#4141st of 42 in Open models
chinese-llama-alpaca-2
Llama-2 models with extended Chinese vocabulary, plus training scripts
Live demo ↗
(opens in a new tab)Documentation ↗
(opens in a new tab)Repository on GitHub ↗
(opens in a new tab)
- Stars
- 7.1k
- License
- Apache-2.0
- Last commit
- Apr 2026
- Last release
- Jan 2024
- Language
- Python
Missed the latest GitHub check; these facts may be older.
Overview
Releases Chinese-LLaMA-2 base and Chinese-Alpaca-2 instruction models (1.3B, 7B, 13B) built on Llama-2, with a 55,296-token Chinese vocabulary and incremental pretraining on 120 GB of Chinese text. Long-context variants reach 16K (PI) and 64K (YaRN), and RLHF versions exist for 1.3B and 7B. Includes pretraining and fine-tuning scripts and guides for llama.cpp, transformers, vLLM and LangChain.
Who it is for: Engineers needing open Chinese-capable Llama-2 models or a fine-tuning recipe
Strengths
- Full, LoRA, GGUF and AWQ weights published on Hugging Face and ModelScope
- Context options of 4K, 16K and 64K, with NTK extension beyond that
- Pretraining and instruction-tuning scripts included, with FlashAttention-2 support
- Documented paths for llama.cpp, transformers, vLLM, text-generation-webui and LangChain
Weaknesses
- Superseded: README recommends upgrading to the Llama-3-based Chinese-LLaMA-Alpaca-3
- Last release was 2024-01-23 (v4.1)
- LoRA weights cannot be used alone and must be merged with original Llama-2
- Vocabulary differs from first-generation models, so the two cannot be mixed
What it needs
- GPU optional
- Needs Llama-2 base weights (for LoRA merge), transformers, llama.cpp
- Models: Chinese-LLaMA-2, Chinese-Alpaca-2, GGUF, AWQ, LoRA
Also in Open models
See all 42| Rank | Project | Score |
|---|---|---|
| 1 | llmfitRust tool that checks which LLMs fit your RAM, CPU and GPU | 87 out of 100 |
| 2 | layaSingle-pass multilingual classifier for typed choice, score and yes/no decisions | 86 out of 100 |
| 3 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 76 out of 100 |
| #4 | speechPyTorch toolkit for training and deploying ASR, TTS and speech LLM models | 73 out of 100 |
| #5 | yueLyrics-to-song model that writes an editable melody and chord score first | 69 out of 100 |