Safetensors Python Packages

fastsafetensors

High-performance safetensors model loader

netron

Visualizer for neural network, deep learning and machine learning models

onnx-safetensors

Use safetensors with ONNX 🤗

aether-vault

Git-like version control for ML — atomic commits for code, weights & datasets with C++ parallel hashing

termux-train

📱 Ultra-lightweight On-Device Tensor & DAG Autograd Deep Learning Framework for Android ARM64 Termux. SafeTensors zero-copy, LoRA fine-tuning, and RoPE Transformers.

hf-mem

A CLI to estimate inference memory requirements for Hugging Face models, written in Python.

hknt

HK is a unified neural model format and framework built in native Zig, designed to replace SafeTensors, GGUF, and PyTorch checkpoints. Features zero-copy mmap loading, dual-mode quantization, 2:4 hardware sparsity, live model expansion, and 137+ architecture support with bindings for Python, Rust, Go, C#, Java, and TypeScript.

axquant

Not another flat 4-bit quantizer. AXQuant mixes precision layer by layer for Apple Silicon—smaller MLX models, protected quality-critical layers, production-ready plans.

needle-rs

600 KB WASM runtime for Cactus Compute's Needle AI tool-calling models, Needle 3, 2 and 1 from one build. Browser, Cloudflare Workers, Node.js, Python, C FFI and no_std. Token-exact with the JAX reference. No backend, no API key.

pygpukit

Minimal GPU runtime for Python - high-performance CUDA kernels, memory management, and LLM inference without heavy dependencies

aivmlib

Aivis Voice Model File (.aivm/.aivmx) Utility Library

bigsmall

Lossless AI model compression - ~34% smaller with bit-identical weights; the autopilot profiles your machine, picks the highest fidelity that runs, and streams models bigger than your RAM.

mlx-model-doctor

Validate an MLX / Hugging Face model repository before you load it.

adaptersentry

Static security scanner for LoRA adapters (.safetensors) — M1 static analyzer for weight-level anomalies.

diffai-python

Python bindings for diffai, an AI/ML model diff tool comparing PyTorch, Safetensors, NumPy and MATLAB tensors

llm-harbor

Stop configuring your AI stack. Start using it. One command brings a complete pre-wired LLM stack with hundreds of services to explore.

pre-reasoning

LLMs take the path of least resistance. Pre-Reasoning hands any model a map of the problem first, so it unlocks the knowledge it already has, faster. For large models, that means hard problems and long-horizon tasks. For small models, consistency. Root blockers, unlock order, conflicts, parallel work. 12M params, runs locally on CPU in ~200ms.

moe-fit

Will this mixture-of-experts model run on my machine, and how fast? Answered from the model's index, before downloading it.

agent-shield-int

Agent Shield catches prompt injection attacks before they reach your LLM.

ptwm

Efficient compression for neural network weights

lmzip

Fast lossless compression for large model weights: 34% off BF16 checkpoints, byte-for-byte identical, with a CUDA decoder at 111 GB/s

knowyourai

Is the model on your machine the model it claims to be? Zero-dependency CLI that verifies open-weight models (GGUF, safetensors, Ollama, Hugging Face) and pins them in a lockfile.

tensorshare

🤝 Trade any tensors over the network

osstables-lance-connector

High-performance Python librarys for connecting AI/ML frameworks with OSS storage.