YHacker News

modded-nanogpt github.com ↗

A competitive speedrun to train a small GPT to a target loss in as few steps as possible.

ToolAI & MLspeedruntraininggpt

Screenshot of github.com, modded-nanogpt's site

Captured

Recommended in 1 thread

Similar tools

  • nanochat ↗AI & ML

    A minimal-code repository for training a small GPT-2-style chat model.

    1 thread
  • CrowdGPT ↗AI & ML

    A distributed volunteer project training a small 1B-parameter LLM from scratch on open data.

    1 thread
  • TabPFN AI & ML

    A transformer model that does tabular prediction via in-context learning.

    1 thread
  • GPT-6.1 Sol ↗AI & ML

    An OpenAI language model offered at several reasoning effort levels.

    1 thread
  • SpeechMap ↗AI & ML

    A site tracking how AI models respond to contested or sensitive prompts.

    1 thread
  • Liquid AI Decision Models ↗AI & ML

    Models from Liquid AI trained specifically to produce decisions with calibrated confidence.

    1 thread
  • MiniCPM ↗AI & ML

    A family of small open-weight language models aimed at edge devices.

    1 thread
  • agent-duel ↗AI & ML

    A system for running local coding agents head to head to rank models.

    0 threads

Also in AI & ML

  • OpenRouter ↗

    A marketplace that routes API requests to many LLM inference providers.

    19 threads
  • DeepSeek ↗

    A family of large language models offered via a low-cost API.

    12 threads
  • llama.cpp ↗

    A C/C++ engine for running LLMs locally.

    12 threads
  • GLM ↗

    An open-weights large language model family from Zhipu AI.

    11 threads
  • Qwen ↗

    Alibaba's family of open-weight large language models.

    9 threads
  • DeepSeek V4.1 Flash ↗

    A low-cost DeepSeek language model available via API.

    8 threads
  • Artificial Analysis ↗

    A site benchmarking and comparing AI models on quality, price and speed.

    7 threads
  • Claude ↗

    Anthropic's AI model and command-line assistant.

    7 threads
  • Qwen 3.8 Flash Next ↗

    A Qwen open-weight model that can be run locally.

    5 threads
  • Strata ↗

    An open-source project for running large MoE models on small-VRAM GPUs with system RAM.

    5 threads
  • OpenCode Go ↗

    A subscription plan for accessing models through the OpenCode coding tool.

    4 threads
  • Pangram ↗

    A service that uses a trained model to detect AI-generated text.

    4 threads

Link from a commenter, not checked or endorsed by us: ↗ https://github.com/kellerjordan/modded-nanogpt