Phonepadith/aidc-llm-laos-10k-gemma-3-12b-it-v2
Phonepadith/aidc-llm-laos-10k-gemma-3-12b-it-v2 is a 12 billion parameter Gemma 3-based instruction-tuned model developed by Phonepadith, specifically fine-tuned for the Lao language. It excels at summarization, question answering, and general chat in Lao, having been trained on over 5,000 high-quality Lao input-output pairs. This model is particularly optimized for generating concise summaries from Lao text, including news, government documents, and meeting minutes.
Loading preview...
Overview
This model, Phonepadith/aidc-llm-laos-10k-gemma-3-12b-it-v2, is a specialized version of Google's Gemma 3-12B Instruct model, fine-tuned by Phonepadith Phoummavong. Its primary distinction is its deep adaptation to the Lao language (๐ฑ๐ฆ), making it highly proficient in understanding and generating Lao text.
Key Capabilities
- Lao Language Generation: Specifically trained for high-quality text generation in Lao.
- Summarization: Designed to create concise summaries from various Lao text types, such as news, government documents, and meeting minutes.
- Question Answering: Capable of answering questions posed in Lao.
- General Chat: Supports general conversational interactions in Lao.
Training Details
The model was fine-tuned using LoRA with PEFT (Unsloth) on a dataset of 10,000 Laos supervised samples, focusing on AIDC (Artificial Intelligence and Digital Content) topics. It utilized a sequence length of 2048 and was trained for 3-5 epochs with early stopping. Evaluation was performed using the BLEU score to assess summary quality against reference summaries.
When to Use This Model
This model is ideal for applications requiring robust natural language processing in Lao. Developers should consider it for:
- Building Lao-language chatbots or virtual assistants.
- Automating summarization of Lao documents or articles.
- Developing tools for content creation or analysis in Lao.