DeciLM 6B is a permissively licensed foundation LLM. DeciLM 6B-Instruct is its intruction-finetuned version. It provides 15 times higher throughput than Llama 2 7B. It has 95% lower hardware cost. It saves 516kg CO2 carbon footprint per model instance yearly on A10G compared to Llama 2 7B. But provides comparable accuracies as evaluated on several benchmark datasets.
In this video, I will talk about the following: What is DeciLM? How does DeciLM perform compared to Llama 2?
For more details, please look at https://deci.ai/blog/decilm-15-times-...