| language: | |
| - en | |
| tags: | |
| - peft | |
| - lora | |
| - medical | |
| - triage | |
| - emergency | |
| - text-classification | |
| base_model: google/medgemma-4b-it | |
| library_name: peft | |
| pipeline_tag: text-classification | |
| license: mit | |
| --- | |
| # ESI-1 LoRA Adapter (MIETIC) for MedGemma 4B | |
| ## Model Summary | |
| This repository contains a **LoRA adapter** (not a full standalone model) for **ESI-1 prediction** in emergency triage settings. | |
| The adapter is trained on **MIETIC** using **few-shot, parameter-efficient fine-tuning (PEFT)** on top of **MedGemma 4B** (`google/medgemma-4b-it`). | |
| ## Model Details | |
| - **Model type:** LoRA adapter | |
| - **Base model:** `google/medgemma-4b-it` | |
| - **Task:** ESI-1 prediction (emergency severity triage) | |
| - **Training approach:** Specialized few-shot PEFT | |
| - **Repository owner:** `AdilA1016` | |
| ## Files in this Repo | |
| - `adapter_config.json` | |
| - `adapter_model.safetensors` | |
| - `chat_template.jinja` | |
| - `processor_config.json` | |
| - `tokenizer_config.json` | |
| - `tokenizer.json` | |
| ## Intended Use | |
| This model is intended for **research and decision-support prototyping** for emergency triage workflows. | |
| It is **not** intended to replace clinician judgment. | |
| ## Out-of-Scope / Limitations | |
| - Not validated as an autonomous clinical decision maker. | |
| - Performance may vary by site, population, and documentation style. | |
| - Should not be used as the sole basis for real-time medical decisions. | |
| ## Training Data | |
| - **Dataset:** MIETIC | |
| - **Domain:** Emergency/clinical triage text | |
| - **Label focus:** ESI-1 identification | |
| > Add a short description of MIETIC access/curation and any preprocessing steps you applied. | |
| ## Training Procedure | |
| - **Method:** LoRA fine-tuning on MedGemma 4B | |
| - **Regime:** Few-shot specialized adaptation | |
| - **Frameworks:** PEFT + Transformers | |
| - **Hardware:** [fill in] | |
| - **Epochs / steps:** [fill in] | |
| - **Learning rate:** [fill in] | |
| - **Batch size:** [fill in] | |
| - **LoRA config (`r`, `alpha`, target modules):** [fill in] | |
| ## Evaluation | |
| - **Validation setup:** [fill in] | |
| - **Primary metrics:** [fill in, e.g., recall/precision/F1 for ESI-1] | |
| - **Key results:** [fill in] | |
| - **Failure modes observed:** [fill in] | |
| ## Usage | |
| ```python | |
| from transformers import AutoModelForCausalLM, AutoTokenizer | |
| from peft import PeftModel | |
| base_id = "google/medgemma-4b-it" | |
| adapter_id = "AdilA1016/esi1trainedmodel" | |
| tokenizer = AutoTokenizer.from_pretrained(adapter_id) | |
| base_model = AutoModelForCausalLM.from_pretrained(base_id) | |
| model = PeftModel.from_pretrained(base_model, adapter_id) | |
| ## Safety and Ethics | |
| This model operates in a high-stakes medical context. Outputs may be incorrect, incomplete, or biased. | |
| Human oversight by qualified clinicians is required for any practical use. | |
| ## Citation | |
| If you use this adapter, please cite: | |
| - MIETIC dataset/source: [fill in] | |
| - MedGemma base model: [fill in official citation/link] | |
| - This repository: AdilA1016/esi1trainedmodel |