Update README.md

README.md

@@ -1,202 +1,261 @@
 ---
-base_model: togethercomputer/Llama-4-Scout-17B-16E-Instruct_bnb_4bit
-library_name: peft
+license: llama3.3
+base_model: meta-llama/Llama-3.3-70B-Instruct
+language:
+- en
+- fr
+- et
+library_name: transformers
+pipeline_tag: text-generation
+tags:
+- legal
+- law
+- multilingual
+- english
+- french
+- estonian
+- legal-reasoning
+- lora
+- instruction-tuning
+- llama-3
+- adaption-labs
+datasets:
+- RayNene/Trilex
 ---
 
-# Model Card for Model ID
+# TriLex
 
-<!-- Provide a quick summary of what the model is/does. -->
+> **Understanding law across borders, languages, and legal traditions.**
 
+*A multilingual legal reasoning model adapted from Llama 3.3 70B Instruct for English, French, and Estonian legal understanding.*
 
+---
 
-## Model Details
-
-### Model Description
-
-<!-- Provide a longer summary of what this model is. -->
-
-
-
-- **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]
+# Why TriLex?
 
-<!-- Provide the basic links for the model. -->
+Law has never been more global.
 
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
+A company incorporated in London may negotiate contracts governed by French law. An AI researcher may compare judicial systems across Europe. Citizens increasingly seek legal information online before consulting legal professionals. Governments themselves frequently study how other jurisdictions address similar policy challenges.
 
-## Uses
+Yet most legal language models remain confined to a single legal system or language.
 
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+TriLex was built to move beyond that limitation.
 
-### Direct Use
+Instead of specializing in only one jurisdiction, TriLex has been adapted to understand three distinct legal traditions:
 
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+- 🇬🇧 **England** — Common Law built through judicial precedent
+- 🇫🇷 **France** — Civil Law founded on comprehensive legal codification
+- 🇪🇪 **Estonia** — A digitally enabled legal system recognized for modern public administration
 
-[More Information Needed]
+By exposing the model to multiple legal frameworks, TriLex aims to produce responses that recognize not only legal terminology, but also the institutional reasoning and policy objectives behind different legal systems.
 
-### Downstream Use [optional]
+---
 
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+# The Problem
 
-[More Information Needed]
+Legal information is often difficult to access.
 
-### Out-of-Scope Use
+For many people, legislation is written in technical language, court structures differ from one country to another, and understanding how laws interact across jurisdictions requires years of formal education.
 
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+Meanwhile, AI is rapidly becoming one of the first places people turn when trying to understand legal concepts.
 
-[More Information Needed]
+That creates an important responsibility.
 
-## Bias, Risks, and Limitations
+Rather than generating generic legal responses, TriLex was adapted to produce explanations that are clearer, more contextual, and more sensitive to differences between legal systems.
 
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+The goal is **not** to replace lawyers.
 
-[More Information Needed]
+The goal is to make legal knowledge easier to understand.
 
-### Recommendations
+---
 
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+# Model Overview
 
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+**Base Model**
 
-## How to Get Started with the Model
+> Meta Llama 3.3 70B Instruct
 
-Use the code below to get started with the model.
+**Adaptation Method**
 
-[More Information Needed]
+- LoRA fine-tuning
+- Adaptive Data pipeline by **Adaption Labs**
 
-## Training Details
+**Languages**
 
-### Training Data
+- English
+- French
+- Estonian
 
-<!-- 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]
+# Areas of Expertise
 
-### Training Procedure
+TriLex has been adapted for legal reasoning tasks including:
 
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+- Comparative law
+- Constitutional law
+- Administrative law
+- Civil law
+- Public policy
+- Government structures
+- Judicial systems
+- Legal interpretation
+- Legal education
+- Legal question answering
+- Cross-jurisdiction legal comparisons
 
-#### Preprocessing [optional]
+---
 
-[More Information Needed]
+# Training Dataset
 
+TriLex was adapted using a multilingual legal instruction dataset containing examples across three legal systems.
 
-#### Training Hyperparameters
+The dataset includes:
 
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+- English legal reasoning
+- French legal reasoning
+- Estonian legal reasoning
+- Comparative legal analysis
+- Enhanced instruction-response pairs
+- Semantically enriched legal examples
 
-#### Speeds, Sizes, Times [optional]
+By training on multiple jurisdictions simultaneously, the model learns to distinguish differences in legal traditions while preserving legal context.
 
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+---
 
-[More Information Needed]
+# How to Use
 
-## Evaluation
+This repository contains **LoRA adapter weights**.
 
-<!-- This section describes the evaluation protocols and provides the results. -->
+Load the base model first, then apply the adapter.
 
-### Testing Data, Factors & Metrics
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel
 
-#### Testing Data
+BASE_MODEL = "meta-llama/Llama-3.3-70B-Instruct"
+ADAPTER = "RayNene/TriLex"
 
-<!-- This should link to a Dataset Card if possible. -->
+tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
 
-[More Information Needed]
+base_model = AutoModelForCausalLM.from_pretrained(
+    BASE_MODEL,
+    device_map="auto"
+)
 
-#### Factors
+model = PeftModel.from_pretrained(
+    base_model,
+    ADAPTER
+)
+```
 
-<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+---
 
-[More Information Needed]
+# Prompting Tips
 
-#### Metrics
+TriLex performs best when legal questions include jurisdictional context.
 
-<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+### Examples
 
-[More Information Needed]
+```
+Explain the differences between English common law and the French civil law system.
+```
 
-### Results
+```
+How does Estonia's digital identity framework support its legal and public administration systems?
+```
 
-[More Information Needed]
+```
+Compare how administrative disputes are handled in France and England.
+```
 
-#### Summary
+```
+Why does England rely heavily on judicial precedent while France relies on codified legislation?
+```
 
+```
+How do tribunal systems improve judicial efficiency in England?
+```
 
+---
 
-## Model Examination [optional]
+# Intended Applications
 
-<!-- Relevant interpretability work for the model goes here -->
+TriLex is designed for:
 
-[More Information Needed]
+- Legal education
+- Comparative law research
+- Government policy analysis
+- Legal assistants
+- Academic research
+- Multilingual legal retrieval
+- Domain-adapted language model research
 
-## Environmental Impact
+---
 
-<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+# Limitations
 
-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).
+TriLex is intended for research and educational purposes.
 
-- **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]
+It does **not** provide legal advice and should never replace consultation with qualified legal professionals. Laws change over time, judicial interpretations evolve, and legal outcomes depend heavily on jurisdiction and individual circumstances.
 
-## Technical Specifications [optional]
+---
 
-### Model Architecture and Objective
+# Built with Adaption Labs
 
-[More Information Needed]
+TriLex was developed as part of the **Adaption Labs AutoScientist Challenge**.
 
-### Compute Infrastructure
+The project leveraged **Adaptive Data** to enhance and refine multilingual legal instruction data before model adaptation. Adaptive Data made it possible to improve instruction quality, enrich legal reasoning examples, and generate a stronger foundation for domain-specific fine-tuning.
 
-[More Information Needed]
+Special thanks to **Adaption Labs** for making advanced domain adaptation workflows more accessible to the open-source AI community.
 
-#### Hardware
+---
 
-[More Information Needed]
+# Vision
 
-#### Software
+The future of legal AI should not stop at answering questions—it should help people understand *why* legal systems differ.
 
-[More Information Needed]
+Every legal system reflects the history, institutions, and values of the society it serves. By bringing together English common law, French civil law, and Estonia's digital-first governance model, TriLex aims to encourage more transparent, multilingual, and internationally aware legal AI.
 
-## Citation [optional]
+By releasing both the dataset and model openly, this project hopes to contribute to research, education, and the broader effort to make legal knowledge more accessible worldwide.
 
-<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+---
 
-**BibTeX:**
+## Citation
 
-[More Information Needed]
+If you use TriLex in research or downstream applications, please cite this repository and the accompanying dataset.
+If you use **TriLex** in your research, applications, or derivative work, please cite this repository.
 
-**APA:**
+### BibTeX
 
-[More Information Needed]
+```bibtex
+@misc{Ray Munene,
+  title        = {TriLex: A Multilingual Legal Reasoning Model for English, French, and Estonian Law},
+  author       = {Munene, Ray},
+  year         = {2026},
+  publisher    = {Hugging Face},
+  howpublished = {\url{https://huggingface.co/RayNene/TriLex}},
+  note         = {LoRA adaptation of Meta Llama 3.3 70B Instruct developed as part of the Adaption Labs AutoScientist Challenge}
+}
+```
 
-## Glossary [optional]
+### APA
 
-<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+Munene, R. (2026). *TriLex: A Multilingual Legal Reasoning Model for English, French, and Estonian Law* [Computer software]. Hugging Face. https://huggingface.co/RayNene/TriLex
 
-[More Information Needed]
+### Dataset
 
-## More Information [optional]
+The accompanying multilingual legal dataset is available at:
 
-[More Information Needed]
+https://huggingface.co/datasets/RayNene/TriLex
 
-## Model Card Authors [optional]
+If you use the training data, please cite both the dataset and the model.
 
-[More Information Needed]
+---
 
-## Model Card Contact
+## Contact
 
-[More Information Needed]
-### Framework versions
+Questions, suggestions, or contributions are welcome.
 
-- PEFT 0.15.1
+- 🤗 Hugging Face: https://huggingface.co/RayNene
+- GitHub: https://github.com/RayMune