Together Llama 2
About
The Together Llama 2 family of large language models extends the capabilities of Meta's Llama 2, maintaining its open-source nature and being available for research and commercial purposes 14. These models feature a noteworthy advancement in context length, allowing for 32K tokens compared to Llama 2's initial 4K, enabling more complex tasks like multi-document question answering and long-form text summarization 45. Among the family's innovations is the Llama-2-7B-32K-Instruct, tailored through high-quality instruction and conversational data for improved performance in dialogue-based interactions 15. Efficiency in training and inference is enhanced through optimizations such as FlashAttention-2, helping the models operate faster and more resource-efficiently 4. By providing training recipes and datasets, Together AI supports broader community engagement and progression in the open-source LLM domain 5.