PKU-Alignment/align-anything

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

4671 stars · 504 forks observed · 2026-09-25

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Align-Anything is a modular Python framework for aligning any-to-any (all-modality) large models with human intentions and values using feedback-based training. It supports fine-tuning across text, image, video, and audio modalities with algorithms like SFT, DPO, PPO, and rule-based RL.

Use cases

• fine-tune a vision-language model with DPO
• run RLHF training on a multimodal model
• align an audio or video model with human preferences
• train an O1-like reasoning model with SFT
• apply rule-based reinforcement learning like DeepSeek-R1
• evaluate aligned multimodal models
• interact with multimodal models via a CLI

When to choose

• you need to align any-to-any multimodal models with human feedback
• you want a modular framework supporting SFT, DPO, PPO, and RLHF
• you are doing research on multimodal model alignment
• you need fine-tuning support for image, video, and audio models

When to avoid

• you only need inference without training or alignment
• you lack GPU resources for large model training
• you need a simple single-modality text fine-tuning pipeline with minimal setup

Facets

framework · maturity active

llm-training machine-learning deep-learning cli large-language-models machine-learning deep-learning artificial-intelligence python rlhf dpo sft multimodal vision-language-model alignment fine-tuning any-to-any-models gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="PKU-Alignment/align-anything")