README.md

metadata

language:
  - zh
  - en
license: apache-2.0
tags:
  - Cantonese
  - Qwen2
  - chat
  - mlx
datasets:
  - jed351/cantonese-wikipedia
  - raptorkwok/cantonese-traditional-chinese-parallel-corpus
pipeline_tag: text-generation
model-index:
  - name: Qwen2-Cantonese-7B-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: 54.35
            name: strict accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=lordjia/Qwen2-Cantonese-7B-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: 32.45
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=lordjia/Qwen2-Cantonese-7B-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: 8.76
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=lordjia/Qwen2-Cantonese-7B-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: 6.04
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=lordjia/Qwen2-Cantonese-7B-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: 7.81
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=lordjia/Qwen2-Cantonese-7B-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: 31.59
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=lordjia/Qwen2-Cantonese-7B-Instruct
          name: Open LLM Leaderboard
library_name: mlx

This is a MLX conversion of lordjia/Qwen2-Cantonese-7B-Instruct

Qwen2-Cantonese-7B-Instruct

Model Overview / 模型概述

Qwen2-Cantonese-7B-Instruct is a Cantonese language model based on Qwen2-7B-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.

Qwen2-Cantonese-7B-Instruct係基於Qwen2-7B-Instruct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: Qwen2-7B-Instruct
• Fine-tuning Method: LoRA instruction tuning
• Training Steps: 4572 steps
• Primary Language: Cantonese / 粵語
• Datasets:
  • jed351/cantonese-wikipedia
  • raptorkwok/cantonese-traditional-chinese-parallel-corpus
• Training Tools: LLaMA-Factory

Quantized Version / 量化版本

A 4-bit quantized version of this model is also available: qwen2-cantonese-7b-instruct-q4_0.gguf.

此外，仲提供此模型嘅4位量化版本：qwen2-cantonese-7b-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：

1. Llama-3-Cantonese-8B-Instruct based on Meta-Llama-3-8B-Instruct.
2. Llama-3.1-Cantonese-8B-Instruct based on Meta-Llama-3.1-8B-Instruct.

License / 許可證

This model is licensed under the Apache 2.0 license. Please review the terms before use.

此模型喺Apache 2.0許可證下獲得許可。 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.	23.50
IFEval (0-Shot)	54.35
BBH (3-Shot)	32.45
MATH Lvl 5 (4-Shot)	8.76
GPQA (0-shot)	6.04
MuSR (0-shot)	7.81
MMLU-PRO (5-shot)	31.59