{
  "id": 5430794,
  "title": "Neve - Towards a Unified Programming Model for the Complete Deep Learning Stack",
  "url": "https://urgent.news/2026/09/03/neve-towards-a-unified-programming-model-for-the-complete-deep",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-09-03T22:23:27.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/no_saved_data/neve-towards-a-unified-programming-model-for-the-complete-deep-learning-stack-1g86"
  },
  "original_language": "en",
  "account": "Neve is a high-level programming language designed to provide an expressive syntax similar to Python/PyTorch while also enabling efficient low-level code. The author, No Saved DATA, created Neve after being impressed by the parallelism in the Efficient Zero reinforcement learning model, which PyTorch did not handle efficiently. The initial version of Neve was mostly implemented in C++ but is now being transitioned to be implemented primarily in Neve.\n\nNeve aims to have features that allow it to be extended to other problem domains. Current results show that it performs close to Python/SentencePiece in text processing and BPE training, and is competitive with NumPy and OpenBLAS in CPU matrix multiplication with pure high-level SIMD code. However, an LSTM implemented in Neve was slower than one in PyTorch, primarily due to the lack of kernel fusion and other optimizations.\n\nThe language has been designed to handle both high-level and low-level tasks within the same framework, including efficient preprocessing, automatic differentiation, parallel data workers, and GPU kernels. Currently, the framework is still in development, and the complete implementation is expected to take some more months.\n\nThe source material also covers specific features of Neve, such as the ability to split vectors across multiple threads, a data channel system for parallel data processing, and the use of anonymous functions for mapping tensor operations to their backward operations.",
  "summary": "Hi folks, this is No Saved DATA. I dedicate this post to describe some of the features I put in Neve to make it an expressive high-level language (close to Python/PyTorch syntax), while also allowing efficient low-level code. I am sharing this now, because I believe the language has already strongs traits that allow it to be extended to other problem domains. Current results: Close to…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}