PrunaAI/pruna

Health v2 · maintenance only

Adoption not part of the score

1275 stars · 102 forks observed · 2026-09-03

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Pruna is an open-source Python model optimization framework that makes AI models faster, smaller, cheaper, and greener via caching, quantization, pruning, distillation, and compilation techniques. It works with popular model stacks like transformers and diffusers and is maintained by Pruna AI, which also offers hosted performance-model endpoints.

Use cases

• speed up inference of my huggingface model
• quantize a diffusion model to reduce memory
• compress an LLM with pruning and distillation
• make my text-to-image model generate faster
• reduce GPU cost for model serving
• benchmark model speed and quality after optimization
• shrink a speech recognition model for deployment

When to choose

• you want a unified API to apply multiple compression techniques (quantization, caching, pruning, distillation, compilation) to transformers or diffusers models
• you need to trade off speed, size, cost, and quality for inference in Python on GPU
• you want to optimize open-source models yourself rather than use a hosted API

When to avoid

• you need training-time optimization or fine-tuning acceleration rather than inference compression
• you want a fully managed inference API without managing models yourself (consider Pruna's hosted endpoints instead)
• your model is not based on a supported stack like transformers or diffusers

Facets

library · maturity active

machine-learning llm-inference image-processing speech-recognition benchmarking machine-learning deep-learning large-language-models computer-vision image-processing performance python cross-platform model-optimization quantization pruning distillation model-compression inference-acceleration diffusion-models transformers huggingface gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="PrunaAI/pruna")