Browsing by Subject "Inference"

Now showing 1 - 3 of 3
  • Thumbnail Image
    Publication
    Open Access
    Characterization of machine learning compilers for LLM inference on NVIDIA GPUs
    (Springer, 2015-05-15) Alejandro Carmona-Martínez,; Bernabé García, Gregorio; José Manuel García; Ingeniería y Tecnología de Computadores; Facultades de la UMU::Facultad de Informática
    AI inference is conflicted between Performance, developer Productivity, and device Portability–the P3 problem. Machine learning compilers (MLCs) aim to address this, but their ecosystem is fragmented, with tools that each prioritize a different issue. This paper evaluates the deployment trade-offs of PyTorch-based LLMs on NVIDIA GPUs using four intertwined prominent MLC tools: torch.compile, TensorRT, XLA, and ONNX Runtime. A dual methodology is used, leveraging synthetic PyTorch models to isolate optimizations and end-to-end benchmarks with State-of-the-Art (SOTA) models (TinyLlama-1.1B, Llama-2-7B) to measure realworld performance. Findings reveal that the peak performance of Ahead-Of-Time (AOT) compilation requires architecture-specific tools such as TensorRT-LLM, which are necessary for SOTA LLMs but are unusable for PyTorch models. As for Just-In-Time (JIT) solutions such as torch.compile and its backends, they are flexible and portable, compatible with all tested models, but they do not consistently accelerate LLMs; therefore, the choice of MLC depends on P3 considerations and model architecture.
  • Thumbnail Image
    Publication
    Open Access
    Una clasificación de las inferencias pragmáticas orientada a la didáctica
    (Asociación Española de Comprensión Lectora, 2015) Ripoll Salcedo, Juan C.
    Nowadays it is accepted that the ability to make inferences is essential to reading comprehension. Pragmatic inferences are classified according to various criteria, none of these taxonomies is commonly accepted and they have been hardly used in education. The aim of this paper is to propose an inference taxonomy with the following characteristics: it must be simple, easy to understand and useful for the classroom. Without a suitable taxonomy there is a risk that teachers do not take into account, when assessing or planning classroom activities, all types of inferences that students should use for a good understanding. The taxonomy was made from the following principle: inferences provide information that does not appear explicitly in the text, so you can ask questions about this information. From this principle it is proposed the existence of five types of inferences that respond to five types of questions. The five main questions are: "what or who is the text alluding to?", "what is the relationship between ... and ...?", "what can be predicted knowing that ...?", "what else can you say about this? "and" what does it all mean? ".According to a review of the literature on inference making the proposed taxonomy is compatible with previous ones, also with those which are focused on teaching. There are several experimental studies showing that children and adolescents perform the proposed five kinds of inferences when they understand texts. Despite these data, the proposed taxonomy is tentative and should be used with caution because it lacks empirical support
  • Thumbnail Image
    Publication
    Open Access
    Review of "Inference and Representation: A Study in Modeling Science"
    (Cambridge University Press; Royal Institute of Philosophy, 2025) Sanches de Oliveira, Guilherme; Filosofía
    Review of Mauricio Suárez (2024) Inference and Representation: A Study in Modeling Science. University of Chicago Press