Update README.md

README.md

@@ -1,3 +1,126 @@
----
-license: mit
----
+---
+language: en
+license: apache-2.0
+library_name: peft
+tags:
+- h2oai
+- causal-lm
+- text-generation
+- adhd
+- cpt-ii
+- clinical-assistant
+base_model: h2oai/h2o-danube3-500m-chat
+---
+
+# ADHD CPT Analyst
+
+## Model Description
+
+This model is a fine-tuned version of `h2oai/h2o-danube3-500m-chat`, specifically adapted for analyzing and interpreting textual reports from the Conners' Continuous Performance Test II (CPT-II). It has been trained using Low-Rank Adaptation (LoRA) on a dataset of CPT-II results to identify patterns relevant to the assessment of ADHD.
+
+The model takes a textual summary of a patient's CPT-II scores as input and can provide analysis, explanations of the metrics, and potential interpretations.
+
+## Intended Uses & Limitations
+
+This model is intended as a research and educational tool. It can be used to:
+- Assist researchers in analyzing patterns across large datasets of CPT-II reports.
+- Help students and trainees learn about the different metrics in a CPT-II report and their potential clinical significance.
+- Provide a preliminary interpretation of a CPT-II report.
+
+**Crucial Disclaimer:** This model is **not a medical device** and should **not** be used for self-diagnosis or as a substitute for professional medical advice, diagnosis, or treatment. Always consult with a qualified healthcare provider for any health-related concerns.
+
+## How to Use
+
+To use this model, you need to load the base model (`h2oai/h2o-danube3-500m-chat`) and then apply the LoRA adapter.
+
+```python
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+from peft import PeftModel
+
+# Model and repository parameters
+base_model_id = "h2oai/h2o-danube3-500m-chat"
+adapter_id = "monkwarrior08/adhd-cpt-analyst"
+
+# Load tokenizer and base model
+tokenizer = AutoTokenizer.from_pretrained(base_model_id)
+base_model = AutoModelForCausalLM.from_pretrained(
+    base_model_id,
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+)
+
+# Load the LoRA adapter
+model = PeftModel.from_pretrained(base_model, adapter_id)
+model.eval()
+
+# Create a prompt with a patient's data
+patient_report = """
+Patient ID: 3.0
+Assessment Status: 3.0
+Assessment Duration: 839999.0 seconds
+
+CPT II Summary Report:
+- Omissions:
+  - General T-Score: 78.75
+  - ADHD T-Score: 70.25
+  - Raw Score: 11.0
+- Commissions:
+  - General T-Score: 65.98
+  - ADHD T-Score: 70.89
+  - Raw Score: 28.0
+- Hit Reaction Time (HitRT):
+  - General T-Score: 36.57
+  - Mean Reaction Time: 325.20 ms
+ADHD Confidence Index: 86.87
+"""
+
+prompt = f"<|prompt|>Analyze this CPT-II report and summarize the findings for potential indicators of ADHD.:\\n{patient_report}<|end|><|answer|>"
+
+# Generate a response
+inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+outputs = model.generate(
+    **inputs, 
+    max_new_tokens=256, 
+    eos_token_id=tokenizer.eos_token_id,
+    do_sample=True,
+    temperature=0.6,
+    top_p=0.9,
+)
+
+response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+# Extract only the model's answer
+answer = response.split('<|answer|>')[1].strip()
+print(answer)
+```
+
+## Training Data
+
+The model was fine-tuned on a private dataset derived from the `ADHD Diagnosis CPT II Data.csv` file. Each record was converted into a textual summary containing the following key metrics:
+- Omissions (T-Scores, Raw Score)
+- Commissions (T-Scores, Raw Score)
+- Hit Reaction Time (T-Score, Mean)
+- Variability of SE
+- d'
+- Beta
+- Perseverations
+- ADHD Confidence Index
+
+## Training Procedure
+
+The model was trained using the `trl` library's `SFTTrainer` with a LoRA configuration. The primary goal was to teach the model to understand the relationship between the various CPT-II metrics and their relevance in ADHD assessment.
+
+### BibTeX Citation
+
+If you use this model in your research, please consider citing it:
+
+```bibtex
+@software{monkwarrior08_2024_adhd_cpt_analyst,
+  author = {monkwarrior08},
+  title = {ADHD CPT Analyst: A Fine-tuned Language Model for CPT-II Report Interpretation},
+  month = {8},
+  year = {2024},
+  url = {https://huggingface.co/monkwarrior08/adhd-cpt-analyst}
+}
+```