Optimizing the performance of large-scale software repositories demands expertise in code reasoning and software engineering (SWE) to reduce runtime while preserving program correctness. However, most benchmarks emphasize what to fix rather than how to fix code. We introduce SWE-fficiency, a benchmark for evaluating repository-level performance optimization on real workloads. Our suite contains 498 tasks across nine widely used data-science, machine-learning, and HPC repositories (e.g., numpy, pandas, scipy): given a complete codebase and a slow workload, an agent must investigate code semantics, localize bottlenecks and relevant tests, and produce a patch that matches or exceeds expert speedup while passing the same unit tests. To enable this how-to-fix evaluation, our automated pipeline scrapes GitHub pull requests for performance-improving edits, combining keyword filtering, static analysis, coverage tooling, and execution validation to both confirm expert speedup baselines and identify relevant repository unit tests. Empirical evaluation of state-of-the-art agents reveals significant underperformance. On average, agents achieve less than 0.15x the expert speedup: agents struggle in localizing optimization opportunities, reasoning about execution across functions, and maintaining correctness in proposed edits. We release the benchmark and accompanying data pipeline to facilitate research on automated performance engineering and long-horizon software reasoning.

Inference latency stands as a critical bottleneck in the large-scale deployment of Large Language Models (LLMs). Speculative decoding methods have recently shown promise in accelerating inference without compromising the output distribution. However, existing work typically relies on a dualistic draft-verify framework and lacks rigorous theoretical grounding. In this paper, we introduce a novel \emph{polybasic} speculative decoding framework, underpinned by a comprehensive theoretical analysis. Specifically, we prove a fundamental theorem that characterizes the optimal inference time for multi-model speculative decoding systems, shedding light on how to extend beyond the dualistic approach to a more general polybasic paradigm. Through our theoretical investigation of multi-model token generation, we expose and optimize the interplay between model capabilities, acceptance lengths, and overall computational cost. Our framework supports both standalone implementation and integration with existing speculative techniques, leading to accelerated performance in practice. Experimental results across multiple model families demonstrate that our approach yields speedup ratios ranging from $3.31\times$ to $4.01\times$ for LLaMA2-Chat 7B, up to $3.87 \times$ for LLaMA3-8B, up to $4.43 \times$ for Vicuna-7B and up to $3.85 \times$ for Qwen2-7B -- all while preserving the original output distribution. We release our theoretical proofs and implementation code to facilitate further investigation into polybasic speculative decoding.

Large Language Models (LLMs) face significant inference latency challenges stemming from their autoregressive design and large size. To address this, speculative decoding emerges as a solution, enabling the simultaneous generation and validation of multiple tokens. While recent approaches like EAGLE-2 and EAGLE-3 improve speculative decoding using dynamic tree structures, they often neglect the impact of crucial system variables such as GPU devices and batch sizes. Therefore, we introduce a new dynamic tree decoding approach called CAST that takes into account inference costs, including factors such as GPU configurations and batch sizes, to dynamically refine the tree structure. Through comprehensive experimentation across six diverse tasks and utilizing six distinct LLMs, our methodology demonstrates remarkable results, achieving speeds up to 5.2 times faster than conventional decoding methods. Moreover, it generally outperforms existing state-of-the-art techniques from 5% to 20%.

Speculative decoding is a widely adopted technique for accelerating inference in large language models (LLMs), yet its application to vision-language models (VLMs) remains underexplored, with existing methods achieving only modest speedups (<1.5x). This gap is increasingly significant as multimodal capabilities become central to large-scale models. We hypothesize that large VLMs can effectively filter redundant image information layer by layer without compromising textual comprehension, whereas smaller draft models struggle to do so. To address this, we introduce Vision-Aware Speculative Decoding (ViSpec), a novel framework tailored for VLMs. ViSpec employs a lightweight vision adaptor module to compress image tokens into a compact representation, which is seamlessly integrated into the draft model's attention mechanism while preserving original image positional information. Additionally, we extract a global feature vector for each input image and augment all subsequent text tokens with this feature to enhance multimodal coherence. To overcome the scarcity of multimodal datasets with long assistant responses, we curate a specialized training dataset by repurposing existing datasets and generating extended outputs using the target VLM with modified prompts. Our training strategy mitigates the risk of the draft model exploiting direct access to the target model's hidden states, which could otherwise lead to shortcut learning when training solely on target model outputs. Extensive experiments validate ViSpec, achieving, to our knowledge, the first substantial speedup in VLM speculative decoding. Code is available at https://github.com/KangJialiang/ViSpec.

To bridge the representation gap between the sub-network and the full model, we train a lightweight and efficient adapter module on top of the sub-network.

Speculative decoding accelerates large language model (LLM) inference by letting a lightweight draft model propose multiple tokens that the target model verifies in parallel. Y et existing training objectives optimize only a single greedy draft path, while decoding follows a tree policy that re-ranks and verifies multiple branches. This draft policy misalignment limits achievable speedups. We introduce Group Tree Optimization (GTO), which aligns training with the decoding-time tree policy through two components: (i) Draft Tree Reward, a sampling-free objective equal to the expected acceptance length of the draft tree under the target model, directly measuring decoding performance; (ii) Group-based Draft Policy Training, a stable optimization scheme that contrasts trees from the current and a frozen reference draft model, forming debiased group-standardized advantages and applying a PPO-style surrogate along the longest accepted sequence for robust updates. We further prove that increasing our Draft Tree Reward provably improves acceptance length and speedup. By bridging draft policy misalignment, GTO offers a practical, general solution for efficient LLM inference. Large language models (LLMs) like GPTs (Achiam et al., 2023) and LLaMAs (Touvron et al., 2023a;b; Dubey et al., 2024) have achieved remarkable success in dialogue (Zheng et al., 2023), coding (Chen et al., 2021), and reasoning (Cobbe et al., 2021).

Large language models (LLMs) deliver impressive generation quality, but incur very high inference cost because each output token is generated auto-regressively through all model layers. Early-exit based self-speculative decoding (EESD) has emerged to mitigate this cost. However, in practice, many approaches struggle to achieve the expected acceleration in such draft-then-verify paradigm even with a well-aligned early-exit head and selected exit position. Our analysis reveals that EESD only pays off when the vast majority of draft tokens are accepted by the LLM. Otherwise, the draft cost may overcome the acceleration gain and lead to a negative speedup. To mitigate this, we propose Pipeline-Parallel Self-Speculative Decoding (PPSD) that fully pipelines the draft and verification work so that no effort is wasted on failed predictions. It has two key innovations. We configure the model layers as a pipeline in which early-exit (draft) computations and remaining-layer (verification) computations overlap. We interleave drafting and verification per token. While the LLM is verifying the current token in its final layers, the early-exit path simultaneously drafts the next token. Such a verify-while-draft scheme keeps all units busy and validates tokens on-the-fly analogous to pipelining the speculation and verification stages. Empirical results confirm that PPSD achieves state-of-the-art acceleration in self-speculative LLM inference. On diverse benchmarks, PPSD achieves speedup ratios in the range of 2.01x~3.81x, which gains almost the optimal acceleration at the fixed acceptance rate and exit position, showcasing its advancement in providing efficient self-speculation.

The synergistic mechanism based on Speculative Decoding (SD) has garnered considerable attention as a simple yet effective approach for accelerating the inference of large language models (LLMs). Nonetheless, the high rejection rates require repeated LLMs calls to validate draft tokens, undermining the overall efficiency gain of SD. In this work, we revisit existing verification mechanisms and propose a novel synergetic mechanism Consultant Decoding (CD). Unlike SD, which relies on a metric derived from importance sampling for verification, CD verifies candidate drafts using token-level likelihoods computed solely by the LLM. CD achieves up to a 2.5-fold increase in inference speed compared to the target model, while maintaining comparable generation quality (around 100% of the target model's performance). Interestingly, this is achieved by combining models whose parameter sizes differ by two orders of magnitude. In addition, CD reduces the call frequency of the large target model to below 10%, particularly in more demanding tasks. CD's performance was even found to surpass that of the large target model, which theoretically represents the upper bound for speculative decoding.

Speech large language models (LLMs) have emerged as a prominent research focus in speech processing. We introduce VocalNet-1B and VocalNet-8B, a series of high-performance, low-latency speech LLMs enabled by a scalable and model-agnostic training framework designed for real-time voice interaction. Central to our contribution is the first application of multi-token prediction (MTP) to speech LLMs. This approach represents a paradigm shift from standard next-token prediction (NTP), offering simultaneous improvements in generation speed and quality. Informed by analysis of MTP's effect on speech generation and experimental comparisons, we designed a straightforward and highly effective MTP implementation. Experiments demonstrate that VocalNet performs on par with mainstream Omni LLMs even with limited training data, and significantly surpasses existing open-source speech LLMs. To foster reproducibility and community advancement, all model weights, inference code, training data, and framework implementations have been made publicly available at https://github.com/SJTU-OmniAgent/VocalNet