising from LLMs' misunderstandings of coding question specifications. Seven categories of non-syntactic mistakes were identified through extensive manual analyses, four of which were missed by previous works. To better understand these mistakes, we proposed six reasons behind these mistakes from various perspectives. Moreover, we explored the effectiveness of LLMs in detecting mistakes and their reasons. Our evaluation demonstrated that GPT-4 with the ReAct prompting technique can achieve an F1 score of up to 0.65 when identifying reasons for LLM's mistakes, such as misleading function signatures. We believe that these findings offer valuable insights into enhancing the quality of LLM-generated code.
Online LLM inference powers many exciting applications such as intelligent chatbots and autonomous agents. Modern LLM inference engines widely rely on request batching to improve inference throughput, aiming to make it cost-efficient when running on expensive GPU accelerators. However, the limited GPU memory has largely limited the batch size achieved in practice, leaving significant GPU compute resources wasted. We present NEO, an online LLM inference system that offloads part of attention compute and KV cache states from the GPU to the local host CPU, effectively increasing the GPU batch size and thus inference throughput. To this end, NEO proposes asymmetric GPU-CPU pipelining and load-aware scheduling to balance GPU and CPU loads and fully utilize their compute and memory resources. We evaluate NEO on a wide range of workloads (i.e., code generation, text summarization), GPUs (i.e., T4, A10G, H100), and LLM models (i.e., 7B, 8B, 70B). NEO achieves up to 7.5$\times$, 26%, and 14% higher throughput compared to GPU-only approach on T4, A10G, and H100 GPUs, respectively, while maintaining the same latency; with more powerful CPUs, NEO achieves up to 79.3% throughput gain on A10G GPU.
Recent work has demonstrated that a promising strategy for teaching robots a wide range of complex skills is by training them on a curriculum of progressively more challenging environments. However, developing an effective curriculum of environment distributions currently requires significant expertise, which must be repeated for every new domain. Our key insight is that environments are often naturally represented as code. Thus, we probe whether effective environment curriculum design can be achieved and automated via code generation by large language models (LLM). In this paper, we introduce Eurekaverse, an unsupervised environment design algorithm that uses LLMs to sample progressively more challenging, diverse, and learnable environments for skill training. We validate Eurekaverse's effectiveness in the domain of quadrupedal parkour learning, in which a quadruped robot must traverse through a variety of obstacle courses. The automatic curriculum designed by Eurekaverse enables gradual learning of complex parkour skills in simulation and can successfully transfer to the real-world, outperforming manual training courses designed by humans.
Static verification is a powerful method for enhancing software quality, but it demands significant human labor and resources. This is particularly true of static verifiers that reason about heap manipulating programs using an ownership logic. LLMs have shown promise in a number of software engineering activities, including code generation, test generation, proof generation for theorem provers, and specification generation for static verifiers. However, prior work has not explored how well LLMs can perform specification generation for specifications based in an ownership logic, such as separation logic. To address this gap, this paper explores OpenAI's GPT-4o model's effectiveness in generating specifications on C programs that are verifiable with VeriFast, a separation logic based static verifier. Our experiment employs three different types of user inputs as well as basic and Chain-of-Thought (CoT) prompting to assess GPT's capabilities. Our results indicate that the specifications generated by GPT-4o preserve functional behavior, but struggle to be verifiable. When the specifications are verifiable they contain redundancies. Future directions are discussed to improve the performance.
Large Language Models (LLMs) are increasingly used to control robotic systems such as drones, but their risks of causing physical threats and harm in real-world applications remain unexplored. Our study addresses the critical gap in evaluating LLM physical safety by developing a comprehensive benchmark for drone control. We classify the physical safety risks of drones into four categories: (1) human-targeted threats, (2) object-targeted threats, (3) infrastructure attacks, and (4) regulatory violations. Our evaluation of mainstream LLMs reveals an undesirable trade-off between utility and safety, with models that excel in code generation often performing poorly in crucial safety aspects. Furthermore, while incorporating advanced prompt engineering techniques such as In-Context Learning and Chain-of-Thought can improve safety, these methods still struggle to identify unintentional attacks. In addition, larger models demonstrate better safety capabilities, particularly in refusing dangerous commands. Our findings and benchmark can facilitate the design and evaluation of physical safety for LLMs. The project page is available at huggingface.co/spaces/TrustSafeAI/LLM-physical-safety.
The remarkable reasoning and code generation capabilities of large language models (LLMs) have spurred significant interest in applying LLMs to enable task automation in digital chip design. In particular, recent work has investigated early ideas of applying these models to formal verification (FV), an approach to verifying hardware implementations that can provide strong guarantees of confidence but demands significant amounts of human effort. While the value of LLM-driven automation is evident, our understanding of model performance, however, has been hindered by the lack of holistic evaluation. In response, we present FVEval, the first comprehensive benchmark and evaluation framework for characterizing LLM performance in tasks pertaining to FV. The benchmark consists of three sub-tasks that measure LLM capabilities at different levels: from the generation of SystemVerilog assertions (SVAs) given natural language descriptions to reasoning about the design RTL and suggesting assertions directly without additional human input. As test instances, we present both collections of expert-written verification collateral and methodologies to scalably generate synthetic examples aligned with industrial FV workflows. A wide range of existing LLMs, both proprietary and open-source, are evaluated against FVEval, based on which we investigate where today's LLMs stand and how we might further enable their application toward improving productivity in digital FV. Our benchmark and evaluation code is available at \url{https://github.com/NVlabs/FVEval}.
Instruction tuning is a supervised fine-tuning approach that significantly improves the ability of large language models (LLMs) to follow human instructions. We propose SelfCodeAlign, the first fully transparent and permissive pipeline for self-aligning code LLMs without extensive human annotations or distillation. SelfCodeAlign employs the same base model for inference throughout the data generation process. It first extracts diverse coding concepts from high-quality seed snippets to generate new tasks. It then samples multiple responses per task, pairs each with test cases, and validates them in a sandbox environment. Finally, passing examples are selected for instruction tuning. In our primary experiments, we use SelfCodeAlign with CodeQwen1.5-7B to generate a dataset of 74k instruction-response pairs. Finetuning on this dataset leads to a model that achieves a 67.1 pass@1 on HumanEval+, surpassing CodeLlama-70B-Instruct despite being ten times smaller. Across all benchmarks, this finetuned model consistently outperforms the original version trained with OctoPack, the previous state-of-the-art method for instruction tuning without human annotations or distillation. Additionally, we show that SelfCodeAlign is effective across LLMs of various sizes, from 3B to 33B, and that the base models can benefit more from alignment with their own data distribution. We further validate each component's effectiveness in our pipeline, showing that SelfCodeAlign outperforms both direct distillation from GPT-4o and leading GPT-3.5-based distillation methods, such as OSS-Instruct and Evol-Instruct. SelfCodeAlign has also led to the creation of StarCoder2-Instruct, the first fully transparent, permissively licensed, and self-aligned code LLM that achieves state-of-the-art coding performance.