Top Results for Mosaicml
MPT-30B is a 30-billion-parameter open-source large language model released by MosaicML in June 2023. It was trained on a massive dataset consisting of 1 trillion tokens and utilized the FlashAttention technique to optimize training and inference efficiency. The model features an 8,000-token context...
Why this score
Solid open 30B with permissive license and long context; later models surpassed performance.
Scoring methodology
MPT-7B is a 7-billion-parameter transformer model developed by MosaicML and released in May 2023. Trained on 1 trillion tokens of text and code, the model was released under the Apache 2.0 license, allowing for commercial use. It was designed to demonstrate that capable open-source large language mo...
Why this score
Popular permissive 7B baseline; useful for builders, weaker than Mistral and Llama successors.
Scoring methodology
