LLM Fine-tuning and Prompt Engineering for Legal RAG & Sentiment Classification
TextFine Tuning
The candidate fine-tuned the Llama 3.1 model using QLoRA with 4-bit quantization for legal and sentiment analysis tasks. They engineered a custom chunking and annotation algorithm for hierarchical legal texts and designed a pipeline for Chain-of-Thought prompting. Extensive prompt engineering and output quality control were implemented to refine the model's legal reasoning and eliminate hallucinations.
• Refined LLMs for aspect-based sentiment and legal query responses in Vietnamese text.
• Developed and annotated training datasets structured by legal context and user review aspects.
• Applied advanced prompting and automated evaluation metrics for model QA.
• Ensured high annotation quality through automated labeling and manual review.
The candidate fine-tuned the Llama 3.1 model using QLoRA with 4-bit quantization for legal and sentiment analysis tasks. They engineered a custom chunking and annotation algorithm for hierarchical legal texts and designed a pipeline for Chain-of-Thought prompting. Extensive prompt engineering and output quality control were implemented to refine the model's legal reasoning and eliminate hallucinations. • Refined LLMs for aspect-based sentiment and legal query responses in Vietnamese text. • Developed and annotated training datasets structured by legal context and user review aspects. • Applied advanced prompting and automated evaluation metrics for model QA. • Ensured high annotation quality through automated labeling and manual review.
2023 - 2024