Update README.md

README.md

@@ -1,199 +1,154 @@
 ---
 library_name: transformers
-tags: []
+tags:
+- clinical
+- healthcare
+- NLP
+- BERT
+- MIMIC-IV
+- MedNLI
+- transformer
+language:
+- en
+metrics:
+- accuracy
+base_model:
+- mosaicml/mosaic-bert-base
+model-index:
+- name: ClinicalMosaic
+  results:
+  - task:
+      type: classification
+    dataset:
+      name: MedNLI
+      type: MedNLI
+    metrics:
+    - name: Accuracy
+      type: accuracy
+      value: 86.5
+license: mit
+datasets:
+- bigbio/mednli
+pipeline_tag: fill-mask
 ---
 
-# Model Card for Model ID
-
-<!-- Provide a quick summary of what the model is/does. -->
-
+# Model Card for Clinical Mosaic
 
+Clinical Mosaic is a transformer-based language model built on the Mosaic BERT architecture. It has been optimized for clinical language understanding by incorporating recent innovations such as Attention with Linear Biases (ALiBi) for long‐sequence extrapolation and Gated Linear Units (GLU) to capture complex clinical patterns. The model is pre-trained on 331,794 deidentified clinical notes from the MIMIC-IV-NOTES 2.2 database and demonstrates strong performance on clinical reasoning tasks.
 
 ## Model Details
-
-### Model Description
-
-<!-- Provide a longer summary of what this model is. -->
-
-This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
-
-- **Developed by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Model type:** [More Information Needed]
-- **Language(s) (NLP):** [More Information Needed]
-- **License:** [More Information Needed]
-- **Finetuned from model [optional]:** [More Information Needed]
-
-### Model Sources [optional]
-
-<!-- Provide the basic links for the model. -->
-
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
+- **Developed by:** Sifal Klioui, Sana Sellami, and Youssef Trardi (Aix-Marseille Univ, LIS, CNRS, Marseille, France)
+- **Funded by:** PICOMALE project (AMIDEX)
+- **Base Model:** Mosaic BERT
+- **License:** MIMIC Data Use Agreement (requires compliance with original DUA)
+- **Repository:** [PatientTrajectoryForecasting](https://github.com/MostHumble/PatientTrajectoryForecasting)
+- **Paper:** *Patient Trajectory Prediction: Integrating Clinical Notes with Transformers* ([PDF](insert-link))
 
 ## Uses
 
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
-
 ### Direct Use
 
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
-
-[More Information Needed]
+Clinical Mosaic can be used directly as a clinical language model for:
+- Natural language inference and clinical reasoning tasks.
+- Serving as a backbone for further fine-tuning on clinical NLP applications (e.g., clinical note summarization, diagnosis classification).
 
-### Downstream Use [optional]
+### Downstream Use
 
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
-
-[More Information Needed]
+The model can be integrated into larger systems for tasks such as patient trajectory prediction or decision support, provided that additional fine-tuning and rigorous validation are performed.
 
 ### Out-of-Scope Use
 
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
-
-[More Information Needed]
+- **Clinical Decision-Making:** The model is for research use only and should not be deployed for direct patient care without further validation and regulatory approval.
+- **Medical Advice Generation:** It is not intended to replace expert clinical judgment.
 
 ## Bias, Risks, and Limitations
 
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
-
-[More Information Needed]
+Clinical Mosaic was pre-trained on deidentified clinical notes from MIMIC-IV-NOTES 2.2—a dataset from a single U.S. institution. This may introduce biases related to local clinical practices and patient demographics. Although extensive care was taken to deidentify data and prevent PHI leakage, users must ensure that the model is not used to inadvertently reidentify sensitive information. When applying the model to new populations or clinical settings, performance may vary, so further fine-tuning and bias audits are recommended.
 
 ### Recommendations
+- **Text Normalization Consistency:** Any input text must undergo identical preprocessing as applied during training
+- **Evaluation:** Users should evaluate the model’s performance on their target population.
+- **Fine-Tuning:** Consider additional fine-tuning or domain adaptation if deploying in a different clinical context.
+- **Bias Audits:** Regularly perform bias audits to monitor for potential disparities in performance.
 
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+## How to Get Started with the Model
 
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+Install the Hugging Face Transformers library and load the model as follows:
 
-## How to Get Started with the Model
+```python
+from transformers import AutoTokenizer, AutoModel
 
-Use the code below to get started with the model.
+tokenizer = AutoTokenizer.from_pretrained("path/to/clinical-mosaic")
+model = AutoModel.from_pretrained("path/to/clinical-mosaic")
+```
 
-[More Information Needed]
+Further instructions and example scripts are provided in the model’s repository.
 
 ## Training Details
 
 ### Training Data
 
-<!-- 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. -->
-
-[More Information Needed]
+- **Data Source:** 331,794 clinical notes from the MIMIC-IV-NOTES 2.2 dataset.
+- **Preprocessing:** 
+  - Following (Alsentzer et al., 2019), clinical notes were preprocessed by unifying medical abbreviations (e.g., “hr”, “hrs” to “hours”), removing accents, converting special characters, and normalizing text to lowercase. These steps help mitigate variations caused by subword tokenizers.
+  - For additional details, please refer to the [PatientTrajectoryForecasting GitHub repository](https://github.com/MostHumble/PatientTrajectoryForecasting/).
 
 ### Training Procedure
 
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
-
-#### Preprocessing [optional]
-
-[More Information Needed]
-
-
-#### Training Hyperparameters
-
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
-
-#### Speeds, Sizes, Times [optional]
-
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
-
-[More Information Needed]
+- **Setup:** Distributed training using 7 NVIDIA A40 GPUs.
+- **Hyperparameters:**
+  - **Effective Batch Size:** 224
+  - **Training Steps:** 80,000
+  - **Sequence Length:** 512 tokens
+  - **Optimizer:** ADAMW
+  - **Initial Learning Rate:** 5e-4
+  - **Learning Rate Schedule:** Linear warmup for 33,000 steps, followed by cosine annealing for 46,000 steps (final LR 1e-5)
+  - **Masking Probability:** 30%
 
 ## 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]
+The model was evaluated on the MedNLI dataset, which comprises 14,049 clinical premise-hypothesis pairs derived from MIMIC-III notes.
 
 #### 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]
+Evaluation disaggregated performance across clinical language understanding, with a focus on natural language inference.
 
 ### Results
 
-[More Information Needed]
-
-#### Summary
-
-
+Clinical Mosaic outperformed comparable models:
+- **BERT:** 77.6%
+- **BioBERT:** 80.8%
+- **Clinical Discharge BERT:** 84.1%
+- **Bio+Clinical BERT:** 82.7%
+- **Clinical Mosaic:** **86.5%**
 
-## Model Examination [optional]
+These results indicate improved clinical reasoning and language understanding.
 
-<!-- Relevant interpretability work for the model goes here -->
+#### Summary
 
-[More Information Needed]
+The model demonstrates robust performance on clinical natural language inference tasks and serves as a strong foundation for further clinical NLP applications.
 
 ## 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
+- **Hardware Type:** 7 NVIDIA A40 GPUs
+- **Hours used:** 1008 (GPU hours)
+- **Provider:** Private Infrastructure
+- **Carbon Emitted:** 108.86 kg CO2 eq.
 
-[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. -->
+## Citation
 
 **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]
+To be added
 
-## Model Card Authors [optional]
+## More Information
 
-[More Information Needed]
+For further details, please refer to the model’s repository and supplementary documentation.
 
 ## Model Card Contact
 
-[More Information Needed]
+For questions or further information, please contact [[email protected]].