ReqLLM.Providers.AmazonBedrock.Meta (ReqLLM v1.0.0)

View Source

Meta Llama model family support for AWS Bedrock.

Handles Meta's Llama models (Llama 3, 3.1, 3.2, 3.3, 4) on AWS Bedrock.

This module acts as a thin adapter between Bedrock's AWS-specific wrapping and Meta's native Llama format. It delegates to ReqLLM.Providers.Meta for all format conversion and response parsing.

Native Format vs OpenAI-Compatible

Unlike most cloud providers (Azure, Google Cloud Vertex AI) and self-hosted deployments (vLLM, Ollama) which wrap Llama in OpenAI-compatible APIs, AWS Bedrock uses Meta's native format with prompt, max_gen_len, and generation fields.

This is why this module delegates to the generic ReqLLM.Providers.Meta rather than ReqLLM.Providers.OpenAI.

Summary

Functions

extract_usage(body, arg2)

Extracts usage metadata from the response body.

format_llama_prompt(messages)

Formats messages into Llama 3 prompt format.

format_request(model_id, context, opts)

Formats a ReqLLM context into Meta Llama request format for Bedrock.

normalize_tool_schema(json_schema)

Normalizes tool schema for Meta Llama models.

parse_response(body, opts)

Parses Meta Llama response from Bedrock into ReqLLM format.

parse_stream_chunk(chunk, opts)

Parses a streaming chunk for Meta Llama models.

supports_converse_tool_choice?()

Returns whether this model family supports toolChoice in Bedrock Converse API.

Functions

extract_usage(body, arg2)

Extracts usage metadata from the response body.

Delegates to the generic Meta provider for usage extraction.

format_llama_prompt(messages)

Formats messages into Llama 3 prompt format.

Delegates to the generic Meta provider. Exposed for testing.

format_request(model_id, context, opts)

Formats a ReqLLM context into Meta Llama request format for Bedrock.

Delegates to the generic Meta provider and returns the formatted request.

normalize_tool_schema(json_schema)

Normalizes tool schema for Meta Llama models.

Meta Llama models on Bedrock have a bug where they return empty content arrays when tool schemas include "additionalProperties": false. This function strips that field recursively from the schema.

See: https://github.com/agentjido/req_llm/issues/XXX

parse_response(body, opts)

Parses Meta Llama response from Bedrock into ReqLLM format.

Delegates to the generic Meta provider for parsing.

parse_stream_chunk(chunk, opts)

Parses a streaming chunk for Meta Llama models.

Each chunk contains a "generation" field with the next text segment. Unwraps Bedrock's AWS Event Stream encoding before processing.

supports_converse_tool_choice?()

Returns whether this model family supports toolChoice in Bedrock Converse API.