OpenGVLab/LLaMA-Adapter

Health v2 · maintenance only

Adoption not part of the score

5914 stars · 378 forks observed · 2026-08-28

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

Official implementation of LLaMA-Adapter and its V2 successor, a parameter-efficient fine-tuning method that adapts LLaMA models to follow instructions using only 1.2M trainable parameters and zero-init attention. V2 extends this to a lightweight multimodal visual instruction model.

Use cases

• fine-tune llama on custom instructions cheaply
• parameter-efficient fine-tuning of large language models
• build a multimodal instruction-following chatbot
• reproduce the llama-adapter paper results
• adapt an open-source LLM without full fine-tuning
• train a visual instruction model on limited GPU budget

When to choose

• you want to instruction-tune LLaMA with minimal trainable parameters and compute
• you need a research-grade reference implementation of the LLaMA-Adapter papers
• you want a lightweight multimodal LLaMA variant

When to avoid

• you need a maintained general toolkit - the team recommends LLaMA2-Accessory instead
• you want production-ready chatbot deployment out of the box
• you work with non-LLaMA model families unsupported by this codebase

Facets

library · maturity maintenance

llm-training machine-learning chatbot sdk large-language-models deep-learning artificial-intelligence python parameter-efficient-fine-tuning lora-alternative instruction-tuning multimodal llama zero-init-attention iclr-2024 research-code natural-language-processing gpu linux docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="OpenGVLab/LLaMA-Adapter")