Tag

Gemma

JAX-native library for loading, sampling, and fine-tuning DeepMind's open-weight Gemma LLM family, with multi-modal sampling and checkpoint utilities — aimed at researchers and engineers running experiments on CPU/GPU/TPU.

Unstract

Extracts structured JSON from unstructured documents — PDFs, scans, images — by defining extraction schemas as natural-language prompts, then shipping them as REST APIs or ETL pipelines. Swappable across LLM providers and vector DBs.

Brush

Produces real-time 3D reconstructions from multi-view images using Gaussian splatting, with on-device training and interactive viewing across native desktops, Android, and the browser. Uses WebGPU and the Burn ML framework to ship dependency-free binaries, a CLI, live training visualization, and streaming .ply support.

bigcode/the-stack-v2

Provides a multilingual, deduplicated corpus of public source code in Parquet for large-scale model training and evaluation. Includes license metadata, language splits, and streaming-friendly packaging for use with Hugging Face Datasets — suited to training code-focused foundation models but requires careful license/provenance review.

supermemory

Memory engine that lets AI apps remember users across conversations: it extracts facts, tracks updates, resolves contradictions, and auto-forgets stale info, returning context in ~50ms. Tops the LongMemEval, LoCoMo and ConvoMem memory benchmarks.

TheRock

Provides a lightweight build platform for HIP and ROCm that supports building ROCm, PyTorch, and JAX from source, multi-architecture nightly releases, and integrated CI/CD and developer tooling for Linux and Windows.

Generates HD short videos from a single topic/keyword — auto-creates script, finds/assembles footage, generates subtitles, TTS and background music. Offers web UI + API, batch mode, multiple LLM/TTS providers and common short-video aspect ratios.

Wren AI

Converts natural-language questions into SQL, charts, and dashboards across 20+ data sources like BigQuery and Snowflake. Runs through AI clients like Claude Code, grounding queries in a Git-versioned semantic layer rather than raw schemas.

Open-weights 314B-parameter Mixture-of-Experts language model (8 experts, 2 active per token, 8,192-token context) released under Apache 2.0. Ships a raw JAX checkpoint plus reference inference code; needs heavy multi-GPU memory to load.

Generates short videos from text, images, or videos and ships a full training/inference pipeline with checkpoints and demos. Key features include multi-stage training (VAE / 3D-VAE), rectified-flow training, video compression modules, and support for 2s–16s clips at up to 720p. Best for researchers and engineers who can provide substantial GPU resources.

OpenHands: AI-Driven Development

Runs coding agents and automations from a self-hosted developer control center, with local, remote, cloud, and ACP-compatible backends for managed engineering workflows.

Marin

End-to-end framework for running and reproducing foundation-model research workflows — from data curation and tokenization to training and evaluation. Emphasizes reproducibility by recording every step (including failed runs) and expressing experiments as dependency-ordered steps.