Open dickren123 opened 2 weeks ago
Thanks for this idea! We will be working on it and keep you posted. To benchmark a custom model, we would simply have to write a predictor (https://aws-samples.github.io/foundation-model-benchmarking-tool/byo_rest_predictor.html) that supports the prediction format for custom model imports (nothing too complex, the function would be similar to what is given in https://aws.amazon.com/cn/blogs/machine-learning/amazon-bedrock-custom-model-import-now-generally-available/, using the converse API.
Now the custom import model is GAed, we need to evaluate the cost performance using FMBench with customized model arn.