Merge pull request #193 from kyaukyuai/feat/llm

Update Azure OpenAI API version and LLM creation logic

.env.sample

@@ -7,8 +7,9 @@ OPENAI_API_KEY=<your-openai-api-key>
 
 # USE when ENDPOINT=AZURE
 AZURE_OPENAI_API_KEY=<your-azure-opnai-api-key>
-AZURE_OPENAI_ENDPOINT=https://<your-azure-openai-endpoint>.openai.azure.com/
+AZURE_OPENAI_API_VERSION=2023-07-01-preview
 AZURE_OPENAI_DEPLOYMENT_NAME=<your-azure-openai-deployment-name>
+AZURE_OPENAI_ENDPOINT=https://<your-azure-openai-endpoint>.openai.azure.com/
 
 # LangSmith
 LANGCHAIN_TRACING_V2=true

gpt_all_star/core/agents/agent.py

@@ -6,24 +6,21 @@
 from dataclasses import dataclass
 from datetime import datetime
 from enum import Enum
-from functools import lru_cache
 
-import openai
 from langchain.agents.agent import AgentExecutor
 from langchain.agents.agent_toolkits.file_management.toolkit import (
     FileManagementToolkit,
 )
 from langchain.agents.openai_tools.base import create_openai_tools_agent
-from langchain_core.language_models import BaseChatModel
 from langchain_core.messages import BaseMessage
 from langchain_core.prompts.chat import ChatPromptTemplate, MessagesPlaceholder
 from langchain_core.prompts.prompt import PromptTemplate
-from langchain_openai import AzureChatOpenAI, ChatOpenAI
 from rich.markdown import Markdown
 from rich.panel import Panel
 from rich.table import Table
 
 from gpt_all_star.cli.console_terminal import ConsoleTerminal
+from gpt_all_star.core.llm import LLM_TYPE, create_llm
 from gpt_all_star.core.message import Message
 from gpt_all_star.core.storage import Storages
 from gpt_all_star.core.tools.shell_tool import ShellTool
@@ -47,7 +44,7 @@ def __init__(
         language: str | None = None,
     ) -> None:
         self.console = ConsoleTerminal()
-        self._llm = _create_llm(os.getenv("OPENAI_API_MODEL_NAME"), 0.1)
+        self._llm = create_llm(LLM_TYPE[os.getenv("ENDPOINT", default="OPENAI")])
 
         self.role: AgentRole = role
         self.name: str = name or self._get_default_profile().name
@@ -169,45 +166,6 @@ def _create_executor(self, tools: list) -> AgentExecutor:
         )
 
 
-def _create_llm(model_name: str, temperature: float) -> BaseChatModel:
-    endpoint = os.getenv("ENDPOINT", default="OPENAI")
-    if endpoint == "AZURE":
-        return _create_azure_chat_openai_instance(
-            os.getenv("AZURE_OPENAI_DEPLOYMENT_NAME")
-        )
-    else:
-        return _create_chat_openai_instance(model_name, temperature)
-
-
-def _create_chat_openai_instance(model_name: str, temperature: float):
-    if model_name not in _get_supported_models():
-        raise ValueError(f"Model {model_name} not supported")
-    return ChatOpenAI(
-        model=model_name,
-        temperature=temperature,
-        streaming=True,
-        client=openai.chat.completions,
-    )
-
-
-def _create_azure_chat_openai_instance(model_name: str):
-    return AzureChatOpenAI(
-        openai_api_version=os.getenv("AZURE_OPENAI_API_VERSION", "2023-07-01-preview"),
-        deployment_name=model_name,
-        streaming=True,
-    )
-
-
-def _get_supported_models() -> list[str]:
-    # cache the models list since it is unlikely to change frequently.
-    @lru_cache(maxsize=1)
-    def _fetch_supported_models():
-        openai.api_type = "openai"
-        return [model.id for model in openai.models.list()]
-
-    return _fetch_supported_models()
-
-
 class AgentRole(str, Enum):
     COPILOT = "copilot"
     PRODUCT_OWNER = "product_owner"

gpt_all_star/core/agents/chain.py

@@ -5,7 +5,8 @@
 from langchain.output_parsers.openai_functions import JsonOutputFunctionsParser
 from langchain_core.prompts.chat import ChatPromptTemplate, MessagesPlaceholder
 
-from gpt_all_star.core.agents.agent import Agent, _create_llm
+from gpt_all_star.core.agents.agent import Agent
+from gpt_all_star.core.llm import LLM_TYPE, create_llm
 
 ACTIONS = [
     "Execute a command",
@@ -17,7 +18,7 @@
 
 class Chain:
     def __init__(self) -> None:
-        self._llm = _create_llm(os.getenv("OPENAI_API_MODEL_NAME"), 0.1)
+        self._llm = create_llm(LLM_TYPE[os.getenv("ENDPOINT", default="OPENAI")])
 
     def create_supervisor_chain(self, members: list[Agent] = []):
         members = [member.name for member in members]

gpt_all_star/core/llm.py

@@ -0,0 +1,99 @@
+import os
+from enum import Enum
+
+import openai
+from langchain_anthropic import ChatAnthropic
+from langchain_anthropic.experimental import ChatAnthropicTools
+from langchain_core.language_models.chat_models import BaseChatModel
+from langchain_openai import AzureChatOpenAI, ChatOpenAI
+
+
+class LLM_TYPE(str, Enum):
+    OPENAI = "OPENAI"
+    AZURE = "AZURE"
+    ANTHROPIC = "ANTHROPIC"
+    ANTHROPIC_TOOLS = "ANTHROPIC_TOOLS"
+
+
+def create_llm(llm_name: LLM_TYPE) -> BaseChatModel:
+    if llm_name == LLM_TYPE.OPENAI:
+        return _create_chat_openai(
+            model_name=os.getenv("OPENAI_API_MODEL", "gpt-4-turbo-preview"),
+            temperature=0.1,
+        )
+    elif llm_name == LLM_TYPE.AZURE:
+        return _create_azure_chat_openai(
+            api_key=os.getenv("AZURE_OPENAI_API_KEY"),
+            azure_endpoint=os.getenv(
+                "AZURE_OPENAI_ENDPOINT", "https://interpreter.openai.azure.com/"
+            ),
+            openai_api_version=os.getenv(
+                "AZURE_OPENAI_API_VERSION", "2023-07-01-preview"
+            ),
+            deployment_name=os.getenv("AZURE_OPENAI_DEPLOYMENT_NAME", "gpt-4-32k"),
+            temperature=0.1,
+        )
+    elif llm_name == LLM_TYPE.ANTHROPIC:
+        return _create_chat_anthropic(
+            anthropic_api_key=os.getenv("ANTHROPIC_API_KEY"),
+            model_name=os.getenv("ANTHROPIC_API_MODEL", "claude-3-opus-20240229"),
+            temperature=0.1,
+        )
+    elif llm_name == LLM_TYPE.ANTHROPIC_TOOLS:
+        return _create_chat_anthropic_tools(
+            anthropic_api_key=os.getenv("ANTHROPIC_API_KEY"),
+            model_name=os.getenv("ANTHROPIC_API_MODEL", "claude-3-opus-20240229"),
+            temperature=0.1,
+        )
+    else:
+        raise ValueError(f"Unsupported LLM type: {llm_name}")
+
+
+def _create_chat_openai(model_name: str, temperature: float) -> ChatOpenAI:
+    openai.api_type = "openai"
+    return ChatOpenAI(
+        model_name=model_name,
+        temperature=temperature,
+        streaming=True,
+        client=openai.chat.completions,
+    )
+
+
+def _create_azure_chat_openai(
+    api_key: str,
+    azure_endpoint: str,
+    openai_api_version: str,
+    deployment_name: str,
+    temperature: float,
+) -> AzureChatOpenAI:
+    openai.api_type = "azure"
+    return AzureChatOpenAI(
+        api_key=api_key,
+        azure_endpoint=azure_endpoint,
+        openai_api_version=openai_api_version,
+        deployment_name=deployment_name,
+        temperature=temperature,
+        streaming=True,
+    )
+
+
+def _create_chat_anthropic(
+    anthropic_api_key: str, model_name: str, temperature: float
+) -> ChatAnthropic:
+    return ChatAnthropic(
+        anthropic_api_key=anthropic_api_key,
+        model=model_name,
+        temperature=temperature,
+        streaming=True,
+    )
+
+
+def _create_chat_anthropic_tools(
+    anthropic_api_key: str, model_name: str, temperature: float
+) -> ChatAnthropicTools:
+    return ChatAnthropicTools(
+        anthropic_api_key=anthropic_api_key,
+        model=model_name,
+        temperature=temperature,
+        streaming=True,
+    )