# MiuLab/Taiwan-LLM

Traditional Mandarin LLMs for Taiwan

Repository: https://github.com/MiuLab/Taiwan-LLM
Canonical: https://ross.abutalabs.com/products/taiwan-llm
Homepage: https://twllm.com
Language: Python
License: Apache-2.0
License Family: permissive
Topics: langauge-model, llm, taiwan, traditional-mandarin, twllm
Last push: 2025-04-20T02:21:55+00:00

## Health v2 (maintenance only)
Score: 26/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 17, release rhythm 8, longevity 79
- inputs: {"age_days": 1119, "days_push": 501, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1426, forks 119 (observed 2026-08-28T04:04:41.645157+00:00)

## What it is
A collection of Traditional Mandarin large language models (Taiwan-LLM / TAME) built on the Llama-3 architecture, fine-tuned on Traditional Mandarin and English corpora for Taiwanese culture and domains. The repository provides model weights, fine-tuning configs for Axolotl, evaluation papers, and an online chat demo.

## Use cases
- fine-tune a Traditional Mandarin chat model on my own Taiwan-specific data
- get a Llama-3-based LLM that understands Traditional Mandarin and Taiwanese context
- run a Traditional Mandarin NLP benchmark to compare Chinese-language models
- deploy a chatbot that speaks Traditional Mandarin for Taiwan users
- adapt an open LLM to legal or medical text in Traditional Mandarin
- train a mixture-of-experts model for multilingual Traditional Mandarin and English tasks

## When to choose
- you need strong Traditional Mandarin (Taiwan) language understanding or generation
- you want open Apache-2.0 licensed Llama-3-based weights to fine-tune or self-host
- you are benchmarking Chinese-language LLMs against Taiwan-specific NLP tasks
- you need a starting point for Axolotl or NVIDIA NeMo fine-tuning pipelines

## When to avoid
- you need Simplified Chinese or mainland-China-specific language models
- you want a ready-to-use hosted API without managing your own inference
- you lack GPU resources and only need a small lightweight model
- you need non-Chinese multilingual coverage beyond English and Mandarin

## Facets
- artifact type: learning-resource
- maturity: active
- function: llm-training, nlp, machine-learning, chatbot
- domain: large-language-models, artificial-intelligence
- platform: python, cloud
- tags: traditional-mandarin, taiwan, llama-3, fine-tuning, axolotl, nemo-framework, mixture-of-experts, chinese-language-model, model-weights, benchmarking, natural-language-processing, gpu, docker

## Member repositories
- MiuLab/Taiwan-LLM (main) score 26

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:41.645157+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 2026-08-30T04:37:25.100626+00:00, confidence not recorded.
  - readme: https://github.com/MiuLab/Taiwan-LLM (fetched 2026-08-28T04:04:41.645157+00:00, sha 37b5f16dc8cc)
- Data as of 2026-08-30T08:39:29.467469+00:00.