id license: apache-2.0 tags:
Indonesian
Chat
Instruct base_model:
meta-llama/Llama-3.2-3B-Instruct datasets:
NekoFi/alpaca-gpt4-indonesia-cleaned pipeline_tag: text-generation model-index:
name: FinMatcha-3B-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: 75.48 name: strict accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=xMaulana/FinMatcha-3B-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: 23.19 name: normalized accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=xMaulana/FinMatcha-3B-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: 12.39 name: exact match source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=xMaulana/FinMatcha-3B-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: 2.57 name: acc_norm source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=xMaulana/FinMatcha-3B-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: 5.02 name: acc_norm source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=xMaulana/FinMatcha-3B-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: 24.24 name: accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=xMaulana/FinMatcha-3B-Instruct name: Open LLM Leaderboard

FinMatcha-3B-Instruct

FinMatcha is a powerful Indonesian-focused large language model (LLM) fine-tuned from the Llama-3.2-3B-Instruct base model. The model has been trained to handle a variety of conversation, with a special emphasis on understanding and generating Indonesian text.

This model has been fine-tuned on a wide array of Indonesian datasets, making it adept at handling the nuances of the Indonesian language, from formal to colloquial speech. It also supports English for bilingual applications.

Model Details

Finetuned from model: Llama-3.2-3B-Instruct
Dataset: NekoFi/alpaca-gpt4-indonesia-cleaned
Model Size: 3B
License: Apache-2.0
Languages: Indonesian, English

How to use

Installation

To use the Finmatcha model, install the required dependencies:

pip install transformers>=4.45

Usage

Google Colab

python

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "xMaulana/FinMatcha-3B-Instruct"
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_id)

inputs = tokenizer("Bagaimanakah sebuah negara dapat terbentuk?", return_tensors="pt").to("cuda")
outputs = model.generate(inputs.input_ids, 
                          max_new_tokens = 2048,
                          pad_token_id=tokenizer.pad_token_id,
                          eos_token_id=tokenizer.eos_token_id,
                          temperature=0.7,
                          do_sample=True, 
                          top_k=5, 
                          top_p=0.9,
                          repetition_penalty=1.1
                         )
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Limitations

The model is primarily focused on the Indonesian language and may not perform as well on non-Indonesian tasks.
As with all LLMs, cultural and contextual biases can be present.

License

The model is licensed under the Apache-2.0.

Contributing

We welcome contributions to enhance and improve Finmatcha. Feel free to open issues or submit pull requests for improvements.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	23.81
IFEval (0-Shot)	75.48
BBH (3-Shot)	23.19
MATH Lvl 5 (4-Shot)	12.39
GPQA (0-shot)	2.57
MuSR (0-shot)	5.02
MMLU-PRO (5-shot)	24.24

RichardErkhov/xMaulana_-_FinMatcha-3B-Instruct-8bits

Original model description: