Models

A community-maintained, regularly updated overview of publicly available Japanese large language models and their benchmarks, listing each model's architecture, training data, publisher and license.

• Models
• Curated lists

A Home Assistant integration and family of small fine-tuned models that let a completely local language model understand voice or chat requests and control a smart home, without cloud services.

• Models

A research method, from Microsoft, for augmenting a language model with an external knowledge base without a separate retrieval step, so the added computation grows linearly rather than quadratically.

• Models

A Core ML and Apple Neural Engine port of the Laya typed-decision model, which answers structured choice, ordinal-score and yes/no questions in a few milliseconds without generating text tokens.

• Models
• Inference

A data repository of per-model metadata, such as release dates and fact-sheet fields, that powers a public gallery of LLM architecture cards.

• Models
• Curated lists

A Python library for training and analyzing sparse autoencoders on language models, used for mechanistic interpretability research.

• Models

A research implementation that adapts Meta's Segment Anything Model to downstream image-segmentation tasks, such as camouflaged-object and polyp detection, using adapters and prompts.

• Models

A TensorFlow toolkit that optimizes Keras and TensorFlow models for deployment through techniques such as quantization and pruning for sparse weights.

• Models

A PyTorch research library for neural network quantization, supporting both post-training quantization and quantization-aware training for common layer types.

• Models

An Intel post-training quantization toolkit that compresses large language and vision-language models to two to four bits with minimal accuracy loss, using sign-gradient descent.

• Models
• Inference

A large-model compression toolkit supporting quantization, speculative decoding and distillation for LLMs, VLMs and diffusion models, aimed at making compressed models more usable and efficient to deploy.

• Models

An educational, end-to-end open-source knowledge base covering LLM fine-tuning, dataset distillation, reinforcement learning and local deployment, with notebooks and guides in four languages.

• Models
• Tutorials

The official research implementation for training large language models to reason using a continuous latent space instead of generating explicit chain-of-thought text.

• Models

A Python library for bulk inference and scalable synthetic-data curation, for preparing post-training datasets and extracting structured data from LLM outputs, with a viewer for monitoring generation.

• Models
• Data engineering

A GPU-accelerated data curation toolkit that builds repeatable pipelines to load, filter, deduplicate and transform large text, image, video and audio datasets for AI training, running the same pipeline on a laptop or a multi-node Ray cluster.

• Models
• Data engineering

CLM-8B is a System One model trained with contrastive learning to score candidate actions against a state, matching a larger reference model's accuracy on computer-use, gaming and tool-calling tasks at up to 9x lower latency.

• Models

The first fully open-source generative-recommendation framework, providing an end-to-end pipeline of semantic ID construction, supervised fine-tuning and reinforcement learning, reproducing OneRec.

• Models

An Apple Silicon port of Karpathy's autoresearch project, running fixed-time autonomous training-experiment loops natively through MLX with no PyTorch or CUDA dependency.

• Agent frameworks
• Models

A tool that turns a folder of documents into a domain-expert training dataset and a fully fine-tuned custom LLM, optionally running entirely offline with no external API key.

• Models

A Core ML model zoo for iOS and macOS that packages PyTorch models converted into ready-to-use .mlpackage files, each with a conversion script and a SwiftUI sample app.

• Models
• Mobile

A bilingual and crosslingual embedding and reranker model pair for retrieval-augmented generation in English and Chinese, developed by NetEase Youdao and used in its own RAG products.

• RAG
• Models

A 9B-parameter multimodal foundation model that pairs a Qwen3.5 language backbone with a vision encoder for general visual understanding, spatial reasoning and agentic tool use.

• Models

A PyTorch implementation of BitNet, a 1-bit Transformer architecture that replaces standard linear layers with a quantized BitLinear layer to scale large language models with lower-precision weights.

• Models

A research method and PyTorch implementation for fine-tuning text-to-image diffusion models on a handful of example images to add a new concept, without retraining the whole model.

• Models
• Generative media