TigerBot is a Python-based, multi-language LLM foundation with open-source releases and a wide set of models focused on Chinese language tasks. It provides model downloads, environment setup, and inference/finetuning capabilities across 70B and 13B variants, plus API components.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
TigerBot is a multi-language, multi-task foundation for large language models, with open-source releases and a collection of models (e.g., tigerbot-70b-base, tigerbot-70b-chat, tigerbot-13b-base, tigerbot-13b-chat) intended for Chinese and multilingual tasks. The repository lists model download options, environment installation steps, and references to related tooling and APIs.
How it works
The README describes a stack with base and chat variants trained on large token data, using methods such as holistic training, rejection sampling for alignment, and integration with search and external tools via TigerBot-API. It also references streaming and function-calling capabilities in some releases. The project includes model download tables and guidance for inference via provided scripts.
Getting started
Environment installation snippet:
conda create --name tigerbot python=3.8
conda activate tigerbot
conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
git clone https://github.com/TigerResearch/TigerBot
cd TigerBot
pip install -r requirements.txt
Model download table lists various model versions and their architectures (llama-2) and disk sizes (GB).
Recent releases
The release section for the latest version indicates: - none The README includes a historical timeline of updates and model releases dating back to 2023, with multiple iterations of 7B/13B/70B models and associated tooling; latest release note in the snippet shows up to 12/28/2024 in the Chinese release log, but the explicit RELEASES (latest 0) entry states none.
Traction
Stars: 2261, Forks: 189, Open issues: 45. Language: Python. License: Apache-2.0.
Behind the repo
TigerBot links to multiple external platforms (Hugging Face, ModelScope, and a medical terminology initiative MedCT) as part of its ecosystem. The README references these external resources and model collections.
Caveats
Environment installation uses Python 3.8 and CUDA setup; model download table shows multiple variants with disk sizes and versions. The latest releases section states none. The license is Apache-2.0.