You need to agree to share your Hugging Face information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

Pearl Chat 500M IT

Pearl Chat is an instruction-tuned causal language model developed by the AI for Uganda Initiative, focused on Luganda text generation.

Model details

  • Architecture: PearlChatV2Model, a causal Transformer.
  • Parameters: 466,682,112 total.
  • Languages: Luganda and English.
  • Framework: JAX and Flax NNX.
  • Tokenizer: Custom Luganda BPE tokenizer.
  • License: Apache 2.0.

Intended use

The model is intended for research and applications involving:

  • Luganda conversational question answering.
  • Luganda text generation and continuation.
  • Luganda-English translation.
  • African language NLP research.

Generated outputs may require human review, particularly for factual, technical, medical, legal, or other high-stakes applications.

Training data

The model was trained using the following datasets:

See the linked dataset cards for additional information about the data.

Evaluation

Evaluation results are provided in benchmark_scorecard.json.

Results should be interpreted in the context of the evaluated tasks, datasets, metrics, and baselines. They do not establish performance across all Luganda use cases.

Limitations

Pearl Chat may produce inaccurate, incomplete, or misleading outputs. Its performance may vary across domains, dialects, and language varieties. It should not be used as the sole basis for high-stakes decisions.

The released weights use the project's native JAX/Flax NNX architecture. The Safetensors export does not, by itself, guarantee compatibility with standard Hugging Face Transformers loading APIs.

Artifacts

The repository includes:

  • model.safetensors: exported model weights.
  • model_state/: native model state.
  • config.json: model configuration.
  • Tokenizer files.
  • model_parameter_audit.json: parameter audit.
  • benchmark_scorecard.json: evaluation results.

Citation

If you use Pearl Chat in research, please cite:

Kambale, W. (2026). Pearl Chat: A Native Luganda Conversational Language Model. AI for Uganda Initiative.

@misc{kambale2026pearlchat,
  author = {Kambale, Wesley},
  title = {Pearl Chat: A Native Luganda Conversational Language Model},
  year = {2026},
  publisher = {AI for Uganda Initiative},
  url = {https://huggingface.co/ai4ug/pearl-chat-500m-it}
}

Project: AI for Uganda Initiative

Model: https://huggingface.co/ai4ug/pearl-chat-500m-it

Downloads last month
5
Safetensors
Model size
0.5B params
Tensor type
F16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Datasets used to train ai4ug/pearl-chat-500m-it