Models

A neural-network quantization and inference toolkit that measures how sensitive each part of a model is, assigns it just enough precision to fit a memory budget, then serves the packed result with CUDA and Metal kernels.

• Models
• Inference

Research code implementing least-squares binary quantization of neural networks in PyTorch, with configs and scripts for reproducing the paper's MNIST, CIFAR-100 and ImageNet results.

• Models
• Papers

A deployment tool for vision-language-action robot policies that verifies a model exports correctly, proves it matches the reference implementation, then promotes it to hardware profiles such as Jetson, RTX or Apple Silicon.

• Models
• Inference

An MLX implementation of Google's TurboQuant method that compresses both model weights and the key-value cache of large language models on Apple Silicon, letting bigger models fit in less memory.

• Models
• Inference

An open-weight, agent-oriented language model with 29 billion total parameters and 4 billion active per token, trained entirely on Ascend NPU hardware with a native 256K-token context window.

• Models

A command-line toolkit for fine-tuning small language models with LoRA, quantizing them, and running them locally on CPU, designed for low-RAM laptops without a GPU.

• Models
• CLI tools

A tool that predicts whether a given GGUF language model will run on your hardware and how fast, then builds a quantized file customized to your machine and prints the exact llama.cpp command to run it.

• Models
• Inference

A research method for adapting true 1-bit large language models after training, using sparse XOR bitmask patches that flip individual binary weights instead of requiring gradients or continuous weights.

• Models
• Papers

A fast, multithreaded CPU library of SIMD-optimized quantization and dequantization kernels for machine learning, more than twice as fast as PyTorch's built-in routines on tested hardware.

• Models
• Libraries

A Python toolkit for composing pretrained LoRA adapter functions into one deployable model, so capabilities such as RAG, safety and factuality can be combined, swapped or upgraded like software libraries.

• Models

A Python library for running and fine-tuning small, task-specific language models on CPU, generating synthetic training data instead of requiring a labeled dataset.

• Models

Research code for a two-stage reinforcement fine-tuning method that trains large language models to reason step by step about time series data before producing a forecast.

• Models
• Papers

A library of Megatron-Core model definitions for large and multimodal models, letting a developer train models such as Qwen3-Next, GLM-5, DeepSeek-V4 and MiniMax with Megatron-Core's parallelism techniques as easily as with Transformers.

• Models
• Libraries

A distributed micro-language-model system that splits a 56 million parameter model across three ESP32-S3 microcontrollers connected over ESP-NOW, with a browser interface for typing prompts and streaming generated text.

• Models
• Inference

A lightweight command-line and GUI tool, written in Zig, that converts safetensors model files into various GGUF quantizations, targeting image and video diffusion model architectures.

• Models
• CLI tools

A vision-language-action foundation model that turns a vision-language backbone into a real-time robot controller for scientific laboratory tasks, trained with the pi0.5 recipe.

• Models

A task-aware MCP server that calculates the VRAM and compute a fine-tuning job needs and returns the cheapest and fastest matching GPUs across a dozen cloud providers, instead of comparing prices by hand.

• MCP servers
• Models

Research code for a reinforcement-learning method that trains reasoning models with dense token-level supervision by contrasting teacher-student gaps under correct and incorrect reference solutions.

• Models
• Papers

A from-scratch decoder-only language model implementation in pure Rust, covering tokenization, training, quantization, alignment, multimodal input, and an inference server.

• Models

A small, dependency-free neural-network framework written in pure Go, with hand-written AVX2 and NEON SIMD kernels for training and inference on ordinary CPUs.

• Models
• Libraries

The official research codebase for a training method that stabilizes distilling a large teacher agent's behavior into a smaller student agent across long, multi-turn tasks.

• Models
• Papers

A single-file, dependency-free Julia port of Karpathy's microgpt that trains and runs a small GPT model from scratch, written to explore Julia's numerical performance.

• Models
• Tutorials

A vision-language model fine-tuned specifically to extract structured text from German business documents such as invoices, receipts, and forms, runnable locally or through a cloud API.

• RAG
• Models

A local research workflow for capturing a language model's internal activations and directly editing its weights to suppress a chosen behavior, without any training step.

• Models