metadata

language:
  - en
  - zh
license: llama3
tags:
  - Cantonese
  - chat
  - Llama3
datasets:
  - jed351/cantonese-wikipedia
  - lordjia/Cantonese_English_Translation
pipeline_tag: text-generation
model-index:
  - name: Llama-3-Cantonese-8B-Instruct
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: IFEval (0-Shot)
          type: HuggingFaceH4/ifeval
          args:
            num_few_shot: 0
        metrics:
          - type: inst_level_strict_acc and prompt_level_strict_acc
            value: 66.69
            name: strict accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=lordjia/Llama-3-Cantonese-8B-Instruct
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: BBH (3-Shot)
          type: BBH
          args:
            num_few_shot: 3
        metrics:
          - type: acc_norm
            value: 26.79
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=lordjia/Llama-3-Cantonese-8B-Instruct
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MATH Lvl 5 (4-Shot)
          type: hendrycks/competition_math
          args:
            num_few_shot: 4
        metrics:
          - type: exact_match
            value: 8.23
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=lordjia/Llama-3-Cantonese-8B-Instruct
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GPQA (0-shot)
          type: Idavidrein/gpqa
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 5.82
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=lordjia/Llama-3-Cantonese-8B-Instruct
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MuSR (0-shot)
          type: TAUR-Lab/MuSR
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 9.48
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=lordjia/Llama-3-Cantonese-8B-Instruct
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU-PRO (5-shot)
          type: TIGER-Lab/MMLU-Pro
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 27.94
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=lordjia/Llama-3-Cantonese-8B-Instruct
          name: Open LLM Leaderboard

Llama-3-Cantonese-8B-Instruct

Model Overview / 模型概述

Llama-3-Cantonese-8B-Instruct is a Cantonese language model based on Meta-Llama-3-8B-Instruct, fine-tuned using LoRA. It aims to enhance Cantonese text generation and comprehension capabilities, supporting various tasks such as dialogue generation, text summarization, and question-answering.

Llama-3-Cantonese-8B-Instruct係基於Meta-Llama-3-8B-Struct嘅粵語語言模型，使用LoRA進行微調。它旨在提高粵語文本的生成和理解能力，支持各種任務，如對話生成、文本摘要和問答。

Model Features / 模型特性

Base Model: Meta-Llama-3-8B-Instruct
Fine-tuning Method: LoRA instruction tuning
Training Steps: 4562 steps
Primary Language: Cantonese / 粵語
Datasets:
- jed351/cantonese-wikipedia
- lordjia/Cantonese_English_Translation
Training Tools: LLaMA-Factory

Quantized Version / 量化版本

A 4-bit quantized version of this model is also available: llama3-cantonese-8b-instruct-q4_0.gguf.

此模型的4位量化版本也可用：llama3-cantonese-8b-instruct-q4_0.gguf。

Alternative Model Recommendations / 備選模型舉薦

For alternatives, consider the following models, both fine-tuned by LordJia on Cantonese language tasks:

揾其他嘅話，可以諗下呢啲模型，全部都係LordJia用廣東話嘅工作調教好嘅：

Qwen2-Cantonese-7B-Instruct based on Qwen2-7B-Instruct.
Llama-3.1-Cantonese-8B-Instruct based on Meta-Llama-3.1-8B-Instruct.

License / 許可證

This model is licensed under the Llama 3 Community License. Please review the terms before use.

此模型根據Llama 3社區許可證獲得許可。請在使用前仔細閱讀呢啲條款。

Contributors / 貢獻

LordJia https://ai.chao.cool

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	24.16
IFEval (0-Shot)	66.69
BBH (3-Shot)	26.79
MATH Lvl 5 (4-Shot)	8.23
GPQA (0-shot)	5.82
MuSR (0-shot)	9.48
MMLU-PRO (5-shot)	27.94