ReidLM

Model Details

ReidLM is a fine-tuned version of Meta's LLaMA 3 model, specifically optimized for generating high-quality, contextually accurate responses in the domain of rare diseases. <br> Utilizing the Evol-Instruct methodology, this model was fine-tuned with a dataset of over 400 rare diseases.

Model Description

• Developed by: MSRIT Students
• Model type: Transformer-based Large Language Model (LLM)
• Language(s) (NLP): English
• License: Llama 3 Community License Agreement
• Finetuned from model: Meta-Llama-3-8B-Instruct

Uses

ReidLM is designed for direct use in generating insightful and reliable information to support healthcare professionals and researchers in diagnosing and managing rare diseases. It can be used as an educational tool for training medical students and professionals about rare diseases.

Out-of-Scope Use

ReidLM is specifically designed for generating information related to rare diseases and should not be used for the following purposes:

• Non-Medical Domains: ReidLM is optimized for rare disease information and may not perform well in other domains such as finance, law, general health conditions, or any other non-medical fields.<br>
• General Conversational AI: While capable of generating detailed information on rare diseases, ReidLM may not be suitable for general conversational AI tasks that require a broad understanding of various topics. <br>

Bias, Risks, and Limitations

ReidLM, like all large language models, has inherent biases and limitations that users should be aware of:<br>

• Ethical Concerns: There is a risk of over-reliance on AI for medical decisions, which should always be validated by healthcare professionals.<br>
• Accuracy: While the model strives for accuracy, it may generate incorrect or incomplete information, especially in highly specialized or novel cases.

Getting Started with the Model

Use the code below to get started with the model.

Use with Transformers AutoModelForCausalLM

import transformers
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load the pre-trained model and tokenizer
model_name = "Tanvi03/ReidLM"  
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

def generate_text(prompt, max_length=1000):
    inputs = tokenizer(prompt, return_tensors="pt")
    outputs = model.generate(inputs.input_ids, max_length=max_length, num_return_sequences=1)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

prompt = "Explain MEN-1 with respect to how it affects the pituitary gland. What is the other name for this syndrome?"
generated_text = generate_text(prompt)
print(generated_text)

<!--## Training Details -->

Training Data

This link provides the Evol-Instruct question-and-answer dataset https://raw.githubusercontent.com/M-e-e-n-a/Synthetic-Dataset-Creation/main/combined_dataset.json

<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. --> <!--### Training Procedure --> <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->

Training Hyperparameters

num_train_epochs=3, <br> per_device_train_batch_size=4,<br> gradient_accumulation_steps=2,<br> optim="paged_adamw_8bit",<br> save_steps=1000,<br> logging_steps=30,<br> learning_rate=2e-4,<br> weight_decay=0.01,<br> fp16=True,<br> max_grad_norm=1.0,<br> warmup_ratio=0.1<br>

<!-- --> <!---#### Speeds, Sizes, Times [optional] <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. --> <!---## Evaluation <!-- This section describes the evaluation protocols and provides the results. --> <!---### Testing Data, Factors & Metrics #### Testing Data <!-- This should link to a Dataset Card if possible. --> <!--[More Information Needed] ---> <!---#### Factors <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. --> <!--[More Information Needed] <!---#### Metrics <!-- These are the evaluation metrics being used, ideally with a description of why. --> <!--[More Information Needed] <!---### Results [More Information Needed] #### Summary ## Model Examination [optional] <!-- Relevant interpretability work for the model goes here --> <!--[More Information Needed] <!---## Environmental Impact Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). - **Hardware Type:** [More Information Needed] - **Hours used:** [More Information Needed] - **Cloud Provider:** [More Information Needed] - **Compute Region:** [More Information Needed] - **Carbon Emitted:** [More Information Needed] ## Technical Specifications [optional] ### Model Architecture and Objective [More Information Needed] ### Compute Infrastructure [More Information Needed] #### Hardware [More Information Needed] #### Software [More Information Needed] ## Citation [optional] <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. --> <!---**BibTeX:** [More Information Needed] **APA:** [More Information Needed] ## Glossary [optional] <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. --> <!---[More Information Needed] ## More Information [optional] [More Information Needed] ## Model Card Authors [optional] [More Information Needed] ## Model Card Contact [More Information Needed]--->

Results

Evaluation Metrics

<table> <thead> <tr> <th>Metrics</th> <th>Llama-2-7b</th> <th>Mistral-7b</th> <th>Mixtral-47B</th> <th>ReidLM</th> </tr> </thead> <tbody> <tr> <td align="center">ROUGE-1</td> <td align="center">0.3117</td> <td align="center">0.3188</td> <td align="center">0.2637</td> <td align="center">0.3281</td> </tr> <tr> <td align="center">ROUGE-2</td> <td align="center">0.1867</td> <td align="center">0.1176</td> <td align="center">0.1573</td> <td align="center">0.1270</td> </tr> <tr> <td align="center">ROUGE-L</td> <td align="center">0.1818</td> <td align="center">0.1449</td> <td align="center">0.2637</td> <td align="center">0.2031</td> </tr> <tr> <td align="center">ROUGE-LSUM</td> <td align="center">0.1818</td> <td align="center">0.1449</td> <td align="center">0.2637</td> <td align="center">0.2031</td> </tr> <tr> <td align="center">METEOR</td> <td align="center">0.0693</td> <td align="center">0.3088</td> <td align="center">0.4377</td> <td align="center">0.3662</td> </tr> <tr> <td align="center">BERTScore</td> <td align="center">0.8262</td> <td align="center">0.8538</td> <td align="center">0.9070</td> <td align="center">0.8782</td> </tr> <tr> <td align="center">G-Eval</td> <td align="center">0.35</td> <td align="center">0.42</td> <td align="center">0.78</td> <td align="center">0.87</td> </tr> <tr> <td align="center">QAG Score</td> <td align="center">0.1046</td> <td align="center">0.2061</td> <td align="center">0.3762</td> <td align="center">0.2609</td> </tr> </tbody> </table>