🦦 Otter, a multi-modal model based on OpenFlamingo (open-sourced version of DeepMind's Flamingo), trained on MIMIC-IT and showcasing improved instruction-following and in-context learning ability.
For the evaluation of classification tasks, such as ImageNet1k, I would like to know if using kv_cathe for evaluation would have any impact on the results? My personal understanding is that if no_kv_caching=False, then the model's evaluation process is more akin to a generation process, whereas if it's True, the model's evaluation process is more similar to the training process. I'm not sure if my understanding is correct; thank you for your response.
For the evaluation of classification tasks, such as ImageNet1k, I would like to know if using kv_cathe for evaluation would have any impact on the results? My personal understanding is that if no_kv_caching=False, then the model's evaluation process is more akin to a generation process, whereas if it's True, the model's evaluation process is more similar to the training process. I'm not sure if my understanding is correct; thank you for your response.