feat: AmazonBedrockChatGenerator - add tools support

[vblagoje]

Why:

Adds Haystack tool support to AmazonBedrockChatGenerator

• fixes Bedrock ChatGenerator - support for Tool haystack#8732

What:

• Added method to convert Haystack tools to Amazon Bedrock toolConfig format, facilitating tool integration.
• Modified AmazonBedrockChatGenerator to accept, serialize, and handle tool specifications as part of the chat messages.
• Updated testing suite to cover tool-related functionalities, ensuring robustness and reliability.

How can it be used:

The changes enable users to pass tools to AmazonBedrockChatGenerator, enhancing the model's capabilities by allowing it to use external tools within chat interactions. Here is a simplified example:

from haystack.dataclasses import ChatMessage
from haystack_integrations.components.generators.amazon_bedrock import AmazonBedrockChatGenerator
from haystack.tools import Tool

def weather(city: str):
        """Get weather for a given city."""
        return f"The weather in {city} is sunny and 32°C"

tool_parameters = {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]}
tool = Tool(
    name="weather",
    description="useful to determine the weather in a given location",
    parameters=tool_parameters,
    function=weather,
)

messages = [ChatMessage.from_user("What's the weather like in Paris?")]
generator = AmazonBedrockChatGenerator(model="model_name", tools=[tool])
response = generator.run(messages=messages)

How did you test it:

Implemented thorough unit and integration tests to validate the addition of tool-related functionalities. Tests included:

• Serialization and deserialization of tool configurations.
• Ensuring tool calls within conversations behave as expected.
• Integration tests with mock and live scenarios for tool usage within chat interactions.

Notes for the reviewer:

Please pay special attention to the serialization and deserialization processes for tool objects and the handling of tool calls within the chat messages. Consider reviewing the edge cases in the test suite that handle different orderings of tool messages in replies.

[anakin87]

I have some questions and small suggestions.

In general, this PR looks good.

[anakin87]

I would transform this into a simpler unit test about serde (without running the pipeline).
Pipeline running with tools is already tested in test_pipeline_with_amazon_bedrock_chat_generator.

[anakin87]

please simplify this test

[vblagoje]

Ah missing the pydoc class tooling example. I'll add it @anakin87

[vblagoje]

@anakin87 lmk what we should do about ids or perhaps we can do that in separate PR? I'm ok with both

[anakin87]

At this point, let's do nothing about ID and wait for @lambda-science's opinion in the issue.

Please take this comment into account: #1304 (comment)
Then this PR is ready to be merged.

[vblagoje]

At this point, let's do nothing about ID and wait for @lambda-science's opinion in the issue.

Please take this comment into account: #1304 (comment) Then this PR is ready to be merged.

Right - I forgot about that one 🙏
Will correct now

[vblagoje]

Ok should be gtg now @anakin87 See 91f2330

[anakin87]

last change request (hopefully 😊)

[anakin87]

Suggested change

	@pytest.mark.parametrize("model_name", [MODELS_TO_TEST_WITH_TOOLS[0]]) # just one model is enough
	@pytest.mark.integration

• let's transform this into a unit test (with mocking?)
• if possible, let's test the dictionary format more extensively
• let's also dump to YAML and reload from YAML (some serialization errors happen in this phase)

An example of what I mean:

haystack-core-integrations/integrations/google_ai/tests/generators/chat/test_chat_gemini.py

Line 258 in 68ec202

def test_serde_in_pipeline(self, monkeypatch):

[vblagoje]

Ok, I agree, how about now? 🤞

[anakin87]

LGTM!