Feature refactor llm function package

council/controllers/llm_controller.py

@@ -4,7 +4,7 @@
 from council.contexts import AgentContext, ChatMessage, ContextBase
 from council.controllers import ControllerBase, ControllerException
 from council.llm import LLMBase, LLMMessage, MonitoredLLM
-from council.llm.llm_answer import LLMAnswer, LLMParsingException, llm_class_validator, llm_property
+from council.llm.base.llm_answer import LLMAnswer, LLMParsingException, llm_class_validator, llm_property
 from council.utils import Option
 from typing_extensions import TypeGuard
 

council/evaluators/llm_evaluator.py

@@ -9,7 +9,7 @@
 from council.contexts import AgentContext, ChatMessage, ContextBase, ScoredChatMessage
 from council.evaluators import EvaluatorBase, EvaluatorException
 from council.llm import LLMAnswer, LLMBase, LLMMessage, MonitoredLLM, llm_property
-from council.llm.llm_answer import LLMParsingException, llm_class_validator
+from council.llm.base.llm_answer import LLMParsingException, llm_class_validator
 from council.utils import Option
 
 

council/filters/llm_filter.py

@@ -9,7 +9,7 @@
 from council.contexts import AgentContext, ContextBase, ScoredChatMessage
 from council.filters import FilterBase, FilterException
 from council.llm import LLMAnswer, LLMBase, LLMMessage, MonitoredLLM, llm_property
-from council.llm.llm_answer import LLMParsingException
+from council.llm.base.llm_answer import LLMParsingException
 from council.utils import Option
 
 

council/llm/__init__.py

@@ -1,99 +1,72 @@
 """This package provides clients to use various LLMs."""
 
-from typing import Optional, Type
-
-from ..utils import read_env_str
-
-from .llm_config_object import LLMProvider, LLMConfigObject, LLMConfigSpec, LLMProviders
-from .llm_answer import llm_property, LLMAnswer, LLMProperty, LLMParsingException
-from .llm_exception import LLMException, LLMCallException, LLMCallTimeoutException, LLMTokenLimitException
-from .llm_message import LLMMessageRole, LLMMessage, LLMMessageTokenCounterBase
-from .llm_base import LLMBase, LLMResult, LLMConfigurationBase
-from .llm_cost import (
-    LLMCostCard,
-    LLMConsumptionCalculatorBase,
+from .base import (
+    AnthropicLLM,
+    AnthropicLLMConfiguration,
+    AzureChatGPTConfiguration,
+    AzureLLM,
     DefaultLLMConsumptionCalculator,
-    DefaultLLMConsumptionCalculatorHelper,
-    TokenKind,
-    LLMCostManagerSpec,
+    GeminiLLM,
+    GeminiLLMConfiguration,
+    GroqLLM,
+    GroqLLMConfiguration,
+    LLMAnswer,
+    LLMBase,
+    LLMCacheControlData,
+    LLMCallException,
+    LLMCallTimeoutException,
+    LLMConfigObject,
+    LLMConfigSpec,
+    LLMConfigurationBase,
+    LLMConsumptionCalculatorBase,
+    LLMCostCard,
     LLMCostManagerObject,
+    LLMException,
+    LLMFallback,
+    LLMMessage,
+    LLMMessageData,
+    LLMMessageRole,
+    LLMMessageTokenCounterBase,
+    LLMOutOfRetriesException,
+    LLMParsingException,
+    LLMProperty,
+    LLMProvider,
+    LLMProviders,
+    LLMResult,
+    LLMTokenLimitException,
+    MonitoredLLM,
+    OllamaLLM,
+    OllamaLLMConfiguration,
+    OpenAIChatGPTConfiguration,
+    OpenAILLM,
+    TokenKind,
+    get_default_llm,
+    get_llm_from_config,
+    get_llm_from_config_obj,
+    llm_property,
 )
-from .llm_fallback import LLMFallback
-from .llm_middleware import (
-    LLMRequest,
-    LLMResponse,
+from .llm_function import (
+    BaseModelResponseParser,
+    CodeBlocksResponseParser,
+    EchoResponseParser,
+    ExecuteLLMRequest,
+    FunctionOutOfRetryError,
+    JSONBlockResponseParser,
+    JSONResponseParser,
+    LLMCachingMiddleware,
+    LLMFileLoggingMiddleware,
+    LLMFunction,
+    LLMFunctionError,
+    LLMFunctionResponse,
+    LLMFunctionWithPrompt,
+    LLMLoggingMiddleware,
+    LLMLoggingStrategy,
     LLMMiddleware,
     LLMMiddlewareChain,
+    LLMRequest,
+    LLMResponse,
     LLMRetryMiddleware,
-    LLMLoggingStrategy,
-    LLMLoggingMiddleware,
-    LLMFileLoggingMiddleware,
-    LLMCachingMiddleware,
-    ExecuteLLMRequest,
-)
-from .llm_response_parser import (
-    EchoResponseParser,
     StringResponseParser,
-    CodeBlocksResponseParser,
-    JSONBlockResponseParser,
-    JSONResponseParser,
     YAMLBlockResponseParser,
     YAMLResponseParser,
 )
-from .llm_function import LLMFunction, LLMFunctionResponse, LLMFunctionError, FunctionOutOfRetryError
-from .llm_function_with_prompt import LLMFunctionWithPrompt
-from .monitored_llm import MonitoredLLM
-
-from .providers import (
-    _build_llm,
-    _PROVIDER_TO_LLM,
-    AzureLLM,
-    AzureChatGPTConfiguration,
-    OpenAILLM,
-    OpenAIChatGPTConfiguration,
-    AnthropicLLM,
-    AnthropicLLMConfiguration,
-    GeminiLLM,
-    GeminiLLMConfiguration,
-    GroqLLM,
-    GroqLLMConfiguration,
-    OllamaLLM,
-    OllamaLLMConfiguration,
-)
-
-
-def get_default_llm(max_retries: Optional[int] = None) -> LLMBase:
-    """Get default LLM based on `COUNCIL_DEFAULT_LLM_PROVIDER` env variable."""
-    provider_str = read_env_str("COUNCIL_DEFAULT_LLM_PROVIDER", default=LLMProviders.OpenAI).unwrap()
-    provider_str = provider_str.lower() + "spec"
-
-    llm_class: Optional[Type[LLMBase]] = next(
-        (llm_class for provider_enum, llm_class in _PROVIDER_TO_LLM.items() if provider_str == provider_enum.lower()),
-        None,
-    )
-
-    if llm_class is None:
-        raise ValueError(f"Provider {provider_str} not supported by Council.")
-
-    llm = llm_class.from_env()
-
-    if max_retries is not None and max_retries > 0:
-        return LLMFallback(llm=llm, fallback=llm, retry_before_fallback=max_retries - 1)
-
-    return llm
-
-
-def get_llm_from_config(filename: str) -> LLMBase:
-    """Get LLM from a yaml LLMConfigObject file."""
-    llm_config = LLMConfigObject.from_yaml(filename)
-    return get_llm_from_config_obj(llm_config)
-
-
-def get_llm_from_config_obj(llm_config: LLMConfigObject):
-    llm = _build_llm(llm_config)
-    fallback_provider = llm_config.spec.fallback_provider
-    if fallback_provider is not None:
-        llm_config.spec.provider = fallback_provider
-        llm_with_fallback = _build_llm(llm_config)
-        return LLMFallback(llm, llm_with_fallback)
-    return llm

council/llm/base/__init__.py

@@ -0,0 +1,77 @@
+from typing import Optional, Type
+
+from .llm_config_object import LLMProvider, LLMConfigObject, LLMConfigSpec, LLMProviders
+from .llm_answer import llm_property, LLMAnswer, LLMProperty, LLMParsingException
+from .llm_exception import (
+    LLMException,
+    LLMCallException,
+    LLMCallTimeoutException,
+    LLMTokenLimitException,
+    LLMOutOfRetriesException,
+)
+from .llm_message import LLMMessageRole, LLMMessage, LLMMessageData, LLMCacheControlData, LLMMessageTokenCounterBase
+from .llm_base import LLMBase, LLMResult, LLMConfigurationBase, T_Configuration
+from .llm_cost import (
+    LLMCostCard,
+    LLMCostManagerObject,
+    TokenKind,
+    LLMConsumptionCalculatorBase,
+    DefaultLLMConsumptionCalculator,
+)
+from .llm_fallback import LLMFallback
+from .monitored_llm import MonitoredLLM
+
+from .providers import (
+    _build_llm,
+    _PROVIDER_TO_LLM,
+    AzureLLM,
+    AzureChatGPTConfiguration,
+    OpenAILLM,
+    OpenAIChatGPTConfiguration,
+    AnthropicLLM,
+    AnthropicLLMConfiguration,
+    GeminiLLM,
+    GeminiLLMConfiguration,
+    GroqLLM,
+    GroqLLMConfiguration,
+    OllamaLLM,
+    OllamaLLMConfiguration,
+)
+from ...utils import read_env_str
+
+
+def get_default_llm(max_retries: Optional[int] = None) -> LLMBase:
+    """Get default LLM based on `COUNCIL_DEFAULT_LLM_PROVIDER` env variable."""
+    provider_str = read_env_str("COUNCIL_DEFAULT_LLM_PROVIDER", default=LLMProviders.OpenAI).unwrap()
+    provider_str = provider_str.lower() + "spec"
+
+    llm_class: Optional[Type[LLMBase]] = next(
+        (llm_class for provider_enum, llm_class in _PROVIDER_TO_LLM.items() if provider_str == provider_enum.lower()),
+        None,
+    )
+
+    if llm_class is None:
+        raise ValueError(f"Provider {provider_str} not supported by Council.")
+
+    llm = llm_class.from_env()
+
+    if max_retries is not None and max_retries > 0:
+        return LLMFallback(llm=llm, fallback=llm, retry_before_fallback=max_retries - 1)
+
+    return llm
+
+
+def get_llm_from_config(filename: str) -> LLMBase:
+    """Get LLM from a yaml LLMConfigObject file."""
+    llm_config = LLMConfigObject.from_yaml(filename)
+    return get_llm_from_config_obj(llm_config)
+
+
+def get_llm_from_config_obj(llm_config: LLMConfigObject):
+    llm = _build_llm(llm_config)
+    fallback_provider = llm_config.spec.fallback_provider
+    if fallback_provider is not None:
+        llm_config.spec.provider = fallback_provider
+        llm_with_fallback = _build_llm(llm_config)
+        return LLMFallback(llm, llm_with_fallback)
+    return llm

council/llm/llm_base.py → council/llm/base/llm_base.py

@@ -4,7 +4,7 @@
 from council.contexts import Consumption, LLMContext, Monitorable
 from typing_extensions import Self
 
-from . import LLMConfigObject, LLMConfigSpec
+from .llm_config_object import LLMConfigObject, LLMConfigSpec
 from .llm_message import LLMMessage, LLMMessageTokenCounterBase
 
 _DEFAULT_TIMEOUT: Final[int] = 30

council/llm/llm_fallback.py → council/llm/base/llm_fallback.py

@@ -4,16 +4,11 @@
 from typing import Any, Sequence
 
 from council.contexts import LLMContext
-from council.llm import (
-    LLMBase,
-    LLMCallException,
-    LLMConfigSpec,
-    LLMConfigurationBase,
-    LLMException,
-    LLMMessage,
-    LLMResult,
-)
-from council.llm.llm_base import T_Configuration
+
+from .llm_base import LLMBase, LLMConfigurationBase, LLMResult, T_Configuration
+from .llm_config_object import LLMConfigSpec
+from .llm_exception import LLMCallException, LLMException
+from .llm_message import LLMMessage
 
 
 class LLMFallbackConfiguration(LLMConfigurationBase):

council/llm/monitored_llm.py → council/llm/base/monitored_llm.py

@@ -1,7 +1,9 @@
 from typing import Any, Optional, Sequence
 
 from council.contexts import Budget, ContextBase, LLMContext, Monitored
-from council.llm import LLMBase, LLMMessage, LLMResult
+
+from .llm_base import LLMBase, LLMResult
+from .llm_message import LLMMessage
 
 
 class MonitoredLLM(Monitored[LLMBase]):

council/llm/providers/anthropic/anthropic.py → council/llm/base/providers/anthropic/anthropic.py

@@ -5,7 +5,8 @@
 from typing import Any, Dict, List, Optional, Sequence
 
 from anthropic.types import Completion
-from council.llm import LLMMessage
+
+from ...llm_message import LLMMessage
 
 
 class Usage:

council/llm/providers/anthropic/anthropic_completion_llm.py → council/llm/base/providers/anthropic/anthropic_completion_llm.py

@@ -2,8 +2,8 @@
 
 from anthropic import Anthropic
 from anthropic._types import NOT_GIVEN
-from council.llm import LLMMessage, LLMMessageRole
 
+from ...llm_message import LLMMessage, LLMMessageRole
 from .anthropic import AnthropicAPIClientResult, AnthropicAPIClientWrapper
 from .anthropic_llm_configuration import AnthropicLLMConfiguration
 

council/llm/providers/anthropic/anthropic_llm.py → council/llm/base/providers/anthropic/anthropic_llm.py

@@ -4,9 +4,11 @@
 
 from anthropic import Anthropic, APIStatusError, APITimeoutError
 from council.contexts import Consumption, LLMContext
-from council.llm import LLMBase, LLMCallException, LLMCallTimeoutException, LLMMessage, LLMResult
 from council.utils.utils import DurationManager
 
+from ...llm_base import LLMBase, LLMResult
+from ...llm_exception import LLMCallException, LLMCallTimeoutException
+from ...llm_message import LLMMessage
 from .anthropic import AnthropicAPIClientWrapper, Usage
 from .anthropic_completion_llm import AnthropicCompletionLLM
 from .anthropic_llm_configuration import AnthropicLLMConfiguration

council/llm/providers/anthropic/anthropic_llm_configuration.py → council/llm/base/providers/anthropic/anthropic_llm_configuration.py

@@ -2,7 +2,6 @@
 
 from typing import Any, Final, Optional
 
-from council.llm import LLMConfigSpec, LLMConfigurationBase, LLMProviders
 from council.utils import (
     Parameter,
     greater_than_validator,
@@ -12,6 +11,9 @@
     zero_to_one_validator,
 )
 
+from ...llm_base import LLMConfigSpec, LLMConfigurationBase
+from ...llm_config_object import LLMProviders
+
 _env_var_prefix: Final[str] = "ANTHROPIC_"
 
 

council/llm/providers/anthropic/anthropic_llm_cost.py → council/llm/base/providers/anthropic/anthropic_llm_cost.py

@@ -4,8 +4,8 @@
 from typing import Final, List, Mapping, Optional
 
 from council.contexts import Consumption
-from council.llm import LLMConsumptionCalculatorBase, LLMCostCard, LLMCostManagerObject, TokenKind
 
+from ...llm_cost import LLMConsumptionCalculatorBase, LLMCostCard, LLMCostManagerObject, TokenKind
 from .anthropic import Usage
 
 ANTHROPIC_COSTS_FILENAME: Final[str] = os.path.join(

council/llm/providers/anthropic/anthropic_messages_llm.py → council/llm/base/providers/anthropic/anthropic_messages_llm.py

@@ -5,9 +5,8 @@
 from anthropic import Anthropic
 from anthropic._types import NOT_GIVEN
 from anthropic.types import MessageParam, TextBlock
-from council.llm import LLMMessage, LLMMessageRole
-from council.llm.llm_message import LLMCacheControlData
 
+from ...llm_message import LLMCacheControlData, LLMMessage, LLMMessageRole
 from .anthropic import AnthropicAPIClientResult, AnthropicAPIClientWrapper, Usage
 from .anthropic_llm_configuration import AnthropicLLMConfiguration
 

council/llm/providers/gemini/gemini_llm.py → council/llm/base/providers/gemini/gemini_llm.py

@@ -4,12 +4,13 @@
 
 import google.generativeai as genai  # type: ignore
 from council.contexts import Consumption, LLMContext
-from council.llm import LLMBase, LLMMessage, LLMMessageRole, LLMResult
 from council.utils.utils import DurationManager
 from google.ai.generativelanguage import FileData
 from google.ai.generativelanguage_v1 import HarmCategory  # type: ignore
 from google.generativeai.types import GenerateContentResponse, HarmBlockThreshold  # type: ignore
 
+from ...llm_base import LLMBase, LLMResult
+from ...llm_message import LLMMessage, LLMMessageRole
 from .gemini_llm_configuration import GeminiLLMConfiguration
 from .gemini_llm_cost import GeminiConsumptionCalculator