13259218501

Committed 11 Feb 2025 09:00AM UTC coverage: 91.459% (-1.3%) from 92.709%

Build # 13259218501

Build Type

Pull #8829

github

Committed by

web-flow

Commit Message

Merge 427e76339 into ad90e106a

Pull Request Pull Request #8829: fix: Look through all streaming chunks for tools calls

Run Details

9413 of 10292 relevant lines covered (91.46%)

0.91 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

96.32

haystack/components/generators/chat/openai.py

# SPDX-FileCopyrightText: 2022-present deepset GmbH <info@deepset.ai>
#
# SPDX-License-Identifier: Apache-2.0

import json
import os
from datetime import datetime
from typing import Any, Callable, Dict, List, Optional, Union

from openai import OpenAI, Stream
from openai.types.chat import ChatCompletion, ChatCompletionChunk, ChatCompletionMessage
from openai.types.chat.chat_completion import Choice
from openai.types.chat.chat_completion_chunk import Choice as ChunkChoice

from haystack import component, default_from_dict, default_to_dict, logging
from haystack.dataclasses import ChatMessage, StreamingChunk, ToolCall
from haystack.tools.tool import Tool, _check_duplicate_tool_names, deserialize_tools_inplace
from haystack.utils import Secret, deserialize_callable, deserialize_secrets_inplace, serialize_callable

logger = logging.getLogger(__name__)


StreamingCallbackT = Callable[[StreamingChunk], None]


@component
class OpenAIChatGenerator:
    """
    Completes chats using OpenAI's large language models (LLMs).

    It works with the gpt-4 and gpt-3.5-turbo models and supports streaming responses
    from OpenAI API. It uses [ChatMessage](https://docs.haystack.deepset.ai/docs/chatmessage)
    format in input and output.

    You can customize how the text is generated by passing parameters to the
    OpenAI API. Use the `**generation_kwargs` argument when you initialize
    the component or when you run it. Any parameter that works with
    `openai.ChatCompletion.create` will work here too.

    For details on OpenAI API parameters, see
    [OpenAI documentation](https://platform.openai.com/docs/api-reference/chat).

    ### Usage example

    ```python
    from haystack.components.generators.chat import OpenAIChatGenerator
    from haystack.dataclasses import ChatMessage

    messages = [ChatMessage.from_user("What's Natural Language Processing?")]

    client = OpenAIChatGenerator()
    response = client.run(messages)
    print(response)
    ```
    Output:
    ```
    {'replies':
        [ChatMessage(content='Natural Language Processing (NLP) is a branch of artificial intelligence
            that focuses on enabling computers to understand, interpret, and generate human language in
            a way that is meaningful and useful.',
         role=<ChatRole.ASSISTANT: 'assistant'>, name=None,
         meta={'model': 'gpt-4o-mini', 'index': 0, 'finish_reason': 'stop',
         'usage': {'prompt_tokens': 15, 'completion_tokens': 36, 'total_tokens': 51}})
        ]
    }
    ```
    """

    def __init__(  # pylint: disable=too-many-positional-arguments
        self,
        api_key: Secret = Secret.from_env_var("OPENAI_API_KEY"),
        model: str = "gpt-4o-mini",
        streaming_callback: Optional[StreamingCallbackT] = None,
        api_base_url: Optional[str] = None,
        organization: Optional[str] = None,
        generation_kwargs: Optional[Dict[str, Any]] = None,
        timeout: Optional[float] = None,
        max_retries: Optional[int] = None,
        tools: Optional[List[Tool]] = None,
        tools_strict: bool = False,
    ):
        """
        Creates an instance of OpenAIChatGenerator. Unless specified otherwise in `model`, uses OpenAI's gpt-4o-mini

        Before initializing the component, you can set the 'OPENAI_TIMEOUT' and 'OPENAI_MAX_RETRIES'
        environment variables to override the `timeout` and `max_retries` parameters respectively
        in the OpenAI client.

        :param api_key: The OpenAI API key.
            You can set it with an environment variable `OPENAI_API_KEY`, or pass with this parameter
            during initialization.
        :param model: The name of the model to use.
        :param streaming_callback: A callback function that is called when a new token is received from the stream.
            The callback function accepts [StreamingChunk](https://docs.haystack.deepset.ai/docs/data-classes#streamingchunk)
            as an argument.
        :param api_base_url: An optional base URL.
        :param organization: Your organization ID, defaults to `None`. See
        [production best practices](https://platform.openai.com/docs/guides/production-best-practices/setting-up-your-organization).
        :param generation_kwargs: Other parameters to use for the model. These parameters are sent directly to
            the OpenAI endpoint. See OpenAI [documentation](https://platform.openai.com/docs/api-reference/chat) for
            more details.
            Some of the supported parameters:
            - `max_tokens`: The maximum number of tokens the output text can have.
            - `temperature`: What sampling temperature to use. Higher values mean the model will take more risks.
                Try 0.9 for more creative applications and 0 (argmax sampling) for ones with a well-defined answer.
            - `top_p`: An alternative to sampling with temperature, called nucleus sampling, where the model
                considers the results of the tokens with top_p probability mass. For example, 0.1 means only the tokens
                comprising the top 10% probability mass are considered.
            - `n`: How many completions to generate for each prompt. For example, if the LLM gets 3 prompts and n is 2,
                it will generate two completions for each of the three prompts, ending up with 6 completions in total.
            - `stop`: One or more sequences after which the LLM should stop generating tokens.
            - `presence_penalty`: What penalty to apply if a token is already present at all. Bigger values mean
                the model will be less likely to repeat the same token in the text.
            - `frequency_penalty`: What penalty to apply if a token has already been generated in the text.
                Bigger values mean the model will be less likely to repeat the same token in the text.
            - `logit_bias`: Add a logit bias to specific tokens. The keys of the dictionary are tokens, and the
                values are the bias to add to that token.
        :param timeout:
            Timeout for OpenAI client calls. If not set, it defaults to either the
            `OPENAI_TIMEOUT` environment variable, or 30 seconds.
        :param max_retries:
            Maximum number of retries to contact OpenAI after an internal error.
            If not set, it defaults to either the `OPENAI_MAX_RETRIES` environment variable, or set to 5.
        :param tools:
            A list of tools for which the model can prepare calls.
        :param tools_strict:
            Whether to enable strict schema adherence for tool calls. If set to `True`, the model will follow exactly
            the schema provided in the `parameters` field of the tool definition, but this may increase latency.
        """
        self.api_key = api_key
        self.model = model
        self.generation_kwargs = generation_kwargs or {}
        self.streaming_callback = streaming_callback
        self.api_base_url = api_base_url
        self.organization = organization
        self.timeout = timeout
        self.max_retries = max_retries
        self.tools = tools
        self.tools_strict = tools_strict

        _check_duplicate_tool_names(tools)

        if timeout is None:
            timeout = float(os.environ.get("OPENAI_TIMEOUT", 30.0))
        if max_retries is None:
            max_retries = int(os.environ.get("OPENAI_MAX_RETRIES", 5))

        self.client = OpenAI(
            api_key=api_key.resolve_value(),
            organization=organization,
            base_url=api_base_url,
            timeout=timeout,
            max_retries=max_retries,
        )

    def _get_telemetry_data(self) -> Dict[str, Any]:
        """
        Data that is sent to Posthog for usage analytics.
        """
        return {"model": self.model}

    def to_dict(self) -> Dict[str, Any]:
        """
        Serialize this component to a dictionary.

        :returns:
            The serialized component as a dictionary.
        """
        callback_name = serialize_callable(self.streaming_callback) if self.streaming_callback else None
        return default_to_dict(
            self,
            model=self.model,
            streaming_callback=callback_name,
            api_base_url=self.api_base_url,
            organization=self.organization,
            generation_kwargs=self.generation_kwargs,
            api_key=self.api_key.to_dict(),
            timeout=self.timeout,
            max_retries=self.max_retries,
            tools=[tool.to_dict() for tool in self.tools] if self.tools else None,
            tools_strict=self.tools_strict,
        )

    @classmethod
    def from_dict(cls, data: Dict[str, Any]) -> "OpenAIChatGenerator":
        """
        Deserialize this component from a dictionary.

        :param data: The dictionary representation of this component.
        :returns:
            The deserialized component instance.
        """
        deserialize_secrets_inplace(data["init_parameters"], keys=["api_key"])
        deserialize_tools_inplace(data["init_parameters"], key="tools")
        init_params = data.get("init_parameters", {})
        serialized_callback_handler = init_params.get("streaming_callback")
        if serialized_callback_handler:
            data["init_parameters"]["streaming_callback"] = deserialize_callable(serialized_callback_handler)
        return default_from_dict(cls, data)

    @component.output_types(replies=List[ChatMessage])
    def run(
        self,
        messages: List[ChatMessage],
        streaming_callback: Optional[StreamingCallbackT] = None,
        generation_kwargs: Optional[Dict[str, Any]] = None,
        *,
        tools: Optional[List[Tool]] = None,
        tools_strict: Optional[bool] = None,
    ):
        """
        Invokes chat completion based on the provided messages and generation parameters.

        :param messages:
            A list of ChatMessage instances representing the input messages.
        :param streaming_callback:
            A callback function that is called when a new token is received from the stream.
        :param generation_kwargs:
            Additional keyword arguments for text generation. These parameters will
            override the parameters passed during component initialization.
            For details on OpenAI API parameters, see [OpenAI documentation](https://platform.openai.com/docs/api-reference/chat/create).
        :param tools:
            A list of tools for which the model can prepare calls. If set, it will override the `tools` parameter set
            during component initialization.
        :param tools_strict:
            Whether to enable strict schema adherence for tool calls. If set to `True`, the model will follow exactly
            the schema provided in the `parameters` field of the tool definition, but this may increase latency.
            If set, it will override the `tools_strict` parameter set during component initialization.

        :returns:
            A dictionary with the following key:
            - `replies`: A list containing the generated responses as ChatMessage instances.
        """
        if len(messages) == 0:
            return {"replies": []}

        streaming_callback = streaming_callback or self.streaming_callback

        api_args = self._prepare_api_call(
            messages=messages,
            streaming_callback=streaming_callback,
            generation_kwargs=generation_kwargs,
            tools=tools,
            tools_strict=tools_strict,
        )
        chat_completion: Union[Stream[ChatCompletionChunk], ChatCompletion] = self.client.chat.completions.create(
            **api_args
        )

        is_streaming = isinstance(chat_completion, Stream)
        assert is_streaming or streaming_callback is None

        if is_streaming:
            completions = self._handle_stream_response(
                chat_completion,  # type: ignore
                streaming_callback,  # type: ignore
            )
        else:
            assert isinstance(chat_completion, ChatCompletion), "Unexpected response type for non-streaming request."
            completions = [
                self._convert_chat_completion_to_chat_message(chat_completion, choice)
                for choice in chat_completion.choices
            ]

        # before returning, do post-processing of the completions
        for message in completions:
            self._check_finish_reason(message.meta)

        return {"replies": completions}

    def _prepare_api_call(  # noqa: PLR0913
        self,
        *,
        messages: List[ChatMessage],
        streaming_callback: Optional[StreamingCallbackT] = None,
        generation_kwargs: Optional[Dict[str, Any]] = None,
        tools: Optional[List[Tool]] = None,
        tools_strict: Optional[bool] = None,
    ) -> Dict[str, Any]:
        # update generation kwargs by merging with the generation kwargs passed to the run method
        generation_kwargs = {**self.generation_kwargs, **(generation_kwargs or {})}

        # adapt ChatMessage(s) to the format expected by the OpenAI API
        openai_formatted_messages = [message.to_openai_dict_format() for message in messages]

        tools = tools or self.tools
        tools_strict = tools_strict if tools_strict is not None else self.tools_strict
        _check_duplicate_tool_names(tools)

        openai_tools = {}
        if tools:
            tool_definitions = [
                {"type": "function", "function": {**t.tool_spec, **({"strict": tools_strict} if tools_strict else {})}}
                for t in tools
            ]
            openai_tools = {"tools": tool_definitions}

        is_streaming = streaming_callback is not None
        num_responses = generation_kwargs.pop("n", 1)
        if is_streaming and num_responses > 1:
            raise ValueError("Cannot stream multiple responses, please set n=1.")

        return {
            "model": self.model,
            "messages": openai_formatted_messages,  # type: ignore[arg-type] # openai expects list of specific message types
            "stream": streaming_callback is not None,
            "n": num_responses,
            **openai_tools,
            **generation_kwargs,
        }

    def _handle_stream_response(self, chat_completion: Stream, callback: StreamingCallbackT) -> List[ChatMessage]:
        chunks: List[StreamingChunk] = []
        chunk = None

        for chunk in chat_completion:  # pylint: disable=not-an-iterable
            assert len(chunk.choices) == 1, "Streaming responses should have only one choice."
            chunk_delta: StreamingChunk = self._convert_chat_completion_chunk_to_streaming_chunk(chunk)
            chunks.append(chunk_delta)

            callback(chunk_delta)

        return [self._convert_streaming_chunks_to_chat_message(chunk, chunks)]

    def _check_finish_reason(self, meta: Dict[str, Any]) -> None:
        if meta["finish_reason"] == "length":
            logger.warning(
                "The completion for index {index} has been truncated before reaching a natural stopping point. "
                "Increase the max_tokens parameter to allow for longer completions.",
                index=meta["index"],
                finish_reason=meta["finish_reason"],
            )
        if meta["finish_reason"] == "content_filter":
            logger.warning(
                "The completion for index {index} has been truncated due to the content filter.",
                index=meta["index"],
                finish_reason=meta["finish_reason"],
            )

    def _convert_streaming_chunks_to_chat_message(self, chunk: Any, chunks: List[StreamingChunk]) -> ChatMessage:
        """
        Connects the streaming chunks into a single ChatMessage.

        :param chunk: The last chunk returned by the OpenAI API.
        :param chunks: The list of all `StreamingChunk` objects.
        """
        text = "".join([chunk.content for chunk in chunks])
        tool_calls = []

        # Process tool calls if present in any chunk
        tool_call_data = {}  # Track tool calls by ID
        for chunk_payload in chunks:
            tool_calls_meta = chunk_payload.meta.get("tool_calls")
            if tool_calls_meta:
                for delta in tool_calls_meta:
                    if not delta.id in tool_call_data:
                        tool_call_data[delta.id] = {"id": delta.id, "name": "", "arguments": ""}

                    if delta.function:
                        if delta.function.name:
                            tool_call_data[delta.id]["name"] = delta.function.name
                        if delta.function.arguments:
                            tool_call_data[delta.id]["arguments"] = delta.function.arguments

        # Convert accumulated tool call data into ToolCall objects
        for call_data in tool_call_data.values():
            try:
                arguments = json.loads(call_data["arguments"])
                tool_calls.append(ToolCall(id=call_data["id"], tool_name=call_data["name"], arguments=arguments))
            except json.JSONDecodeError:
                logger.warning(
                    "Skipping malformed tool call due to invalid JSON. Set `tools_strict=True` for valid JSON. "
                    "Tool call ID: {_id}, Tool name: {_name}, Arguments: {_arguments}",
                    _id=call_data["id"],
                    _name=call_data["name"],
                    _arguments=call_data["arguments"],
                )

        meta = {
            "model": chunk.model,
            "index": 0,
            "finish_reason": chunk.choices[0].finish_reason,
            "completion_start_time": chunks[0].meta.get("received_at"),  # first chunk received
            "usage": {},  # we don't have usage data for streaming responses
        }

        return ChatMessage.from_assistant(text=text, tool_calls=tool_calls, meta=meta)

    def _convert_chat_completion_to_chat_message(self, completion: ChatCompletion, choice: Choice) -> ChatMessage:
        """
        Converts the non-streaming response from the OpenAI API to a ChatMessage.

        :param completion: The completion returned by the OpenAI API.
        :param choice: The choice returned by the OpenAI API.
        :return: The ChatMessage.
        """
        message: ChatCompletionMessage = choice.message
        text = message.content
        tool_calls = []
        if openai_tool_calls := message.tool_calls:
            for openai_tc in openai_tool_calls:
                arguments_str = openai_tc.function.arguments
                try:
                    arguments = json.loads(arguments_str)
                    tool_calls.append(ToolCall(id=openai_tc.id, tool_name=openai_tc.function.name, arguments=arguments))
                except json.JSONDecodeError:
                    logger.warning(
                        "OpenAI returned a malformed JSON string for tool call arguments. This tool call "
                        "will be skipped. To always generate a valid JSON, set `tools_strict` to `True`. "
                        "Tool call ID: {_id}, Tool name: {_name}, Arguments: {_arguments}",
                        _id=openai_tc.id,
                        _name=openai_tc.function.name,
                        _arguments=arguments_str,
                    )

        chat_message = ChatMessage.from_assistant(text=text, tool_calls=tool_calls)
        chat_message._meta.update(
            {
                "model": completion.model,
                "index": choice.index,
                "finish_reason": choice.finish_reason,
                "usage": dict(completion.usage or {}),
            }
        )
        return chat_message

    def _convert_chat_completion_chunk_to_streaming_chunk(self, chunk: ChatCompletionChunk) -> StreamingChunk:
        """
        Converts the streaming response chunk from the OpenAI API to a StreamingChunk.

        :param chunk: The chunk returned by the OpenAI API.

        :returns:
            The StreamingChunk.
        """
        # we stream the content of the chunk if it's not a tool or function call
        choice: ChunkChoice = chunk.choices[0]
        content = choice.delta.content or ""
        chunk_message = StreamingChunk(content)
        # but save the tool calls and function call in the meta if they are present
        # and then connect the chunks in the _convert_streaming_chunks_to_chat_message method
        chunk_message.meta.update(
            {
                "model": chunk.model,
                "index": choice.index,
                "tool_calls": choice.delta.tool_calls,
                "finish_reason": choice.finish_reason,
                "received_at": datetime.now().isoformat(),
            }
        )
        return chunk_message

1	# SPDX-FileCopyrightText: 2022-present deepset GmbH <info@deepset.ai>
2	#
3	# SPDX-License-Identifier: Apache-2.0
4
5	import json	1✔
6	import os	1✔
7	from datetime import datetime	1✔
8	from typing import Any, Callable, Dict, List, Optional, Union	1✔
9
10	from openai import OpenAI, Stream	1✔
11	from openai.types.chat import ChatCompletion, ChatCompletionChunk, ChatCompletionMessage	1✔
12	from openai.types.chat.chat_completion import Choice	1✔
13	from openai.types.chat.chat_completion_chunk import Choice as ChunkChoice	1✔
14
15	from haystack import component, default_from_dict, default_to_dict, logging	1✔
16	from haystack.dataclasses import ChatMessage, StreamingChunk, ToolCall	1✔
17	from haystack.tools.tool import Tool, _check_duplicate_tool_names, deserialize_tools_inplace	1✔
18	from haystack.utils import Secret, deserialize_callable, deserialize_secrets_inplace, serialize_callable	1✔
19
20	logger = logging.getLogger(__name__)	1✔
21
22
23	StreamingCallbackT = Callable[[StreamingChunk], None]	1✔
24
25
26	@component	1✔
27	class OpenAIChatGenerator:	1✔
28	"""
29	Completes chats using OpenAI's large language models (LLMs).
30
31	It works with the gpt-4 and gpt-3.5-turbo models and supports streaming responses
32	from OpenAI API. It uses [ChatMessage](https://docs.haystack.deepset.ai/docs/chatmessage)
33	format in input and output.
34
35	You can customize how the text is generated by passing parameters to the
36	OpenAI API. Use the `**generation_kwargs` argument when you initialize
37	the component or when you run it. Any parameter that works with
38	`openai.ChatCompletion.create` will work here too.
39
40	For details on OpenAI API parameters, see
41	[OpenAI documentation](https://platform.openai.com/docs/api-reference/chat).
42
43	### Usage example
44
45	```python
46	from haystack.components.generators.chat import OpenAIChatGenerator
47	from haystack.dataclasses import ChatMessage
48
49	messages = [ChatMessage.from_user("What's Natural Language Processing?")]
50
51	client = OpenAIChatGenerator()
52	response = client.run(messages)
53	print(response)
54	```
55	Output:
56	```
57	{'replies':
58	[ChatMessage(content='Natural Language Processing (NLP) is a branch of artificial intelligence
59	that focuses on enabling computers to understand, interpret, and generate human language in
60	a way that is meaningful and useful.',
61	role=<ChatRole.ASSISTANT: 'assistant'>, name=None,
62	meta={'model': 'gpt-4o-mini', 'index': 0, 'finish_reason': 'stop',
63	'usage': {'prompt_tokens': 15, 'completion_tokens': 36, 'total_tokens': 51}})
64	]
65	}
66	```
67	"""
68
69	def __init__( # pylint: disable=too-many-positional-arguments	1✔
70	self,
71	api_key: Secret = Secret.from_env_var("OPENAI_API_KEY"),
72	model: str = "gpt-4o-mini",
73	streaming_callback: Optional[StreamingCallbackT] = None,
74	api_base_url: Optional[str] = None,
75	organization: Optional[str] = None,
76	generation_kwargs: Optional[Dict[str, Any]] = None,
77	timeout: Optional[float] = None,
78	max_retries: Optional[int] = None,
79	tools: Optional[List[Tool]] = None,
80	tools_strict: bool = False,
81	):
82	"""
83	Creates an instance of OpenAIChatGenerator. Unless specified otherwise in `model`, uses OpenAI's gpt-4o-mini
84
85	Before initializing the component, you can set the 'OPENAI_TIMEOUT' and 'OPENAI_MAX_RETRIES'
86	environment variables to override the `timeout` and `max_retries` parameters respectively
87	in the OpenAI client.
88
89	:param api_key: The OpenAI API key.
90	You can set it with an environment variable `OPENAI_API_KEY`, or pass with this parameter
91	during initialization.
92	:param model: The name of the model to use.
93	:param streaming_callback: A callback function that is called when a new token is received from the stream.
94	The callback function accepts [StreamingChunk](https://docs.haystack.deepset.ai/docs/data-classes#streamingchunk)
95	as an argument.
96	:param api_base_url: An optional base URL.
97	:param organization: Your organization ID, defaults to `None`. See
98	[production best practices](https://platform.openai.com/docs/guides/production-best-practices/setting-up-your-organization).
99	:param generation_kwargs: Other parameters to use for the model. These parameters are sent directly to
100	the OpenAI endpoint. See OpenAI [documentation](https://platform.openai.com/docs/api-reference/chat) for
101	more details.
102	Some of the supported parameters:
103	- `max_tokens`: The maximum number of tokens the output text can have.
104	- `temperature`: What sampling temperature to use. Higher values mean the model will take more risks.
105	Try 0.9 for more creative applications and 0 (argmax sampling) for ones with a well-defined answer.
106	- `top_p`: An alternative to sampling with temperature, called nucleus sampling, where the model
107	considers the results of the tokens with top_p probability mass. For example, 0.1 means only the tokens
108	comprising the top 10% probability mass are considered.
109	- `n`: How many completions to generate for each prompt. For example, if the LLM gets 3 prompts and n is 2,
110	it will generate two completions for each of the three prompts, ending up with 6 completions in total.
111	- `stop`: One or more sequences after which the LLM should stop generating tokens.
112	- `presence_penalty`: What penalty to apply if a token is already present at all. Bigger values mean
113	the model will be less likely to repeat the same token in the text.
114	- `frequency_penalty`: What penalty to apply if a token has already been generated in the text.
115	Bigger values mean the model will be less likely to repeat the same token in the text.
116	- `logit_bias`: Add a logit bias to specific tokens. The keys of the dictionary are tokens, and the
117	values are the bias to add to that token.
118	:param timeout:
119	Timeout for OpenAI client calls. If not set, it defaults to either the
120	`OPENAI_TIMEOUT` environment variable, or 30 seconds.
121	:param max_retries:
122	Maximum number of retries to contact OpenAI after an internal error.
123	If not set, it defaults to either the `OPENAI_MAX_RETRIES` environment variable, or set to 5.
124	:param tools:
125	A list of tools for which the model can prepare calls.
126	:param tools_strict:
127	Whether to enable strict schema adherence for tool calls. If set to `True`, the model will follow exactly
128	the schema provided in the `parameters` field of the tool definition, but this may increase latency.
129	"""
130	self.api_key = api_key	1✔
131	self.model = model	1✔
132	self.generation_kwargs = generation_kwargs or {}	1✔
133	self.streaming_callback = streaming_callback	1✔
134	self.api_base_url = api_base_url	1✔
135	self.organization = organization	1✔
136	self.timeout = timeout	1✔
137	self.max_retries = max_retries	1✔
138	self.tools = tools	1✔
139	self.tools_strict = tools_strict	1✔
140
141	_check_duplicate_tool_names(tools)	1✔
142
143	if timeout is None:	1✔
144	timeout = float(os.environ.get("OPENAI_TIMEOUT", 30.0))	1✔
145	if max_retries is None:	1✔
146	max_retries = int(os.environ.get("OPENAI_MAX_RETRIES", 5))	1✔
147
148	self.client = OpenAI(	1✔
149	api_key=api_key.resolve_value(),
150	organization=organization,
151	base_url=api_base_url,
152	timeout=timeout,
153	max_retries=max_retries,
154	)
155
156	def _get_telemetry_data(self) -> Dict[str, Any]:	1✔
157	"""
158	Data that is sent to Posthog for usage analytics.
159	"""
160	return {"model": self.model}	×
161
162	def to_dict(self) -> Dict[str, Any]:	1✔
163	"""
164	Serialize this component to a dictionary.
165
166	:returns:
167	The serialized component as a dictionary.
168	"""
169	callback_name = serialize_callable(self.streaming_callback) if self.streaming_callback else None	1✔
170	return default_to_dict(	1✔
171	self,
172	model=self.model,
173	streaming_callback=callback_name,
174	api_base_url=self.api_base_url,
175	organization=self.organization,
176	generation_kwargs=self.generation_kwargs,
177	api_key=self.api_key.to_dict(),
178	timeout=self.timeout,
179	max_retries=self.max_retries,
180	tools=[tool.to_dict() for tool in self.tools] if self.tools else None,
181	tools_strict=self.tools_strict,
182	)
183
184	@classmethod	1✔
185	def from_dict(cls, data: Dict[str, Any]) -> "OpenAIChatGenerator":	1✔
186	"""
187	Deserialize this component from a dictionary.
188
189	:param data: The dictionary representation of this component.
190	:returns:
191	The deserialized component instance.
192	"""
193	deserialize_secrets_inplace(data["init_parameters"], keys=["api_key"])	1✔
194	deserialize_tools_inplace(data["init_parameters"], key="tools")	1✔
195	init_params = data.get("init_parameters", {})	1✔
196	serialized_callback_handler = init_params.get("streaming_callback")	1✔
197	if serialized_callback_handler:	1✔
198	data["init_parameters"]["streaming_callback"] = deserialize_callable(serialized_callback_handler)	1✔
199	return default_from_dict(cls, data)	1✔
200
201	@component.output_types(replies=List[ChatMessage])	1✔
202	def run(	1✔
203	self,
204	messages: List[ChatMessage],
205	streaming_callback: Optional[StreamingCallbackT] = None,
206	generation_kwargs: Optional[Dict[str, Any]] = None,
207	*,
208	tools: Optional[List[Tool]] = None,
209	tools_strict: Optional[bool] = None,
210	):
211	"""
212	Invokes chat completion based on the provided messages and generation parameters.
213
214	:param messages:
215	A list of ChatMessage instances representing the input messages.
216	:param streaming_callback:
217	A callback function that is called when a new token is received from the stream.
218	:param generation_kwargs:
219	Additional keyword arguments for text generation. These parameters will
220	override the parameters passed during component initialization.
221	For details on OpenAI API parameters, see [OpenAI documentation](https://platform.openai.com/docs/api-reference/chat/create).
222	:param tools:
223	A list of tools for which the model can prepare calls. If set, it will override the `tools` parameter set
224	during component initialization.
225	:param tools_strict:
226	Whether to enable strict schema adherence for tool calls. If set to `True`, the model will follow exactly
227	the schema provided in the `parameters` field of the tool definition, but this may increase latency.
228	If set, it will override the `tools_strict` parameter set during component initialization.
229
230	:returns:
231	A dictionary with the following key:
232	- `replies`: A list containing the generated responses as ChatMessage instances.
233	"""
234	if len(messages) == 0:	1✔
235	return {"replies": []}	×
236
237	streaming_callback = streaming_callback or self.streaming_callback	1✔
238
239	api_args = self._prepare_api_call(	1✔
240	messages=messages,
241	streaming_callback=streaming_callback,
242	generation_kwargs=generation_kwargs,
243	tools=tools,
244	tools_strict=tools_strict,
245	)
246	chat_completion: Union[Stream[ChatCompletionChunk], ChatCompletion] = self.client.chat.completions.create(	1✔
247	**api_args
248	)
249
250	is_streaming = isinstance(chat_completion, Stream)	1✔
251	assert is_streaming or streaming_callback is None	1✔
252
253	if is_streaming:	1✔
254	completions = self._handle_stream_response(	1✔
255	chat_completion, # type: ignore
256	streaming_callback, # type: ignore
257	)
258	else:
259	assert isinstance(chat_completion, ChatCompletion), "Unexpected response type for non-streaming request."	1✔
260	completions = [	1✔
261	self._convert_chat_completion_to_chat_message(chat_completion, choice)
262	for choice in chat_completion.choices
263	]
264
265	# before returning, do post-processing of the completions
266	for message in completions:	1✔
267	self._check_finish_reason(message.meta)	1✔
268
269	return {"replies": completions}	1✔
270
271	def _prepare_api_call( # noqa: PLR0913	1✔
272	self,
273	*,
274	messages: List[ChatMessage],
275	streaming_callback: Optional[StreamingCallbackT] = None,
276	generation_kwargs: Optional[Dict[str, Any]] = None,
277	tools: Optional[List[Tool]] = None,
278	tools_strict: Optional[bool] = None,
279	) -> Dict[str, Any]:
280	# update generation kwargs by merging with the generation kwargs passed to the run method
281	generation_kwargs = {self.generation_kwargs, (generation_kwargs or {})}	1✔
282
283	# adapt ChatMessage(s) to the format expected by the OpenAI API
284	openai_formatted_messages = [message.to_openai_dict_format() for message in messages]	1✔
285
286	tools = tools or self.tools	1✔
287	tools_strict = tools_strict if tools_strict is not None else self.tools_strict	1✔
288	_check_duplicate_tool_names(tools)	1✔
289
290	openai_tools = {}	1✔
291	if tools:	1✔
292	tool_definitions = [	1✔
293	{"type": "function", "function": {t.tool_spec, ({"strict": tools_strict} if tools_strict else {})}}
294	for t in tools
295	]
296	openai_tools = {"tools": tool_definitions}	1✔
297
298	is_streaming = streaming_callback is not None	1✔
299	num_responses = generation_kwargs.pop("n", 1)	1✔
300	if is_streaming and num_responses > 1:	1✔
301	raise ValueError("Cannot stream multiple responses, please set n=1.")	×
302
303	return {	1✔
304	"model": self.model,
305	"messages": openai_formatted_messages, # type: ignore[arg-type] # openai expects list of specific message types
306	"stream": streaming_callback is not None,
307	"n": num_responses,
308	**openai_tools,
309	**generation_kwargs,
310	}
311
312	def _handle_stream_response(self, chat_completion: Stream, callback: StreamingCallbackT) -> List[ChatMessage]:	1✔
313	chunks: List[StreamingChunk] = []	1✔
314	chunk = None	1✔
315
316	for chunk in chat_completion: # pylint: disable=not-an-iterable	1✔
317	assert len(chunk.choices) == 1, "Streaming responses should have only one choice."	1✔
318	chunk_delta: StreamingChunk = self._convert_chat_completion_chunk_to_streaming_chunk(chunk)	1✔
319	chunks.append(chunk_delta)	1✔
320
321	callback(chunk_delta)	1✔
322
323	return [self._convert_streaming_chunks_to_chat_message(chunk, chunks)]	1✔
324
325	def _check_finish_reason(self, meta: Dict[str, Any]) -> None:	1✔
326	if meta["finish_reason"] == "length":	1✔
327	logger.warning(	1✔
328	"The completion for index {index} has been truncated before reaching a natural stopping point. "
329	"Increase the max_tokens parameter to allow for longer completions.",
330	index=meta["index"],
331	finish_reason=meta["finish_reason"],
332	)
333	if meta["finish_reason"] == "content_filter":	1✔
334	logger.warning(	1✔
335	"The completion for index {index} has been truncated due to the content filter.",
336	index=meta["index"],
337	finish_reason=meta["finish_reason"],
338	)
339
340	def _convert_streaming_chunks_to_chat_message(self, chunk: Any, chunks: List[StreamingChunk]) -> ChatMessage:	1✔
341	"""
342	Connects the streaming chunks into a single ChatMessage.
343
344	:param chunk: The last chunk returned by the OpenAI API.
345	:param chunks: The list of all `StreamingChunk` objects.
346	"""
347	text = "".join([chunk.content for chunk in chunks])	1✔
348	tool_calls = []	1✔
349
350	# Process tool calls if present in any chunk
351	tool_call_data = {} # Track tool calls by ID	1✔
352	for chunk_payload in chunks:	1✔
353	tool_calls_meta = chunk_payload.meta.get("tool_calls")	1✔
354	if tool_calls_meta:	1✔
355	for delta in tool_calls_meta:	1✔
356	if not delta.id in tool_call_data:	1✔
357	tool_call_data[delta.id] = {"id": delta.id, "name": "", "arguments": ""}	1✔
358
359	if delta.function:	1✔
360	if delta.function.name:	1✔
361	tool_call_data[delta.id]["name"] = delta.function.name	1✔
362	if delta.function.arguments:	1✔
363	tool_call_data[delta.id]["arguments"] = delta.function.arguments	1✔
364
365	# Convert accumulated tool call data into ToolCall objects
366	for call_data in tool_call_data.values():	1✔
367	try:	1✔
368	arguments = json.loads(call_data["arguments"])	1✔
369	tool_calls.append(ToolCall(id=call_data["id"], tool_name=call_data["name"], arguments=arguments))	1✔
370	except json.JSONDecodeError:	×
371	logger.warning(	×
372	"Skipping malformed tool call due to invalid JSON. Set `tools_strict=True` for valid JSON. "
373	"Tool call ID: {_id}, Tool name: {_name}, Arguments: {_arguments}",
374	_id=call_data["id"],
375	_name=call_data["name"],
376	_arguments=call_data["arguments"],
377	)
378
379	meta = {	1✔
380	"model": chunk.model,
381	"index": 0,
382	"finish_reason": chunk.choices[0].finish_reason,
383	"completion_start_time": chunks[0].meta.get("received_at"), # first chunk received
384	"usage": {}, # we don't have usage data for streaming responses
385	}
386
387	return ChatMessage.from_assistant(text=text, tool_calls=tool_calls, meta=meta)	1✔
388
389	def _convert_chat_completion_to_chat_message(self, completion: ChatCompletion, choice: Choice) -> ChatMessage:	1✔
390	"""
391	Converts the non-streaming response from the OpenAI API to a ChatMessage.
392
393	:param completion: The completion returned by the OpenAI API.
394	:param choice: The choice returned by the OpenAI API.
395	:return: The ChatMessage.
396	"""
397	message: ChatCompletionMessage = choice.message	1✔
398	text = message.content	1✔
399	tool_calls = []	1✔
400	if openai_tool_calls := message.tool_calls:	1✔
401	for openai_tc in openai_tool_calls:	1✔
402	arguments_str = openai_tc.function.arguments	1✔
403	try:	1✔
404	arguments = json.loads(arguments_str)	1✔
405	tool_calls.append(ToolCall(id=openai_tc.id, tool_name=openai_tc.function.name, arguments=arguments))	1✔
406	except json.JSONDecodeError:	1✔
407	logger.warning(	1✔
408	"OpenAI returned a malformed JSON string for tool call arguments. This tool call "
409	"will be skipped. To always generate a valid JSON, set `tools_strict` to `True`. "
410	"Tool call ID: {_id}, Tool name: {_name}, Arguments: {_arguments}",
411	_id=openai_tc.id,
412	_name=openai_tc.function.name,
413	_arguments=arguments_str,
414	)
415
416	chat_message = ChatMessage.from_assistant(text=text, tool_calls=tool_calls)	1✔
417	chat_message._meta.update(	1✔
418	{
419	"model": completion.model,
420	"index": choice.index,
421	"finish_reason": choice.finish_reason,
422	"usage": dict(completion.usage or {}),
423	}
424	)
425	return chat_message	1✔
426
427	def _convert_chat_completion_chunk_to_streaming_chunk(self, chunk: ChatCompletionChunk) -> StreamingChunk:	1✔
428	"""
429	Converts the streaming response chunk from the OpenAI API to a StreamingChunk.
430
431	:param chunk: The chunk returned by the OpenAI API.
432
433	:returns:
434	The StreamingChunk.
435	"""
436	# we stream the content of the chunk if it's not a tool or function call
437	choice: ChunkChoice = chunk.choices[0]	1✔
438	content = choice.delta.content or ""	1✔
439	chunk_message = StreamingChunk(content)	1✔
440	# but save the tool calls and function call in the meta if they are present
441	# and then connect the chunks in the _convert_streaming_chunks_to_chat_message method
442	chunk_message.meta.update(	1✔
443	{
444	"model": chunk.model,
445	"index": choice.index,
446	"tool_calls": choice.delta.tool_calls,
447	"finish_reason": choice.finish_reason,
448	"received_at": datetime.now().isoformat(),
449	}
450	)
451	return chunk_message	1✔

deepset-ai / haystack / 13259218501

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous