- Newest
- Most votes
- Most comments
Did you try this example? https://python.langchain.com/v0.2/docs/integrations/llms/bedrock/
# Guardrails for Amazon Bedrock with trace
llm = BedrockLLM(
credentials_profile_name="bedrock-admin",
model_id="<Model_ID>",
model_kwargs={},
guardrails={"id": "<Guardrail_ID>", "version": "<Version>", "trace": True},
callbacks=[BedrockAsyncCallbackHandler()],
You can enable the trace and view what exactly happened
Since LangChain is 3rd party and not AWS tool, it may be hard to address this challenge. If you don't have any special reason to use LangChain, I suggest you to check the new unified AWS Converse API, which also supports Guardrails
https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails-use-converse-api.html
I’m building a RAG application, so LangChain is required. In the system prompt, I pass the user’s query and need to implement guardrails for the user's query. The guardrails are effective for denied topics and word filters but are not addressing prompt injection. I have also tried using input tags.
https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails-tagging.html
Maybe you want to check Amazon Bedrock Knowledge Bases. It is very easy implementation for RAG with a friendly API that also supports guardrails and preserves the context. https://aws.amazon.com/bedrock/knowledge-bases/ https://docs.aws.amazon.com/bedrock/latest/userguide/kb-test-query.html
Relevant content
- AWS OFFICIALUpdated 10 months ago
- AWS OFFICIALUpdated 3 months ago
Hi, I tried using ChatBedrock instead of BedrockLLM since BedrockLLM doesn't support Claude v3 models. However, prompt injection is still possible even after providing my guardrail details. I also tried using input tags like <amazon-bedrock-guardrails-guardContent_xyz>, but it remains prone to prompt injection.