// Install AgentJet swarm server using Conda. Handles Python 3.10 environment creation, dependency installation with the verl training backbone, flash-attn compilation, and optional PyPI mirror for China users.
| name | conda-install-agentjet-swarm-server |
| description | Install AgentJet swarm server using Conda. Handles Python 3.10 environment creation, dependency installation with the verl training backbone, flash-attn compilation, and optional PyPI mirror for China users. |
| license | Complete terms in LICENSE.txt |
when the user only need to run agentjet client, and do not have to run models locally (e.g. user in their laptop), ONLY install AgentJet basic requirements is enough (pip install -e .). see
install-agentjet-clientskill
Check Python version requirement (3.10) and conda availability.
Verify user has conda installed:
conda --version
Clone the AgentJet repository from GitHub and navigate into the project directory:
git clone https://github.com/modelscope/AgentJet.git
cd AgentJet
Create a new conda environment with Python 3.10:
conda create -n ajet-verl python=3.10
conda activate ajet-verl
Install AgentJet with the verl training backbone:
pip install -e .[verl]
Use Aliyun PyPI mirror for faster downloads:
pip install -i https://mirrors.aliyun.com/pypi/simple/ -e .[verl]
Before installing flash-attn, test the connection to GitHub:
curl -I --connect-timeout 10 https://github.com 2>/dev/null | head -n 1
If the connection test fails or times out:
!!! danger "IMPORTANT: GitHub Connection Issue Detected" ⚠️ WARNING: You cannot connect to GitHub stably!
flash-attn requires downloading pre-compiled wheels from GitHub. Without a stable GitHub connection, pip will attempt to build flash-attn from source, which can take **30 minutes to several hours**.
**Recommended actions:**
1. Set up a stable connection to GitHub (VPN, proxy, etc.)
2. Use a mirror that hosts flash-attn pre-compiled wheels
3. If you must proceed, set `MAX_JOBS` to limit CPU usage during build:
```bash
export MAX_JOBS=4 # Adjust based on your CPU cores
```
**Do you want to proceed anyway?** (This may take a very long time)
flash-attn must be installed after other dependencies:
pip install --verbose flash-attn --no-deps --no-build-isolation --no-cache
pip install -i https://mirrors.aliyun.com/pypi/simple/ --verbose flash-attn --no-deps --no-build-isolation --no-cache
!!! warning "flash-attn Installation"
- flash-attn must be installed after other dependencies.
- Ensure a healthy connection to GitHub to install pre-compiled wheels.
- If you find your machine spend a long time installing flash-attn, ensure a healthy connection to GitHub.
- To build faster, export MAX_JOBS=${N_CPU}.
Verify the installation:
python -c "import ajet; print(ajet.__version__)"
Execute AgentJet reinforcement learning experiments using experiment blueprints in classic (non-swarm) mode. Handles full lifecycle: launch experiment in tmux, monitor progress, analyze errors, collect results, and write finish flag. Use when the user wants to run AgentJet training experiments without the swarm distributed framework.
Execute AgentJet reinforcement learning experiments using experiment blueprints in swarm mode. Handles full lifecycle: generate blueprint if needed, launch experiment in tmux, monitor progress, analyze errors, collect results, and write finish flag. Use when the user wants to run or debug AgentJet training experiments.
Install and run the AgentJet Swarm Server in a Docker container with NVIDIA GPU support. Use when the user wants to deploy a swarm server on a GPU machine via Docker, including GPU driver setup, Docker mirror configuration, model weight mounting, and server startup.
Download per-step time-series metric data (reward, entropy, response length, etc.) from a SwanLab cloud run URL as a pandas.DataFrame. Use when the user provides a SwanLab URL and wants to fetch or analyze training curves.
Install AgentJet client for connecting to a swarm server. Use when the user only needs to run the AgentJet client (not a swarm server) and does not need to run models locally, e.g. on a laptop. Installs basic requirements via `pip install -e .`.
Map VERL training configuration to AgentJet configuration. Find VERL config in verl_default.yaml, check for existing mappings in config_auto_convertion_verl.jsonc, add new mappings to ajet_default.yaml and the conversion schema, and optionally add parameters to AgentJetJob.
