bagasbgs2516/llama3.2-3B-cs-Tokyo8

Model Card for Tokyo8 Customer Service Model (Indonesia)

This is the model card of a 🤗 transformers model based on the Llama-3.2-3B architecture, which has been finetuned by PT Clevio on a dataset of customer service conversations for the Tokyo8 product in Indonesia.

Model Description

This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.

• Developed by: PT Clevio
• Funded by : PT Clevio
• Shared by : PT Clevio
• Model type: Causal Language Model
• Language(s) (NLP): English, Indonesian
• License: Apache-2.0
• Finetuned from model : meta-llama/Llama-3.2-3B

Model Sources [optional]

• Repository: bagasbgs2516/llama3.2-3B-cs-Tokyo8

Uses

Direct Use

The Tokyo8 Customer Service Model (Indonesia) can be used for generating responses to customer inquiries and providing helpful information about the Tokyo8 product in the Indonesian market.

Downstream Use

This model can be integrated into customer service chatbots or virtual assistants to provide automated responses to customer questions about the Tokyo8 product in Indonesia.

Out-of-Scope Use

The model should not be used for generating content unrelated to the Tokyo8 product or for purposes outside of customer service in Indonesia. The model may also have biases or limitations that should be considered before deployment.

Bias, Risks, and Limitations

The model was trained on a dataset of customer service conversations in Indonesia, which may contain biases or reflect certain demographic or linguistic patterns. The model's performance may also be limited to the specific domain of the Tokyo8 product in Indonesia and may not generalize well to other products or domains.

Recommendations

Users should be aware of the potential biases and limitations of the model, and should carefully evaluate its performance and appropriateness for their use case. Monitoring for unintended outputs or biases is recommended, and the model should be used with appropriate safeguards and oversight.

How to Get Started with the Model

Use the code below to get started with the Tokyo8 Customer Service Model (Indonesia).

# Load model From Hugging Face Hub
model = AutoModelForCausalLM.from_pretrained("bagasbgs2516/llama3.2-3B-cs-Tokyo8")
tokenizer = AutoTokenizer.from_pretrained("bagasbgs2516/llama3.2-3B-cs-Tokyo8")

# Try for inferensi
input_text = "Halo, Apa itu Tokyo8?"
input_ids = tokenizer.encode(input_text, return_tensors="pt")

output = model.generate(input_ids, max_length=50, num_return_sequences=1)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Training Details

Training Data

The model was finetuned on a dataset of customer service conversations related to the Tokyo8 product in Indonesia, which was provided in a CSV file.

Training Procedure

Preprocessing

The dataset was preprocessed by tokenizing the text and splitting the conversations into input-response pairs.

Training Hyperparameters

• Training regime: The model was finetuned for 6 epochs, with a batch size of 1, a gradient accumulation of 8 steps, a learning rate of 5e-4, and a weight decay of 0.01.

Speeds, Sizes, Times

The finetuning process took approximately 1.5 hours on a NVIDIA T4 GPU with 15GB of memory.

Evaluation

Testing Data, Factors & Metrics

Testing Data

The model was evaluated on a held-out portion of the customer service conversation dataset, which was not used for training.

Factors

The evaluation considered factors such as the quality and relevance of the generated responses, as well as the model's ability to handle different types of customer inquiries in the Indonesian market.

Metrics

The model's performance was evaluated using metrics such as perplexity, BLEU score, and human evaluation. The model achieved a BLEU score of 0.4227 and the precision scores for different n-gram levels ranged from 0.1419 to 0.2038.

Results

The Tokyo8 Customer Service Model (Indonesia) achieved a BLEU score of 0.4227 and demonstrated promising performance in generating relevant and helpful responses to customer inquiries based on the evaluation metrics.

Summary

The Tokyo8 Customer Service Model (Indonesia) has been successfully finetuned by PT Clevio on a dataset of customer service conversations related to the Tokyo8 product in the Indonesian market. The model demonstrates good performance in generating relevant and helpful responses to customer inquiries, but may have biases or limitations that should be carefully considered before deployment.

Model Architecture and Objective

The Tokyo8 Customer Service Model (Indonesia) is based on the Llama-3.2-3B architecture, which is a large language model trained for causal language modeling. The objective of the finetuning process was to adapt the model to the specific domain of customer service conversations for the Tokyo8 product in the Indonesian market.

Compute Infrastructure

The model was finetuned on a system with 1 NVIDIA T4 GPU with 15GB of memory, 51GB of RAM.

Hardware

• NVIDIA T4 GPU (15GB)
• RAM 51GB

Software

• PyTorch
• Transformers library
• Peft

Citation

BibTeX:

@misc{tokyo8-customer-service-model-indonesia,
    title={Tokyo8 Customer Service Model (Indonesia)},
    author={PT Clevio},
    year={2024},
    howpublished={\url{https://huggingface.co/bagasbgs2516/llama3.2-3B-cs-Tokyo8}},
}

APA:

PT Clevio. (2024). Tokyo8 Customer Service Model (Indonesia). Hugging Face. https://huggingface.co/bagasbgs2516/llama3.2-3B-cs-Tokyo8

Model Card Contact

For more information or inquiries about the Tokyo8 Customer Service Model (Indonesia), please contact: Bagas Tri Adiwira [email protected] PT Clevio