Large Language Models (LLMs) have demonstrated remarkable capabilities across a variety of software engineering and coding tasks. However, their application in the domain of code and compiler optimization remains underexplored. Training LLMs is resource-intensive, requiring substantial GPU hours and extensive data collection, which can be prohibitive. To address this gap, we introduce LLM Compiler, a suite of robust, openly available, pre-trained models specifically designed for compiler tasks. Built on the foundation of Code Llama, LLM Compiler enhances the understanding of compiler intermediate representations (IRs), assembly language, and optimization techniques. The models have been trained on a vast corpus of 546 billion tokens of LLVM-IR and assembly code and have undergone instruction fine-tuning to interpret compiler behavior. To demonstrate the utility of these research tools, we also present fine-tuned versions of the models with enhanced capabilities in optimizing code size and disassembling from x86_64 and ARM assembly back into LLVM-IR. These achieve 77% of the optimising potential of an autotuning search, and 45% disassembly round trip (14% exact match). LLM Compiler is released under a bespoke commercial license to allow wide reuse and is available in two sizes: 7 billion and 13 billion parameters. Our aim is to provide scalable, cost-effective foundational models for further research and development in compiler optimization by both academic researchers and industry practitioners. Since we released LLM Compiler the community has quantized, repackaged, and downloaded the models over 250k times.

Compression is at the heart of intelligence. A theoretically optimal way to compress any sequence of data is to find the shortest program that outputs that sequence and then halts. However, such Kolmogorov compression is uncomputable, and code generating LLMs struggle to approximate this theoretical ideal, as it requires reasoning, planning and search capabilities beyond those of current models. In this work, we introduce the *KoLMogorov-Test* (KT), a compression-as-intelligence intelligence test for code generation LLMs. In KT a model is presented with a sequence of data at inference time, and asked to generate the shortest program that produces the sequence. We identify several benefits of KT for both evaluation and training: an essentially infinite number of problem instances of varying difficulty is readily available, strong baselines already exist, the evaluation metric (compression) cannot be gamed, and pretraining data contamination is highly unlikely. To evaluate current models, we use audio, text, and DNA data, as well as sequences produced by random synthetic programs. Current flagship models perform poorly - both GPT4-o and Llama-3.1-405B struggle on our natural and synthetic sequences. On our synthetic distribution, we are able to train code generation models with lower compression rates than previous approaches. Moreover, we show that gains on synthetic data generalize poorly to real data, suggesting that new innovations are necessary for additional gains on KT.

Prompting techniques such as chain-of-thought have established themselves as a popular vehicle for improving the outputs of large language models (LLMs). For code generation, however, their exact mechanics and efficacy are under-explored using unified metrics and benchmarks. We thus investigate the effects of a wide range of prompting strategies with a focus on automatic re-prompting over multiple turns and computational requirements. After systematically decomposing reasoning, instruction, and execution feedback prompts, we conduct an extensive grid search on the competitive programming benchmarks CodeContests and TACO for multiple LLM families and sizes (Llama 3.0 and 3.1, 8B, 70B, 405B, and GPT-4o). Our study reveals strategies that consistently improve performance across all models with small and large sampling budgets. We then show how finetuning with such an optimal configuration allows models to internalize the induced reasoning process and obtain improvements in performance and scalability for multi-turn code generation.

Large language models (LLMs) deployed as agents solve user-specified tasks over multiple steps while keeping the required manual engagement to a minimum. Crucially, such LLMs need to ground their generations in any feedback obtained to reliably achieve the desired outcomes. We propose an end-to-end reinforcement learning method for teaching models to leverage execution feedback in the realm of code synthesis, where state-of-the-art LLMs struggle to improve code iteratively compared to independent sampling. We benchmark on competitive programming tasks and achieve large performance gains with both small (8B parameters) and large (70B) models, outperforming previous work while reducing the number of samples required by an order of magnitude. Our analysis of inference-time behavior demonstrates that our method produces LLMs that effectively leverage automatic feedback over multiple steps.

In reinforcement learning, pre-trained low-level skills have the potential to greatly facilitate exploration. However, prior knowledge of the downstream task is required to strike the right balance between generality (fine-grained control) and specificity (faster learning) in skill design. In previous work on continuous control, the sensitivity of methods to this trade-off has not been addressed explicitly, as locomotion provides a suitable prior for navigation tasks, which have been of foremost interest. In this work, we analyze this trade-off for low-level policy pre-training with a new benchmark suite of diverse, sparse-reward tasks for bipedal robots. We alleviate the need for prior knowledge by proposing a hierarchical skill learning framework that acquires skills of varying complexity in an unsupervised manner. For utilization on downstream tasks, we present a three-layered hierarchical learning algorithm to automatically trade off between general and specific skills as required by the respective task. In our experiments, we show that our approach performs this trade-off effectively and achieves better results than current state-of-the-art methods for end-to-end hierarchical reinforcement learning and unsupervised skill discovery.

We formulate the problem of defogging as state estimation and future state prediction from previous, partial observations in the context of real-time strategy games. We propose to employ encoder-decoder neural networks for this task, and introduce proxy tasks and baselines for evaluation to assess their ability of capturing basic game rules and high-level dynamics. By combining convolutional neural networks and recurrent networks, we exploit spatial and sequential correlations and train well-performing models on a large dataset of human games of StarCraft: Brood War. Finally, we demonstrate the relevance of our models to downstream tasks by applying them for enemy unit prediction in a state-of-the-art, rule-based StarCraft bot. We observe improvements in win rates against several strong community bots.

The prevalent approach to neural machine translation relies on bi-directional LSTMs to encode the source sentence.We present a faster and simpler architecture based on a succession of convolutional layers.This allows to encode the source sentence simultaneously compared to recurrent networks for which computation is constrained by temporal dependencies.On WMT'16 English-Romanian translation we achieve competitive accuracy to the state-of-the-art and on WMT'15 English-German we outperform several recently published results.Our models obtain almost the same accuracy as a very deep LSTM setup on WMT'14 English-French translation.We speed up CPU decoding by more than two times at the same or higher accuracy as a strong bidirectional LSTM. 1

The prevalent approach to sequence to sequence learning maps an input sequence to a variable length output sequence via recurrent neural networks. We introduce an architecture based entirely on convolutional neural networks. Compared to recurrent models, computations over all elements can be fully parallelized during training to better exploit the GPU hardware and optimization is easier since the number of non-linearities is fixed and independent of the input length. Our use of gated linear units eases gradient propagation and we equip each decoder layer with a separate attention module. We outperform the accuracy of the deep LSTM setup of Wu et al. (2016) on both WMT’14 English-German and WMT’14 English-French translation at an order of magnitude faster speed, both on GPU and CPU.

In this work, we propose a deep bottleneck feature architecture that is able to leverage data from multiple languages. We also show that tonal features are helpful for non-tonal languages. Evaluations are performed on a low-resource conversational telephone speech transcription task in Bengali, while additional data for DBNF training is provided in Assamese, Pashto, Tagalog, Turkish, and Vietnamese. We obtain relative reductions of up to 17.3% and 9.4% WER over mono-lingual GMMs and DBNFs, respectively.

In this work, we propose several deep neural network architectures that are able to leverage data from multiple languages. Modularity is achieved by training networks for extracting high-level features and for estimating phoneme state posteriors separately, and then combining them for decoding in a hybrid DNN/HMM setup. This approach has been shown to achieve superior performance for single-language systems, and here we demonstrate that feature extractors benefit significantly from being trained as multi-lingual networks with shared hidden representations. We also show that existing mono-lingual networks can be re-used in a modular fashion to achieve a similar level of performance without having to train new networks on multi-lingual data. Furthermore, we investigate in extending these architectures to make use of language-specific acoustic features. Evaluations are performed on a low-resource conversational telephone speech transcription task in Vietnamese, while additional data for acoustic model training is provided in Pashto, Tagalog, Turkish, and Cantonese. Improvements of up to 17.4% and 13.8% over mono-lingual GMMs and DNNs, respectively, are obtained.

Conventional wisdom in automatic speech recognition asserts that pitch information is not helpful in building speech recognizers for non-tonal languages and contributes only modestly to performance in speech recognizers for tonal languages. To maintain consistency between different systems, pitch is therefore often ignored, trading the slight performance benefits for greater system uniformity/ simplicity. In this paper, we report results that challenge this conventional approach. We present new models of tone that deliver consistent performance improvements for tonal languages (Cantonese, Vietnamese) and even modest improvements for non-tonal languages. Using neural networks for feature integration and fusion, these models achieve significant gains throughout, and provide us with system uniformity and standardization across all languages, tonal and non-tonal.

In this work, a novel training scheme for generating bottleneck features from deep neural networks is proposed. A stack of denoising auto-encoders is first trained in a layer-wise, unsupervised manner. Afterwards, the bottleneck layer and an additional layer are added and the whole network is fine-tuned to predict target phoneme states. We perform experiments on a Cantonese conversational telephone speech corpus and find that increasing the number of auto-encoders in the network produces more useful features, but requires pre-training, especially when little training data is available. Using more unlabeled data for pre-training only yields additional gains. Evaluations on larger datasets and on different system setups demonstrate the general applicability of our approach. In terms of word error rate, relative improvements of 9.2% (Cantonese, ML training), 9.3% (Tagalog, BMMI-SAT training), 12% (Tagalog, confusion network combinations with MFCCs), and 8.7% (Switchboard) are achieved.

In this work, we propose a modular combination of two pop-ular applications of neural networks to large-vocabulary con-tinuous speech recognition. First, a deep neural network is trained to extract bottleneck features from frames of mel scale filterbank coefficients. In a similar way as is usually done for GMM/HMM systems, this network is then applied as a non-linear discriminative feature-space transformation for a hybrid setup where acoustic modeling is performed by a deep belief network. This effectively results in a very large network, where the layers of the bottleneck network are fixed and applied to suc-cessive windows of feature frames in a time-delay fashion. We show that bottleneck features improve the recognition perfor-mance of DBN/HMM hybrids, and that the modular combina-tion enables the acoustic model to benefit from a larger tempo-ral context. Our architecture is evaluated on a recently released and challenging Tagalog corpus containing conversational tele-phone speech.