Meditron, a suite of open-source large multimodal foundation models tailored for the medical field, leverages Meta's Llama 2 and 3 to assist with clinical decision-making and diagnosis in low-resource settings. Developed by researchers at EPFL and Yale School of Medicine in collaboration with humanitarian organizations like the International Committee of the Red Cross, Meditron provides evidence-based, contextually aware recommendations and diagnostic support. It automates tasks such as generating treatment regimens, determining radiation modalities, and assigning diagnostic codes, thereby enhancing clinical efficiency and improving patient outcomes. With over 30,000 downloads and top performance on medical benchmarks, Meditron democratizes access to advanced medical AI tools, enabling equitable healthcare innovation in underserved regions.

The Barcelona Supercomputing Center has developed Aloe, a family of fine-tuned open healthcare LLMs built on Meta Llama 3. Aloe features advanced training and inference mechanisms, achieving over 10 accuracy points improvement and setting new standards for ethical performance in healthcare AI through policy alignment and Direct Preference Optimization, leading to state-of-the-art results for open healthcare 7B LLMs.

Melooha developed a hyper-personalized Vedic astrological engine using Meta's Llama 3 70B model. The model enhances the accuracy and depth of astrological insights by blending ancient wisdom with modern AI, allowing for profound exploration of astrological complexities and delivering credible, personalized predictions.

ClearPoint integrated Meta's Llama 3 with Google's code scanner lint to perform an automated refactor of over 500 code files in a large Kotlin codebase. With less than 500 lines of automation code, they produced a migration that runs with a single build command without human intervention, saving an estimated 11 engineering weeks and improving efficiency without risking IP or vendor lock-in.

Elyza, a Japanese AI startup, uses Meta Llama 2 to develop a Japanese large language model designed for advanced natural language processing applications, including language translation, chatbots, and text generation, thereby enhancing AI-driven communication and language services in Japan.

Tune AI is a fine-tuning and deployment platform assisting large enterprises. For an information services enterprise, Tune AI selected Meta's Llama 3 to index a massive 7B+ page digital library, reducing costs from $18 to $0.60 per document and processing time from 1 hour to 7 minutes, achieving a 99% accuracy rate, significantly enhancing efficiency and cost-effectiveness.

KissanAI has released Dhenu Llama 3, a multilingual AI agriculture assistant tailored for Indian farmers. Built on Meta’s Llama 3 8B architecture, Dhenu understands voice and text inputs and is integrated with platforms like WhatsApp, supporting 22 languages including 9 Indian languages. This assists farmers in accessing agricultural advice and information, enhancing farming practices through advanced AI capabilities.

Taiwan LLM is a pioneering large language model focused on Traditional Chinese used in Taiwan. Built on Meta Llama 2, it incorporates cultural context and advanced pre-training on comprehensive datasets, enabling it to understand linguistic nuances and cultural references, thus revolutionizing Traditional Chinese language processing in NLP applications.

Odia Generative AI is an open-source project aimed at expanding LLM capabilities for the Odia language, spoken by 40 million in India. Using Meta Llama 2, OdiaGenAI has developed the Odia Llama, a fine-tuned LLM for Odia, and is exploring applications like AI chatbots and AI tutors, thereby enhancing accessibility and digital inclusion for Odia speakers.

Researchers at UC Berkeley have introduced RAFT (Retrieval-Augmented Fine-Tuning) by leveraging Meta Llama 2 on Azure AI Studio. RAFT enhances domain adaptation in language models by improving their ability to retrieve and integrate relevant information. This novel approach benefits specialized applications by making Meta Llama 2 more versatile and adaptable to various domain-specific tasks.

ML Commons has integrated Meta's Llama 2 70B parameter model into version 4.0 of its MLPerf Inference benchmark. The published results demonstrate the performance potential of various platforms for running one of the most demanding and capable large language models, highlighting the efficiency and scalability of Meta’s Llama 2 in high-performance computing environments.

Persado's Motivation AI platform utilizes Meta's Llama models, fine-tuned with proprietary data, to generate emotion-informed marketing messages that are 96% more effective in driving action. The models produce effective and on-brand multilingual marketing communications, enhancing marketing effectiveness and customer engagement.

KPMG leveraged Meta's Llama to help the wholesale credit team at a major U.S. bank explore secure, open-source LLMs, enabling the client to review and interrogate complex loan applications more quickly and efficiently. This tailored GenAI solution improves the speed and accuracy of document review and processing, enhancing automation to the next level.

Pratham developed 'BaalSakhi', a chatbot powered by Meta's Llama 2 (70B), integrating via WhatsApp to provide personalized, on-demand support for early childhood care and education to parents and caregivers in resource-poor settings. BaalSakhi draws insights from trusted sources like government departments, UNICEF, WHO, and childcare experts, helping users navigate their child's early development journey independently.

Upstage fine-tuned Meta's Llama 2 to create Solar 70B, a large language model that secured the top position on the Open LLM Leaderboard upon its debut, showcasing the potential of open innovation in AI.

Netsafe, a nonprofit organization, uses Meta Llama models to develop a robust redaction tool for harmful digital communications. The tool employs a tuned LLM and the AI4Privacy dataset to automatically redact sensitive information, facilitating faster and more effective harm resolution. Netsafe's data pipeline, built on Dagster, Postgres, and BigQuery, sources, redacts, and aggregates data from various channels, reducing victim impact and augmenting the skills of Digital Harm Resolution Officers.

Clerkbridge, a North Carolina-based startup, uses Meta Llama to anonymize attorney-privileged data and streamline legal document generation. Additionally, Clerkbridge leverages Llama to monitor and track terms of service, privacy policies, costs, and data policies, alerting users to changes and recommending new models and applications based on usage data, thereby enhancing compliance and operational efficiency.

Mathpresso's QANDA platform utilizes Meta's Llama 2 to create MathGPT, a math-specific large language model that enables highly personalized learning experiences. By leveraging open-source models, Mathpresso offers flexible, domain-specific educational products that democratize access to quality education, allowing learners worldwide to receive tailored math support.

NoHarm.ai, a healthcare non-profit startup in Brazil, uses Meta Llama 2 and an open-source web application to improve hospital discharge summaries for patient care transitions. The NoHarm Discharge Summary tool extracts key information from inpatient records using Named Entity Recognition (NER) and generates discharge summaries in Portuguese for physicians to validate, enhancing data organization and communication within the healthcare system.