A Bedrock playground, contributed by @vaibhavkr993630-droid.
Bedrock
Bedrock now has its own view, with three tabs:
- Playground: pick a text model, write a prompt (and a system prompt if you like), set max tokens, temperature and top P, and send it through the Converse API. You get the reply, the stop reason, token usage and latency. (#184, part of #134)
- Models: the text foundation models the endpoint offers, with their providers and modalities.
- Custom models: the models trained or imported in the account.
It works with any endpoint that serves Bedrock, including AWS Emulator, and with real AWS.
On real AWS, each prompt runs on your account and is billed per token. The playground says so next to the button. Max tokens starts at 512, and Converse calls are never retried, so a slow answer isn't billed twice. Like Lambda invokes, prompts stay allowed in read-only mode, since they don't change any resource.
When an endpoint doesn't serve Bedrock, the view says so. Other errors (credentials, TLS, an unreachable endpoint) show the same explanations as the rest of the console.
Thanks @vaibhavkr993630-droid for building it and for working through two rounds of review!
Full changelog: v0.4.11...v0.4.12