Add model card and metadata for PVminerLLM2_8B

README.md

@@ -0,0 +1,42 @@
+---
+library_name: transformers
+pipeline_tag: text-generation
+---
+
+# PVminerLLM2_8B
+
+This repository contains the 8B parameter model for **PVminerLLM2**, as presented in the paper [PVminerLLM2: Improving Structured Extraction of Patient Voice via Preference Optimization](https://huggingface.co/papers/2606.16074).
+
+## Model Description
+PVminerLLM2 is a set of Large Language Models (LLMs) specialized for the structured extraction of "patient voice" from patient-generated text (e.g., lived experiences, social context). The model improves upon previous iterations by applying preference optimization to address token-critical errors that are often missed by standard supervised fine-tuning (SFT). 
+
+Key innovations include:
+- A preference objective with a token-level gated stabilization term.
+- Confusion-aware preference pair construction to better capture low-separation distinctions.
+- Token-importance weighting and inverse-frequency reweighing to address class skew.
+
+## Resources
+- **Paper:** [PVminerLLM2: Improving Structured Extraction of Patient Voice via Preference Optimization](https://huggingface.co/papers/2606.16074)
+- **GitHub Repository:** [Data-Mining-Lab-Yale/PVminerLLM2](https://github.com/Data-Mining-Lab-Yale/PVminerLLM2)
+
+## Usage
+
+The official implementation for inference and building confusion summaries is available in the GitHub repository. To run inference using the provided scripts:
+
+```bash
+python infer_vllm_and_confusion.py \
+  --model lm2445/PVminerLLM2_8B \
+  --data /path/to/hf_dataset \
+  --out_code_csv outputs/code_confusion_summary.csv \
+  --out_subcode_csv outputs/subcode_confusion_summary.csv \
+  --out_pred_jsonl outputs/pred_dump.jsonl \
+  --tp 1
+```
+
+## Citation
+If you use this model in your research, please cite:
+```bibtex
+@article{PVminerLLM2,
+  title={PVminerLLM2: Improving Structured Extraction of Patient Voice via Preference Optimization}
+}
+```