The organization continued fine-tuning work on their large language models, producing final chat models through multiple rounds of alignment. They used synthetic data generation to produce the majority of their supervised fine-tuning examples and invested in data processing techniques to maintain high-quality training datasets. The team also conducted recurring red teaming exercises to discover risks via adversarial prompting and used their learnings to improve benchmark measurements.
AIMeta | Rating: 57 | 2024-07-23 05:39:36 PM |
Meta is releasing Llama 3.1, the world's largest and most capable openly available foundation model, with over 300 million total downloads. The new model, Meta Llama 3.1 405B, is multilingual and has a longer context length of 128K, state-of-the-art tool use, and stronger reasoning capabilities. This enables advanced use cases such as long-form text summarization.
Duplicated with: 1 | 2024-07-23 05:39:36 PM |