Meditron3 8B

🚀 Model Card: Llama-3.1 Meditron-3[8B]

Meditron-3[8B] is a large language model specialized in clinical medicine, co - designed with experts, aiming at various settings including limited - resource ones.

🚀 Quick Start

This README provides detailed information about the Llama-3.1 Meditron-3[8B] model, including its features, training data, usage, and legal disclaimers.

✨ Features

• Specialization in Medicine: Meditron is a suite of large language models specialized in clinical medicine, co - designed with a diverse range of expert clinicians and humanitarian practitioners.
• Equitable Representation: Its training emphasizes equitable representation, contextual diversity, and actionable real - world evidence - based guidelines, especially representing limited - resource and humanitarian settings, neglected populations, and diseases.
• Research - only Model: Meditron-3 is a research - only model to study and evaluate the potential of LLMs in enhancing clinical decision - making and access to evidence - based medical information.

📦 Installation

No installation steps are provided in the original document, so this section is skipped.

💻 Usage Examples

No code examples are provided in the original document, so this section is skipped.

📚 Documentation

Model details

Uses

Direct Use

⚠️ Important Note

Meditron-3 is a research - only model. It is not validated for medical use.

Meditron-3 is a research - only model to study and evaluate the potential of LLMs in enhancing clinical decision - making and access to evidence - based medical information.

Downstream Use

Meditron-3 is a suite of foundation models that have NOT been fine - tuned or instruction - tuned. However, these models can be adapted to specific downstream tasks or applications using techniques such as Reinforcement Learning from Human Feedback (RLHF) or Direct Preference Optimization (DPO). In the evaluation of the models, two different methods for downstream question - answering tasks have been used:

1. In - context learning with k demonstrations added to the prompt.
2. Model fine - tuning for Q&A tasks using specific training datasets.

Training Data

This new data mixture comprises expert - curated publicly available data and combines various sources:

• Clinical Guidelines: a dataset of internationally - recognized clinical practice guidelines from various healthcare - related sources across the world, including hospitals and international organizations.
• Peer - Reviewed Medical Publications: full - text medical articles.
• Synthetic Differential Diagnoses: synthetic conversation - like data for differential diagnosis.
• Replay Data: general domain pretraining data sampled from multiple state - of - the - art pretraining and instruction tuning.
• LLM - enhanced Medical MCQ: medical multiple - choice questions enriched with LLMs.

Additional information about the datasets will be included in the Meditron - 3 publication.

Evaluation

Evaluation results for the Llama[3.1] - Meditron - 3[8B] are coming soon!

We evaluated Meditron on medical multiple - choice questions using [lm - harness](https://github.com/EleutherAI/lm - evaluation - harness) for reproducibility.

⚠️ Important Note

While MCQs are valuable for assessing exam - like performance, they fall short of capturing the model's real - world utility, especially in terms of contextual adaptation in under - represented settings. Medicine is not multiple choice and we need to go beyond accuracy to assess finer - grained issues like empathy, alignment to local guidelines, structure, completeness and safety. To address this, a platform has been developed to collect feedback directly from experts to continuously adapt to the changing contexts of clinical practice.

Paper

The Meditron - 3 publication is currently in progress and will be released at a later date.

Legal Disclaimer

⚠️ Important Note

THIS SOFTWARE AND MODEL ARE PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS, CONTRIBUTORS, OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES, OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT, OR OTHERWISE, ARISING FROM, OUT OF, OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. These models are a research tool intended for use in the field of computational linguistics and medicine. They are not intended to be used as diagnostic tools or for clinical decision - making without appropriate validation and regulatory approval. The content and data provided with the models do not replace the expertise of healthcare professionals. Healthcare professionals should use their professional judgment in evaluating the outputs of the LLaMA models. Patients should not use the model outputs for self - diagnosis or treatment without consulting a qualified healthcare provider. THE INFORMATION IS NOT INTENDED FOR CLINICAL DECISION - MAKING, IS NOT INTENDED TO BE USED IN THE DIAGNOSIS OR TREATMENT OF PATIENTS, AND MAY NOT BE USEFUL OR APPROPRIATE FOR ANY CLINICAL PURPOSE. UNDER NO CIRCUMSTANCES CAN USERS USE THE NAME “YALE” OR "EPFL" OR “YALE UNIVERSITY,” OR ANY AFFILIATED INSTITUTION NOR ANY VARIATION OR ADAPTATION THEREOF, NOR ANY TRADEMARK, TRADENAME OR OTHER DESIGNATION OWNED BY YALE, NOR THE NAMES OF ANY OF ITS TRUSTEES, OFFICERS, FACULTY, STUDENTS, EMPLOYEES OR AGENTS, FOR ANY PURPOSE WITHOUT THE PRIOR WRITTEN CONSENT OF YALE IN EACH INSTANCE, SUCH CONSENT TO BE GRANTED OR WITHHELD BY YALE IN ITS SOLE DISCRETION.

Llama[3.1] - Meditron[8B] is licensed under the Llama 3.1 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved. By downloading and using this model, you agree to the terms of the LLaMA license [available here](https://github.com/meta - llama/llama - models/blob/main/models/llama3_1/LICENSE).

🔧 Technical Details

No technical details that meet the requirements are provided in the original document, so this section is skipped.

📄 License

Llama[3.1] - Meditron[8B] is licensed under the Llama 3.1 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved. By downloading and using this model, you agree to the terms of the LLaMA license [available here](https://github.com/meta - llama/llama - models/blob/main/models/llama3_1/LICENSE).