Ai2 OLMoE 1.1

The Ai2 OLMoE app emerges as a resourceful tool for researchers and developers focusing on the refinement of on-device intelligence. Its design facilitates the rapid prototyping of AI experiences without necessitating any cloud connectivity.

This app serves multiple purposes:

• Demonstrates the capabilities of cutting-edge local models for real-world tasks.
• Provides a platform for investigating methods to enhance smaller AI models.
• Allows users to test their own models locally, utilizing an open-source codebase.
• Enables integration of OLMoE into various iOS applications.

Privacy is a significant feature, as all interactions with OLMoE remain fully private. Conversations are stored on the device unless the user actively chooses to share them.

Key Features:

• Efficiency: OLMoE utilizes a highly efficient mixture-of-experts architecture from the Ai2 OLMo model family.
• Performance: It is trained using the comprehensive and state-of-the-art Tülu 3 recipe.
• Privacy: The app prioritizes user privacy and security by functioning entirely on-device.
• Shareable: Users can seamlessly share conversational outputs with colleagues or friends.
• Truly open: Both the OLMoE model and application code are fully open-source, promoting transparency in development.

OLMoE is positioned by Ai2 as a robust research tool aimed at enhancing community understanding of large language models (LLMs) and their operational mechanics. Users are encouraged to exercise discernment, verify critical information, and consider context when interpreting the content generated by the app.

For further exploration, the open-source code repository is available at: https://github.com/allenai/OLMoE.swift

Terms of service can be found at: https://allenai.org/terms/2024-09-25

The privacy policy is available at: https://allenai.org/privacy-policy/2022-07-21