# THUDM/AgentTuning

AgentTuning: Enabling Generalized Agent Abilities for LLMs

Repository: https://github.com/THUDM/AgentTuning
Canonical: https://ross.abutalabs.com/products/agenttuning
Homepage: https://thudm.github.io/AgentTuning/
Language: Python
License Family: other
Last push: 2023-10-31T15:34:43+00:00

## Health v2 (maintenance only)
Score: 27/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 75
- inputs: {"age_days": 1050, "days_push": 1037, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1504, forks 104 (observed 2026-08-28T04:04:54.709851+00:00)

## What it is
AgentTuning is a research project from Tsinghua University that instruction-tunes LLMs on multi-task agent interaction trajectories to improve their agent abilities. It releases the AgentInstruct dataset (1,866 filtered GPT-4 trajectories across 6 tasks) and the AgentLM-7B/13B/70B models fine-tuned from Llama 2.

## Use cases
- fine-tune an open-source LLM to act as an agent
- get training data of agent interaction trajectories
- improve tool-use and planning abilities of Llama 2 models
- find an open alternative to GPT-3.5 for agent tasks
- download AgentInstruct dataset for agent instruction tuning
- evaluate open LLMs on held-in and held-out agent benchmarks

## When to choose
- you need an open-weight LLM with stronger agent/tool-use capabilities
- you want a curated instruction-tuning dataset of agent trajectories
- you are researching how instruction tuning affects agent generalization

## When to avoid
- you need a production-ready agent framework rather than models and data
- you cannot host or fine-tune large models on GPUs
- you need a permissively licensed artifact - the repo has no license

## Facets
- artifact type: dataset
- maturity: maintenance
- function: llm-training, agent-framework, machine-learning, data-generation
- domain: large-language-models, machine-learning
- platform: python
- tags: instruction-tuning, agent-trajectories, llama2, huggingface, research, agentlm, agentinstruct, fine-tuning, ai-agents, natural-language-processing, gpu, linux

## Member repositories
- THUDM/AgentTuning (main) score 27

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:54.709851+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:32:44.540279+00:00, confidence not recorded.
  - readme: https://github.com/THUDM/AgentTuning (fetched 2026-08-28T04:04:54.709851+00:00, sha 150bb09574eb)
  - homepage: https://thudm.github.io/AgentTuning/ (fetched 2026-08-29T11:37:21.666180+00:00, sha 1c45f7fe5437)
- Data as of 2026-08-30T08:39:29.467469+00:00.