OpenHands/openhands/llm/llm.py

import copy
import os
import time
import warnings
from functools import partial
from typing import Any

import requests

from openhands.core.config import LLMConfig

with warnings.catch_warnings():
    warnings.simplefilter('ignore')
    import litellm

from litellm import Message as LiteLLMMessage
from litellm import ModelInfo, PromptTokensDetails
from litellm import completion as litellm_completion
from litellm import completion_cost as litellm_completion_cost
from litellm.exceptions import (
    APIConnectionError,
    APIError,
    InternalServerError,
    RateLimitError,
    ServiceUnavailableError,
)
from litellm.types.utils import CostPerToken, ModelResponse, Usage
from litellm.utils import create_pretrained_tokenizer

from openhands.core.exceptions import CloudFlareBlockageError
from openhands.core.logger import openhands_logger as logger
from openhands.core.message import Message
from openhands.llm.debug_mixin import DebugMixin
from openhands.llm.fn_call_converter import (
    STOP_WORDS,
    convert_fncall_messages_to_non_fncall_messages,
    convert_non_fncall_messages_to_fncall_messages,
)
from openhands.llm.metrics import Metrics
from openhands.llm.retry_mixin import RetryMixin

__all__ = ['LLM']

# tuple of exceptions to retry on
LLM_RETRY_EXCEPTIONS: tuple[type[Exception], ...] = (
    APIConnectionError,
    # FIXME: APIError is useful on 502 from a proxy for example,
    # but it also retries on other errors that are permanent
    APIError,
    InternalServerError,
    RateLimitError,
    ServiceUnavailableError,
)

# cache prompt supporting models
# remove this when we gemini and deepseek are supported
CACHE_PROMPT_SUPPORTED_MODELS = [
    'claude-3-5-sonnet-20241022',
    'claude-3-5-sonnet-20240620',
    'claude-3-5-haiku-20241022',
    'claude-3-haiku-20240307',
    'claude-3-opus-20240229',
]

# function calling supporting models
FUNCTION_CALLING_SUPPORTED_MODELS = [
    'claude-3-5-sonnet',
    'claude-3-5-sonnet-20240620',
    'claude-3-5-sonnet-20241022',
    'claude-3.5-haiku',
    'claude-3-5-haiku-20241022',
    'gpt-4o-mini',
    'gpt-4o',
]


class LLM(RetryMixin, DebugMixin):
    """The LLM class represents a Language Model instance.

    Attributes:
        config: an LLMConfig object specifying the configuration of the LLM.
    """

    def __init__(
        self,
        config: LLMConfig,
        metrics: Metrics | None = None,
    ):
        """Initializes the LLM. If LLMConfig is passed, its values will be the fallback.

        Passing simple parameters always overrides config.

        Args:
            config: The LLM configuration.
            metrics: The metrics to use.
        """
        self._tried_model_info = False
        self.metrics: Metrics = (
            metrics if metrics is not None else Metrics(model_name=config.model)
        )
        self.cost_metric_supported: bool = True
        self.config: LLMConfig = copy.deepcopy(config)

        # litellm actually uses base Exception here for unknown model
        self.model_info: ModelInfo | None = None

        if self.config.log_completions:
            if self.config.log_completions_folder is None:
                raise RuntimeError(
                    'log_completions_folder is required when log_completions is enabled'
                )
            os.makedirs(self.config.log_completions_folder, exist_ok=True)

        # call init_model_info to initialize config.max_output_tokens
        # which is used in partial function
        with warnings.catch_warnings():
            warnings.simplefilter('ignore')
            self.init_model_info()
        if self.vision_is_active():
            logger.debug('LLM: model has vision enabled')
        if self.is_caching_prompt_active():
            logger.debug('LLM: caching prompt enabled')
        if self.is_function_calling_active():
            logger.debug('LLM: model supports function calling')

        # if using a custom tokenizer, make sure it's loaded and accessible in the format expected by litellm
        if self.config.custom_tokenizer is not None:
            self.tokenizer = create_pretrained_tokenizer(self.config.custom_tokenizer)
        else:
            self.tokenizer = None

        # set up the completion function
        self._completion = partial(
            litellm_completion,
            model=self.config.model,
            api_key=self.config.api_key,
            base_url=self.config.base_url,
            api_version=self.config.api_version,
            custom_llm_provider=self.config.custom_llm_provider,
            max_tokens=self.config.max_output_tokens,
            timeout=self.config.timeout,
            temperature=self.config.temperature,
            top_p=self.config.top_p,
            drop_params=self.config.drop_params,
        )

        self._completion_unwrapped = self._completion

        @self.retry_decorator(
            num_retries=self.config.num_retries,
            retry_exceptions=LLM_RETRY_EXCEPTIONS,
            retry_min_wait=self.config.retry_min_wait,
            retry_max_wait=self.config.retry_max_wait,
            retry_multiplier=self.config.retry_multiplier,
        )
        def wrapper(*args, **kwargs):
            """Wrapper for the litellm completion function. Logs the input and output of the completion function."""
            from openhands.core.utils import json

            messages: list[dict[str, Any]] | dict[str, Any] = []
            mock_function_calling = kwargs.pop('mock_function_calling', False)

            # some callers might send the model and messages directly
            # litellm allows positional args, like completion(model, messages, **kwargs)
            if len(args) > 1:
                # ignore the first argument if it's provided (it would be the model)
                # design wise: we don't allow overriding the configured values
                # implementation wise: the partial function set the model as a kwarg already
                # as well as other kwargs
                messages = args[1] if len(args) > 1 else args[0]
                kwargs['messages'] = messages

                # remove the first args, they're sent in kwargs
                args = args[2:]
            elif 'messages' in kwargs:
                messages = kwargs['messages']

            # ensure we work with a list of messages
            messages = messages if isinstance(messages, list) else [messages]
            original_fncall_messages = copy.deepcopy(messages)
            mock_fncall_tools = None
            if mock_function_calling:
                assert (
                    'tools' in kwargs
                ), "'tools' must be in kwargs when mock_function_calling is True"
                messages = convert_fncall_messages_to_non_fncall_messages(
                    messages, kwargs['tools']
                )
                kwargs['messages'] = messages
                kwargs['stop'] = STOP_WORDS
                mock_fncall_tools = kwargs.pop('tools')

            # if we have no messages, something went very wrong
            if not messages:
                raise ValueError(
                    'The messages list is empty. At least one message is required.'
                )

            # log the entire LLM prompt
            self.log_prompt(messages)

            if self.is_caching_prompt_active():
                # Anthropic-specific prompt caching
                if 'claude-3' in self.config.model:
                    kwargs['extra_headers'] = {
                        'anthropic-beta': 'prompt-caching-2024-07-31',
                    }

            try:
                # Record start time for latency measurement
                start_time = time.time()

                # we don't support streaming here, thus we get a ModelResponse
                resp: ModelResponse = self._completion_unwrapped(*args, **kwargs)

                # Calculate and record latency
                latency = time.time() - start_time
                response_id = resp.get('id', 'unknown')
                self.metrics.add_response_latency(latency, response_id)

                non_fncall_response = copy.deepcopy(resp)
                if mock_function_calling:
                    assert len(resp.choices) == 1
                    assert mock_fncall_tools is not None
                    non_fncall_response_message = resp.choices[0].message
                    fn_call_messages_with_response = (
                        convert_non_fncall_messages_to_fncall_messages(
                            messages + [non_fncall_response_message], mock_fncall_tools
                        )
                    )
                    fn_call_response_message = fn_call_messages_with_response[-1]
                    if not isinstance(fn_call_response_message, LiteLLMMessage):
                        fn_call_response_message = LiteLLMMessage(
                            **fn_call_response_message
                        )
                    resp.choices[0].message = fn_call_response_message

                message_back: str = resp['choices'][0]['message']['content'] or ''
                tool_calls = resp['choices'][0]['message'].get('tool_calls', [])
                if tool_calls:
                    for tool_call in tool_calls:
                        fn_name = tool_call.function.name
                        fn_args = tool_call.function.arguments
                        message_back += f'\nFunction call: {fn_name}({fn_args})'

                # log the LLM response
                self.log_response(message_back)

                # post-process the response first to calculate cost
                cost = self._post_completion(resp)

                # log for evals or other scripts that need the raw completion
                if self.config.log_completions:
                    assert self.config.log_completions_folder is not None
                    log_file = os.path.join(
                        self.config.log_completions_folder,
                        # use the metric model name (for draft editor)
                        f'{self.metrics.model_name.replace("/", "__")}-{time.time()}.json',
                    )

                    # set up the dict to be logged
                    _d = {
                        'messages': messages,
                        'response': resp,
                        'args': args,
                        'kwargs': {k: v for k, v in kwargs.items() if k != 'messages'},
                        'timestamp': time.time(),
                        'cost': cost,
                    }

                    # if non-native function calling, save messages/response separately
                    if mock_function_calling:
                        # Overwrite response as non-fncall to be consistent with messages
                        _d['response'] = non_fncall_response

                        # Save fncall_messages/response separately
                        _d['fncall_messages'] = original_fncall_messages
                        _d['fncall_response'] = resp
                    with open(log_file, 'w') as f:
                        f.write(json.dumps(_d))

                return resp
            except APIError as e:
                if 'Attention Required! | Cloudflare' in str(e):
                    raise CloudFlareBlockageError(
                        'Request blocked by CloudFlare'
                    ) from e
                raise

        self._completion = wrapper

    @property
    def completion(self):
        """Decorator for the litellm completion function.

        Check the complete documentation at https://litellm.vercel.app/docs/completion
        """
        return self._completion

    def init_model_info(self):
        if self._tried_model_info:
            return
        self._tried_model_info = True
        try:
            if self.config.model.startswith('openrouter'):
                self.model_info = litellm.get_model_info(self.config.model)
        except Exception as e:
            logger.debug(f'Error getting model info: {e}')

        if self.config.model.startswith('litellm_proxy/'):
            # IF we are using LiteLLM proxy, get model info from LiteLLM proxy
            # GET {base_url}/v1/model/info with litellm_model_id as path param
            response = requests.get(
                f'{self.config.base_url}/v1/model/info',
                headers={'Authorization': f'Bearer {self.config.api_key}'},
            )
            resp_json = response.json()
            if 'data' not in resp_json:
                logger.error(
                    f'Error getting model info from LiteLLM proxy: {resp_json}'
                )
            all_model_info = resp_json.get('data', [])
            current_model_info = next(
                (
                    info
                    for info in all_model_info
                    if info['model_name']
                    == self.config.model.removeprefix('litellm_proxy/')
                ),
                None,
            )
            if current_model_info:
                self.model_info = current_model_info['model_info']

        # Last two attempts to get model info from NAME
        if not self.model_info:
            try:
                self.model_info = litellm.get_model_info(
                    self.config.model.split(':')[0]
                )
            # noinspection PyBroadException
            except Exception:
                pass
        if not self.model_info:
            try:
                self.model_info = litellm.get_model_info(
                    self.config.model.split('/')[-1]
                )
            # noinspection PyBroadException
            except Exception:
                pass
        logger.debug(f'Model info: {self.model_info}')

        if self.config.model.startswith('huggingface'):
            # HF doesn't support the OpenAI default value for top_p (1)
            logger.debug(
                f'Setting top_p to 0.9 for Hugging Face model: {self.config.model}'
            )
            self.config.top_p = 0.9 if self.config.top_p == 1 else self.config.top_p

        # Set the max tokens in an LM-specific way if not set
        if self.config.max_input_tokens is None:
            if (
                self.model_info is not None
                and 'max_input_tokens' in self.model_info
                and isinstance(self.model_info['max_input_tokens'], int)
            ):
                self.config.max_input_tokens = self.model_info['max_input_tokens']
            else:
                # Safe fallback for any potentially viable model
                self.config.max_input_tokens = 4096

        if self.config.max_output_tokens is None:
            # Safe default for any potentially viable model
            self.config.max_output_tokens = 4096
            if self.model_info is not None:
                # max_output_tokens has precedence over max_tokens, if either exists.
                # litellm has models with both, one or none of these 2 parameters!
                if 'max_output_tokens' in self.model_info and isinstance(
                    self.model_info['max_output_tokens'], int
                ):
                    self.config.max_output_tokens = self.model_info['max_output_tokens']
                elif 'max_tokens' in self.model_info and isinstance(
                    self.model_info['max_tokens'], int
                ):
                    self.config.max_output_tokens = self.model_info['max_tokens']

    def vision_is_active(self) -> bool:
        with warnings.catch_warnings():
            warnings.simplefilter('ignore')
            return not self.config.disable_vision and self._supports_vision()

    def _supports_vision(self) -> bool:
        """Acquire from litellm if model is vision capable.

        Returns:
            bool: True if model is vision capable. Return False if model not supported by litellm.
        """
        # litellm.supports_vision currently returns False for 'openai/gpt-...' or 'anthropic/claude-...' (with prefixes)
        # but model_info will have the correct value for some reason.
        # we can go with it, but we will need to keep an eye if model_info is correct for Vertex or other providers
        # remove when litellm is updated to fix https://github.com/BerriAI/litellm/issues/5608
        # Check both the full model name and the name after proxy prefix for vision support
        return (
            litellm.supports_vision(self.config.model)
            or litellm.supports_vision(self.config.model.split('/')[-1])
            or (
                self.model_info is not None
                and self.model_info.get('supports_vision', False)
            )
        )

    def is_caching_prompt_active(self) -> bool:
        """Check if prompt caching is supported and enabled for current model.

        Returns:
            boolean: True if prompt caching is supported and enabled for the given model.
        """
        return (
            self.config.caching_prompt is True
            and (
                self.config.model in CACHE_PROMPT_SUPPORTED_MODELS
                or self.config.model.split('/')[-1] in CACHE_PROMPT_SUPPORTED_MODELS
            )
            # We don't need to look-up model_info, because only Anthropic models needs the explicit caching breakpoint
        )

    def is_function_calling_active(self) -> bool:
        # Check if model name is in supported list before checking model_info
        model_name_supported = (
            self.config.model in FUNCTION_CALLING_SUPPORTED_MODELS
            or self.config.model.split('/')[-1] in FUNCTION_CALLING_SUPPORTED_MODELS
            or any(m in self.config.model for m in FUNCTION_CALLING_SUPPORTED_MODELS)
        )
        return model_name_supported

    def _post_completion(self, response: ModelResponse) -> float:
        """Post-process the completion response.

        Logs the cost and usage stats of the completion call.
        """
        try:
            cur_cost = self._completion_cost(response)
        except Exception:
            cur_cost = 0

        stats = ''
        if self.cost_metric_supported:
            # keep track of the cost
            stats = 'Cost: %.2f USD | Accumulated Cost: %.2f USD\n' % (
                cur_cost,
                self.metrics.accumulated_cost,
            )

        # Add latency to stats if available
        if self.metrics.response_latencies:
            latest_latency = self.metrics.response_latencies[-1]
            stats += 'Response Latency: %.3f seconds\n' % latest_latency.latency

        usage: Usage | None = response.get('usage')

        if usage:
            # keep track of the input and output tokens
            input_tokens = usage.get('prompt_tokens')
            output_tokens = usage.get('completion_tokens')

            if input_tokens:
                stats += 'Input tokens: ' + str(input_tokens)

            if output_tokens:
                stats += (
                    (' | ' if input_tokens else '')
                    + 'Output tokens: '
                    + str(output_tokens)
                    + '\n'
                )

            # read the prompt cache hit, if any
            prompt_tokens_details: PromptTokensDetails = usage.get(
                'prompt_tokens_details'
            )
            cache_hit_tokens = (
                prompt_tokens_details.cached_tokens if prompt_tokens_details else None
            )
            if cache_hit_tokens:
                stats += 'Input tokens (cache hit): ' + str(cache_hit_tokens) + '\n'

            # For Anthropic, the cache writes have a different cost than regular input tokens
            # but litellm doesn't separate them in the usage stats
            # so we can read it from the provider-specific extra field
            model_extra = usage.get('model_extra', {})
            cache_write_tokens = model_extra.get('cache_creation_input_tokens')
            if cache_write_tokens:
                stats += 'Input tokens (cache write): ' + str(cache_write_tokens) + '\n'

        # log the stats
        if stats:
            logger.debug(stats)

        return cur_cost

    def get_token_count(self, messages: list[dict] | list[Message]) -> int:
        """Get the number of tokens in a list of messages. Use dicts for better token counting.

        Args:
            messages (list): A list of messages, either as a list of dicts or as a list of Message objects.
        Returns:
            int: The number of tokens.
        """
        # attempt to convert Message objects to dicts, litellm expects dicts
        if (
            isinstance(messages, list)
            and len(messages) > 0
            and isinstance(messages[0], Message)
        ):
            logger.info(
                'Message objects now include serialized tool calls in token counting'
            )
            messages = self.format_messages_for_llm(messages)  # type: ignore

        # try to get the token count with the default litellm tokenizers
        # or the custom tokenizer if set for this LLM configuration
        try:
            return litellm.token_counter(
                model=self.config.model,
                messages=messages,
                custom_tokenizer=self.tokenizer,
            )
        except Exception as e:
            # limit logspam in case token count is not supported
            logger.error(
                f'Error getting token count for\n model {self.config.model}\n{e}'
                + (
                    f'\ncustom_tokenizer: {self.config.custom_tokenizer}'
                    if self.config.custom_tokenizer is not None
                    else ''
                )
            )
            return 0

    def _is_local(self) -> bool:
        """Determines if the system is using a locally running LLM.

        Returns:
            boolean: True if executing a local model.
        """
        if self.config.base_url is not None:
            for substring in ['localhost', '127.0.0.1' '0.0.0.0']:
                if substring in self.config.base_url:
                    return True
        elif self.config.model is not None:
            if self.config.model.startswith('ollama'):
                return True
        return False

    def _completion_cost(self, response) -> float:
        """Calculate the cost of a completion response based on the model.  Local models are treated as free.
        Add the current cost into total cost in metrics.

        Args:
            response: A response from a model invocation.

        Returns:
            number: The cost of the response.
        """
        if not self.cost_metric_supported:
            return 0.0

        extra_kwargs = {}
        if (
            self.config.input_cost_per_token is not None
            and self.config.output_cost_per_token is not None
        ):
            cost_per_token = CostPerToken(
                input_cost_per_token=self.config.input_cost_per_token,
                output_cost_per_token=self.config.output_cost_per_token,
            )
            logger.debug(f'Using custom cost per token: {cost_per_token}')
            extra_kwargs['custom_cost_per_token'] = cost_per_token

        try:
            # try directly get response_cost from response
            cost = getattr(response, '_hidden_params', {}).get('response_cost', None)
            if cost is None:
                cost = litellm_completion_cost(
                    completion_response=response, **extra_kwargs
                )
            self.metrics.add_cost(cost)
            return cost
        except Exception:
            self.cost_metric_supported = False
            logger.debug('Cost calculation not supported for this model.')
        return 0.0

    def __str__(self):
        if self.config.api_version:
            return f'LLM(model={self.config.model}, api_version={self.config.api_version}, base_url={self.config.base_url})'
        elif self.config.base_url:
            return f'LLM(model={self.config.model}, base_url={self.config.base_url})'
        return f'LLM(model={self.config.model})'

    def __repr__(self):
        return str(self)

    def reset(self) -> None:
        self.metrics.reset()

    def format_messages_for_llm(self, messages: Message | list[Message]) -> list[dict]:
        if isinstance(messages, Message):
            messages = [messages]

        # set flags to know how to serialize the messages
        for message in messages:
            message.cache_enabled = self.is_caching_prompt_active()
            message.vision_enabled = self.vision_is_active()
            message.function_calling_enabled = self.is_function_calling_active()

        # let pydantic handle the serialization
        return [message.model_dump() for message in messages]