#2222nd of 23 in Open models
chinese-llama-alpaca
Chinese-vocabulary LLaMA and Alpaca LoRA weights with pretraining and tuning scripts
Live demo ↗
(opens in a new tab)Documentation ↗
(opens in a new tab)Repository on GitHub ↗
(opens in a new tab)
- Stars
- 18.9k
- License
- Apache-2.0
- Last commit
- Apr 2026
- Last release
- Jul 2023
- Language
- Python
Missed the latest GitHub check; these facts may be older.
Overview
Releases LoRA weights that extend LLaMA with a Chinese vocabulary (49,953 tokens) and add Chinese pretraining and instruction tuning, in 7B, 13B and 33B sizes. The LoRA files must be merged with the original LLaMA weights, which are not distributed. Includes merge, quantization, inference and training scripts. The authors recommend moving to their later Llama-3-based project.
Who it is for: Researchers and self-hosters working with Chinese LLaMA-based models
Strengths
- Extended Chinese vocabulary improves Chinese encoding efficiency over the original LLaMA tokenizer
- Pretraining and instruction-tuning scripts are published, so models can be trained further
- Works with llama.cpp, transformers, text-generation-webui, LlamaChat, LangChain and privateGPT
- Weights on Hugging Face and ModelScope, with SHA256 checksums listed
Weaknesses
- Original LLaMA weights are required and not provided; LoRA merge is a manual step
- Built on first-generation LLaMA, whose license forbids commercial use
- Superseded by the authors' Alpaca-2 and Alpaca-3 projects; last release July 2023
- README is Chinese-first; C-Eval scores for the 7B models are low
What it needs
- GPU optional
- Needs Original LLaMA weights, transformers, PEFT, llama.cpp
- Models: LLaMA 7B/13B/33B LoRA, Alpaca Plus/Pro, HF transformers, llama.cpp 8-bit and 4-bit quantized
Also in Open models
See all 23| Rank | Project | Score |
|---|---|---|
| 1 | llmfitRust tool that checks which LLMs fit your RAM, CPU and GPU | 86 out of 100 |
| 2 | layaSingle-pass multilingual classifier for typed choice, score and yes/no decisions | 83 out of 100 |
| 3 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 74 out of 100 |
| #4 | unlimited-ocrOCR model that parses long documents and multi-page PDFs in one pass | 69 out of 100 |
| #5 | speechPyTorch toolkit for training and deploying ASR, TTS and speech LLM models | 67 out of 100 |