Research map / Large language models / Instruction and fine-tuning
PandemIQ Llama: A Domain-Adapted Foundation Model for Enhanced Pandemic Intelligence
Jingmei Yang, Mahtab Talaei, Britta Lassmann, Nahid Bhadelia, Ioannis Ch. Paschalidis
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add continuous pre-training on the Pandemic Corpus before LoRA fine-tuning (ablation) | Llama-3.1-8B · BAND QA | overall accuracy (%): 77.38 -> 86.07 | Improved |
| add continuous pre-training on the Pandemic Corpus before LoRA fine-tuning (ablation) | Llama-3.1-8B · Disease Event Classification | Macro-F1 (%): 74.51 -> 79.80 | Improved |
| use customized epidemiology prompt instead of default prompt (ablation) | GPT-4o · BAND QA | overall accuracy (%): 64.61 -> 79.21 | Improved |
| use customized prompt with category descriptions instead of default prompt (ablation) | GPT-4o · Disease Event Classification | Macro-F1 (%): 53.61 -> 66.68 | Improved |
| use customized prompt instead of default prompt (ablation) | Llama-3.2-3B-Instruct · BAND QA | overall accuracy (%): 57.58 -> 58.56 | No clear effect |
| use customized prompt instead of default prompt (ablation) | Llama-3.2-1B-Instruct · Disease Event Classification | Macro-F1 (%): 19.55 -> 13.16 | Hurt |
| increase LoRA rank from 8 to 256 (ablation) | Llama-3.1-8B-Instruct · BAND QA | overall accuracy (%): 63.55 -> 77.38 | Improved |
| increase LoRA rank from 8 to 256 (ablation) | Llama-3.1-8B-Instruct · Disease Event Classification | Macro-F1 (%): 61.46 -> 74.51 | Improved |
| continuous pre-training gain shrinks as LoRA rank grows (ablation) | PandemIQ Llama vs Llama-3.1-8B-Instruct · BAND QA | accuracy gain (points): +17.32 (rank 8) -> +8.69 (rank 256) | Mixed |
| scale model size in prompt-engineered Llama models (ablation) | Llama zero-shot · BAND QA | overall accuracy (%): 45.84 (3.2-1B) -> 67.51 (3.3-70B), customized prompt | Improved |
| continuous pre-training effect on Violation category (ablation) | PandemIQ Llama · Event classification, Violation F1 | F1 (%) | No clear effect |
Closest papers by meaning
- Multilingual Medical Language Models: A Path to Improving Lay Health Worker Effectiveness (Student Abstract) (AAAI 2024 · similarity 0.85)
- Cross-Care: Assessing the Healthcare Implications of Pre-training Data on Language Model Bias (NeurIPS 2024 · similarity 0.84)
- Evaluating AI-based Scientific Knowledge Synthesis with Epidemiological Systematic Reviews (NeurIPS 2026 · similarity 0.83)
- MedJourney: Benchmark and Evaluation of Large Language Models over Patient Clinical Journey (NeurIPS 2024 · similarity 0.83)
- Multi-OphthaLingua: A Multilingual Benchmark for Assessing and Debiasing LLM Ophthalmological QA in LMICs (AAAI 2025 · similarity 0.82)
- Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks (NeurIPS 2026 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
