Setting truncate to False does not guarantee AmazonBedrockGenerator will not truncate model output?

[tlam-lyra]

Apologies if this is more of a question for haystack-core-integrations repo instead of the main repo, but when I initialized AmazonBedrockGenerator like so

AmazonBedrockGenerator(
    "anthropic.claude-3-sonnet-20240229-v1:0",
    truncate=False,
)

I expected the prompt and the generated text to be not truncated. However, that did not turn out to be the case for some of the prompts. Is this the right assumption and there is something else in my LLM pipeline that is causing this truncation issue or am I missing additional keyword arguments to init AmazonBedrockGenerator?

[julian-risch]

Hello @tlam-lyra was your prompt maybe longer than API's maximum number of characters it allows? Or the model's maximum number of tokens it can process? We had a similar question recently answered by @anakin87 here: deepset-ai/haystack-core-integrations#912 (comment)