flwr new @zexili/flwr-nlp

FlowerTune LLM (General NLP) — Federated Instruction Tuning

This Flower app performs federated instruction tuning for a pretrained LLM on the General NLP task of the FlowerTune LLM leaderboard.

• Dataset (default): vicgalle/alpaca-gpt4 via Flower Datasets
• Fine-tuning: LoRA (🤗 PEFT)
• Orchestration: Flower Simulation Engine
• Aggregation: FedAvg

All app settings are configured in pyproject.toml under [tool.flwr.app.config].

Quickstart

1) Create environment & install

conda create -n flwr-tune python=3.9
conda activate flwr-tune

pip install -e .

2) Run

flwr run

Tip: If the base model requires gated access on Hugging Face, make sure you have access and have authenticated with hf auth login.

Project structure

flowertune-llm-general-nlp/
├─ flwr_nlp/
│  ├─ client_app.py
│  ├─ server_app.py
│  ├─ dataset.py
│  ├─ models.py
│  └─ strategy.py
├─ pyproject.toml
└─ README.md

Notes

• This repository is intentionally minimal for leaderboard submission: it contains only the federated fine-tuning app code and required configuration.
• Logs, checkpoints, and downstream evaluation code are excluded.