mirror of
https://github.com/GNS3/gns3-server.git
synced 2026-08-27 20:40:13 +03:00
feat(agent): enhance LLM config loading with priority system
Refactor model factory to support three configuration sources in priority order: 1. Direct llm_config dictionary parameter 2. Fetch from llm_model_configs system via connector_factory (requires user_id and jwt_token) 3. Environment variables as fallback for backward compatibility This improves flexibility by allowing runtime configuration while maintaining compatibility with existing environment-based setups.
This commit is contained in:
YueGuobin
parent
15dc1e7ea0
commit
db9db56e7a
@ -2,33 +2,45 @@
|
||||
Model Factory for FlowNet-Lab Agent
|
||||
|
||||
This module provides factory functions to create fresh LLM model instances.
|
||||
Configuration can be loaded from:
|
||||
1. Passed llm_config dictionary (from new llm_model_configs system)
|
||||
2. Environment variables (fallback for backward compatibility)
|
||||
Configuration is loaded from the llm_model_configs system via connector_factory.
|
||||
"""
|
||||
|
||||
import logging
|
||||
import os
|
||||
from typing import Any, Optional
|
||||
from uuid import UUID
|
||||
|
||||
from langchain.chat_models import init_chat_model
|
||||
|
||||
from gns3_copilot.gns3_client import get_llm_config
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
def _load_llm_config(llm_config: Optional[dict[str, Any]] = None) -> dict[str, str]:
|
||||
def _load_llm_config(
|
||||
user_id: Optional[UUID] = None,
|
||||
jwt_token: Optional[str] = None,
|
||||
llm_config: Optional[dict[str, Any]] = None,
|
||||
) -> dict[str, str]:
|
||||
"""
|
||||
Load model configuration from llm_config dict or environment variables.
|
||||
Load model configuration from llm_config dict or fetch from llm_model_configs system.
|
||||
|
||||
Priority order:
|
||||
1. Provided llm_config dictionary (highest priority)
|
||||
2. Fetch from llm_model_configs system via connector_factory (requires user_id and jwt_token)
|
||||
3. Environment variables (fallback for backward compatibility)
|
||||
|
||||
Args:
|
||||
llm_config: Optional configuration dictionary from llm_model_configs system.
|
||||
If not provided, will load from environment variables.
|
||||
user_id: User UUID for fetching config from database
|
||||
jwt_token: JWT token for API authentication
|
||||
llm_config: Optional configuration dictionary to use directly
|
||||
|
||||
Returns:
|
||||
Dictionary containing model configuration.
|
||||
"""
|
||||
# Priority 1: Use provided llm_config dictionary
|
||||
if llm_config:
|
||||
# Use provided configuration (from llm_model_configs system)
|
||||
logger.info("Using provided llm_config dictionary")
|
||||
return {
|
||||
"model_name": llm_config.get("model", ""),
|
||||
"model_provider": llm_config.get("provider", ""),
|
||||
@ -36,32 +48,66 @@ def _load_llm_config(llm_config: Optional[dict[str, Any]] = None) -> dict[str, s
|
||||
"base_url": llm_config.get("base_url", ""),
|
||||
"temperature": str(llm_config.get("temperature", "0")),
|
||||
}
|
||||
else:
|
||||
# Fallback to environment variables
|
||||
return {
|
||||
"model_name": os.getenv("MODEL_NAME", ""),
|
||||
"model_provider": os.getenv("MODE_PROVIDER", ""),
|
||||
"api_key": os.getenv("MODEL_API_KEY", ""),
|
||||
"base_url": os.getenv("BASE_URL", ""),
|
||||
"temperature": os.getenv("TEMPERATURE", "0"),
|
||||
}
|
||||
|
||||
# Priority 2: Fetch from llm_model_configs system via connector_factory
|
||||
if user_id and jwt_token:
|
||||
logger.info(f"Fetching LLM config from database for user {user_id}")
|
||||
config = get_llm_config(user_id=user_id, jwt_token=jwt_token)
|
||||
|
||||
if config:
|
||||
logger.info(
|
||||
f"Successfully loaded LLM config from database: "
|
||||
f"provider={config.get('provider')}, model={config.get('model')}"
|
||||
)
|
||||
return {
|
||||
"model_name": config.get("model", ""),
|
||||
"model_provider": config.get("provider", ""),
|
||||
"api_key": config.get("api_key", ""),
|
||||
"base_url": config.get("base_url", ""),
|
||||
"temperature": str(config.get("temperature", "0")),
|
||||
}
|
||||
else:
|
||||
logger.warning(
|
||||
f"No LLM config found in database for user {user_id}, falling back to environment variables"
|
||||
)
|
||||
|
||||
# Priority 3: Fallback to environment variables
|
||||
logger.info("Loading LLM config from environment variables (fallback)")
|
||||
return {
|
||||
"model_name": os.getenv("MODEL_NAME", ""),
|
||||
"model_provider": os.getenv("MODE_PROVIDER", ""),
|
||||
"api_key": os.getenv("MODEL_API_KEY", ""),
|
||||
"base_url": os.getenv("BASE_URL", ""),
|
||||
"temperature": os.getenv("TEMPERATURE", "0"),
|
||||
}
|
||||
|
||||
|
||||
def create_base_model() -> Any:
|
||||
def create_base_model(
|
||||
user_id: Optional[UUID] = None,
|
||||
jwt_token: Optional[str] = None,
|
||||
llm_config: Optional[dict[str, Any]] = None,
|
||||
) -> Any:
|
||||
"""
|
||||
Create a fresh base LLM model instance from current environment variables.
|
||||
Create a fresh base LLM model instance.
|
||||
|
||||
This function reads environment variables fresh every time it's called,
|
||||
allowing configuration changes to take effect immediately.
|
||||
Configuration priority:
|
||||
1. llm_config dictionary (if provided)
|
||||
2. Fetch from llm_model_configs system via connector_factory (if user_id and jwt_token provided)
|
||||
3. Environment variables (fallback)
|
||||
|
||||
Args:
|
||||
user_id: User UUID for fetching config from database
|
||||
jwt_token: JWT token for API authentication
|
||||
llm_config: Optional configuration dictionary to use directly
|
||||
|
||||
Returns:
|
||||
Any: A new LLM model instance configured with current env vars.
|
||||
Any: A new LLM model instance configured with current settings.
|
||||
The actual type depends on the provider (e.g., ChatOpenAI, etc.).
|
||||
|
||||
Raises:
|
||||
ValueError: If required environment variables are missing or invalid.
|
||||
ValueError: If required configuration fields are missing or invalid.
|
||||
"""
|
||||
env_vars = _load_llm_config()
|
||||
env_vars = _load_llm_config(user_id, jwt_token, llm_config)
|
||||
|
||||
# Log the loaded configuration (mask sensitive data)
|
||||
logger.info(
|
||||
@ -98,7 +144,11 @@ def create_base_model() -> Any:
|
||||
raise RuntimeError(f"Failed to create base model: {e}") from e
|
||||
|
||||
|
||||
def create_title_model() -> Any:
|
||||
def create_title_model(
|
||||
user_id: Optional[UUID] = None,
|
||||
jwt_token: Optional[str] = None,
|
||||
llm_config: Optional[dict[str, Any]] = None,
|
||||
) -> Any:
|
||||
"""
|
||||
Create a fresh title generation model instance.
|
||||
|
||||
@ -106,14 +156,24 @@ def create_title_model() -> Any:
|
||||
It uses the same configuration as the base model but with a higher temperature
|
||||
for more creative output.
|
||||
|
||||
Configuration priority:
|
||||
1. llm_config dictionary (if provided)
|
||||
2. Fetch from llm_model_configs system via connector_factory (if user_id and jwt_token provided)
|
||||
3. Environment variables (fallback)
|
||||
|
||||
Args:
|
||||
user_id: User UUID for fetching config from database
|
||||
jwt_token: JWT token for API authentication
|
||||
llm_config: Optional configuration dictionary to use directly
|
||||
|
||||
Returns:
|
||||
Any: A new LLM model instance for title generation.
|
||||
The actual type depends on the provider.
|
||||
|
||||
Raises:
|
||||
ValueError: If required environment variables are missing or invalid.
|
||||
ValueError: If required configuration fields are missing or invalid.
|
||||
"""
|
||||
env_vars = _load_llm_config()
|
||||
env_vars = _load_llm_config(user_id, jwt_token, llm_config)
|
||||
|
||||
logger.info(
|
||||
"Creating title model: name=%s, provider=%s, base_url=%s, temperature=1.0",
|
||||
@ -174,305 +234,35 @@ def create_model_with_tools(
|
||||
raise RuntimeError(f"Failed to bind tools to model: {e}") from e
|
||||
|
||||
|
||||
def create_note_organizer_model() -> Any:
|
||||
"""
|
||||
Create a fresh model instance for note organization.
|
||||
|
||||
This creates a model instance suitable for organizing and formatting notes.
|
||||
It uses the same configuration as the base model but with a lower temperature
|
||||
for more consistent and predictable output.
|
||||
|
||||
Returns:
|
||||
Any: A new LLM model instance for note organization.
|
||||
The actual type depends on the provider.
|
||||
|
||||
Raises:
|
||||
ValueError: If required environment variables are missing or invalid.
|
||||
"""
|
||||
env_vars = _load_llm_config()
|
||||
|
||||
logger.info(
|
||||
"Creating note organizer model: name=%s, provider=%s, base_url=%s, temperature=0.3",
|
||||
env_vars["model_name"],
|
||||
env_vars["model_provider"],
|
||||
env_vars["base_url"] if env_vars["base_url"] else "default",
|
||||
)
|
||||
|
||||
# Validate required fields
|
||||
if not env_vars["model_name"]:
|
||||
raise ValueError("MODEL_NAME environment variable is required")
|
||||
|
||||
if not env_vars["model_provider"]:
|
||||
raise ValueError("MODE_PROVIDER environment variable is required")
|
||||
|
||||
try:
|
||||
model = init_chat_model(
|
||||
env_vars["model_name"],
|
||||
model_provider=env_vars["model_provider"],
|
||||
api_key=env_vars["api_key"],
|
||||
base_url=env_vars["base_url"],
|
||||
temperature="0.3", # Lower temperature for more consistent note organization
|
||||
configurable_fields="any",
|
||||
config_prefix="foo",
|
||||
)
|
||||
|
||||
logger.info("Note organizer model created successfully")
|
||||
return model
|
||||
|
||||
except Exception as e:
|
||||
logger.error("Failed to create note organizer model: %s", e)
|
||||
raise RuntimeError(f"Failed to create note organizer model: {e}") from e
|
||||
|
||||
|
||||
def create_base_model_with_tools(tools: list[Any]) -> Any:
|
||||
def create_base_model_with_tools(
|
||||
tools: list[Any],
|
||||
user_id: Optional[UUID] = None,
|
||||
jwt_token: Optional[str] = None,
|
||||
llm_config: Optional[dict[str, Any]] = None,
|
||||
) -> Any:
|
||||
"""
|
||||
Create a fresh base model instance with tools bound.
|
||||
|
||||
This is a convenience function that combines creating the base model
|
||||
and binding tools to it.
|
||||
|
||||
Configuration priority:
|
||||
1. llm_config dictionary (if provided)
|
||||
2. Fetch from llm_model_configs system via connector_factory (if user_id and jwt_token provided)
|
||||
3. Environment variables (fallback)
|
||||
|
||||
Args:
|
||||
tools: List of tools to bind to the model.
|
||||
user_id: User UUID for fetching config from database
|
||||
jwt_token: JWT token for API authentication
|
||||
llm_config: Optional configuration dictionary to use directly
|
||||
|
||||
Returns:
|
||||
Any: A new model instance with tools bound (type varies by provider).
|
||||
|
||||
Raises:
|
||||
ValueError: If required environment variables are missing.
|
||||
ValueError: If required configuration fields are missing.
|
||||
RuntimeError: If model creation or tool binding fails.
|
||||
"""
|
||||
base_model = create_base_model()
|
||||
base_model = create_base_model(user_id, jwt_token, llm_config)
|
||||
return create_model_with_tools(base_model, tools)
|
||||
|
||||
|
||||
def create_window_agent_base_model() -> Any:
|
||||
"""
|
||||
Create a base model instance for Window Agent (without tools).
|
||||
|
||||
This creates a model instance suitable for Window Agent (voice mode).
|
||||
Uses temperature=0 for precise, deterministic responses.
|
||||
|
||||
Returns:
|
||||
Any: A new base LLM model instance (without tools bound).
|
||||
The actual type depends on the provider.
|
||||
|
||||
Raises:
|
||||
ValueError: If required environment variables are missing.
|
||||
RuntimeError: If model creation fails.
|
||||
"""
|
||||
env_vars = _load_llm_config()
|
||||
|
||||
logger.info(
|
||||
"Creating Window Agent base model: name=%s, provider=%s, base_url=%s, temperature=0",
|
||||
env_vars["model_name"],
|
||||
env_vars["model_provider"],
|
||||
env_vars["base_url"] if env_vars["base_url"] else "default",
|
||||
)
|
||||
|
||||
# Validate required fields
|
||||
if not env_vars["model_name"]:
|
||||
raise ValueError("MODEL_NAME environment variable is required")
|
||||
|
||||
if not env_vars["model_provider"]:
|
||||
raise ValueError("MODE_PROVIDER environment variable is required")
|
||||
|
||||
try:
|
||||
# Create model with temperature=0 for voice mode (precise, concise responses)
|
||||
model = init_chat_model(
|
||||
env_vars["model_name"],
|
||||
model_provider=env_vars["model_provider"],
|
||||
api_key=env_vars["api_key"],
|
||||
base_url=env_vars["base_url"],
|
||||
temperature="0", # Force temperature=0 for voice mode
|
||||
configurable_fields="any",
|
||||
config_prefix="foo",
|
||||
)
|
||||
|
||||
logger.info("Window Agent base model created successfully (no tools)")
|
||||
return model
|
||||
|
||||
except Exception as e:
|
||||
logger.error("Failed to create Window Agent base model: %s", e)
|
||||
raise RuntimeError(f"Failed to create Window Agent base model: {e}") from e
|
||||
|
||||
|
||||
def create_window_agent_model_with_tools(tools: list[Any]) -> Any:
|
||||
"""
|
||||
Create a model instance with tools for Window Agent.
|
||||
|
||||
Window Agent runs in VOICE mode and needs precise, deterministic responses.
|
||||
Uses temperature=0 for consistency.
|
||||
|
||||
Args:
|
||||
tools: List of tools to bind to the model.
|
||||
|
||||
Returns:
|
||||
Any: A new model instance with tools bound (type varies by provider).
|
||||
|
||||
Raises:
|
||||
ValueError: If required environment variables are missing.
|
||||
RuntimeError: If model creation or tool binding fails.
|
||||
"""
|
||||
env_vars = _load_llm_config()
|
||||
|
||||
logger.info(
|
||||
"Creating Window Agent model: name=%s, provider=%s, base_url=%s, temperature=0",
|
||||
env_vars["model_name"],
|
||||
env_vars["model_provider"],
|
||||
env_vars["base_url"] if env_vars["base_url"] else "default",
|
||||
)
|
||||
|
||||
# Validate required fields
|
||||
if not env_vars["model_name"]:
|
||||
raise ValueError("MODEL_NAME environment variable is required")
|
||||
|
||||
if not env_vars["model_provider"]:
|
||||
raise ValueError("MODE_PROVIDER environment variable is required")
|
||||
|
||||
try:
|
||||
# Create model with temperature=0 for voice mode (precise, concise responses)
|
||||
model = init_chat_model(
|
||||
env_vars["model_name"],
|
||||
model_provider=env_vars["model_provider"],
|
||||
api_key=env_vars["api_key"],
|
||||
base_url=env_vars["base_url"],
|
||||
temperature="0", # Force temperature=0 for voice mode
|
||||
configurable_fields="any",
|
||||
config_prefix="foo",
|
||||
)
|
||||
|
||||
# Bind tools
|
||||
model_with_tools = model.bind_tools(tools)
|
||||
logger.info("Window Agent model created with %d tools", len(tools))
|
||||
return model_with_tools
|
||||
|
||||
except Exception as e:
|
||||
logger.error("Failed to create Window Agent model: %s", e)
|
||||
raise RuntimeError(f"Failed to create Window Agent model: {e}") from e
|
||||
|
||||
|
||||
def create_experiment_planner_model() -> Any:
|
||||
"""
|
||||
Create a fresh model instance for experiment planning.
|
||||
|
||||
This creates a model instance suitable for generating GNS3 lab experiment plans.
|
||||
It uses the same configuration as the base model but with a moderate temperature
|
||||
to balance creativity and accuracy in lab design.
|
||||
|
||||
Returns:
|
||||
Any: A new LLM model instance for experiment planning.
|
||||
The actual type depends on the provider.
|
||||
|
||||
Raises:
|
||||
ValueError: If required environment variables are missing or invalid.
|
||||
"""
|
||||
env_vars = _load_llm_config()
|
||||
|
||||
logger.info(
|
||||
"Creating experiment planner model: name=%s, provider=%s, base_url=%s, temperature=0.7",
|
||||
env_vars["model_name"],
|
||||
env_vars["model_provider"],
|
||||
env_vars["base_url"] if env_vars["base_url"] else "default",
|
||||
)
|
||||
|
||||
# Validate required fields
|
||||
if not env_vars["model_name"]:
|
||||
raise ValueError("MODEL_NAME environment variable is required")
|
||||
|
||||
if not env_vars["model_provider"]:
|
||||
raise ValueError("MODE_PROVIDER environment variable is required")
|
||||
|
||||
try:
|
||||
model = init_chat_model(
|
||||
env_vars["model_name"],
|
||||
model_provider=env_vars["model_provider"],
|
||||
api_key=env_vars["api_key"],
|
||||
base_url=env_vars["base_url"],
|
||||
temperature="0.7", # Moderate temperature for balanced creativity and accuracy
|
||||
configurable_fields="any",
|
||||
config_prefix="foo",
|
||||
)
|
||||
|
||||
logger.info("Experiment planner model created successfully")
|
||||
return model
|
||||
|
||||
except Exception as e:
|
||||
logger.error("Failed to create experiment planner model: %s", e)
|
||||
raise RuntimeError(f"Failed to create experiment planner model: {e}") from e
|
||||
|
||||
|
||||
def create_vision_model():
|
||||
"""
|
||||
Create a fresh Qwen-VL vision model instance for network topology recognition.
|
||||
|
||||
This creates a vision model instance using Qwen-VL through DashScope SDK
|
||||
for recognizing network topology diagrams from images.
|
||||
|
||||
Returns:
|
||||
QwenVisionModel: A new Qwen-VL vision model instance.
|
||||
|
||||
Raises:
|
||||
ImportError: If dashscope package is not installed.
|
||||
ValueError: If QWEN_API_KEY is not configured.
|
||||
RuntimeError: If model creation fails.
|
||||
"""
|
||||
from gns3_copilot.agent.qwen_vision_model import create_qwen_vision_model
|
||||
|
||||
logger.info("Creating Qwen-VL vision model")
|
||||
|
||||
try:
|
||||
model = create_qwen_vision_model()
|
||||
logger.info("Qwen-VL vision model created successfully")
|
||||
return model
|
||||
except Exception as e:
|
||||
logger.error("Failed to create Qwen-VL vision model: %s", e)
|
||||
raise RuntimeError(f"Failed to create Qwen-VL vision model: {e}") from e
|
||||
|
||||
|
||||
def create_presentation_eval_model() -> Any:
|
||||
"""
|
||||
Create a fresh model instance for presentation evaluation.
|
||||
|
||||
This creates a model instance suitable for evaluating network engineer
|
||||
presentations. Uses moderate temperature for balanced objective assessment.
|
||||
|
||||
Returns:
|
||||
Any: A new LLM model instance for presentation evaluation.
|
||||
|
||||
Raises:
|
||||
ValueError: If required environment variables are missing.
|
||||
RuntimeError: If model creation fails.
|
||||
"""
|
||||
env_vars = _load_llm_config()
|
||||
|
||||
logger.info(
|
||||
"Creating presentation evaluator model: name=%s, provider=%s, base_url=%s, temperature=0.5",
|
||||
env_vars["model_name"],
|
||||
env_vars["model_provider"],
|
||||
env_vars["base_url"] if env_vars["base_url"] else "default",
|
||||
)
|
||||
|
||||
# Validate required fields
|
||||
if not env_vars["model_name"]:
|
||||
raise ValueError("MODEL_NAME environment variable is required")
|
||||
|
||||
if not env_vars["model_provider"]:
|
||||
raise ValueError("MODE_PROVIDER environment variable is required")
|
||||
|
||||
try:
|
||||
model = init_chat_model(
|
||||
env_vars["model_name"],
|
||||
model_provider=env_vars["model_provider"],
|
||||
api_key=env_vars["api_key"],
|
||||
base_url=env_vars["base_url"],
|
||||
temperature="0.5", # Moderate temperature for balanced evaluation
|
||||
configurable_fields="any",
|
||||
config_prefix="foo",
|
||||
)
|
||||
|
||||
logger.info("Presentation evaluator model created successfully")
|
||||
return model
|
||||
|
||||
except Exception as e:
|
||||
logger.error("Failed to create presentation evaluator model: %s", e)
|
||||
raise RuntimeError(f"Failed to create presentation evaluator model: {e}") from e
|
||||
|
||||
@ -31,6 +31,7 @@ from .connector_factory import (
|
||||
get_gns3_connector,
|
||||
get_gns3_connector_with_llm_config,
|
||||
get_gns3_server_host,
|
||||
get_llm_config,
|
||||
)
|
||||
from .custom_gns3fy import (
|
||||
CONSOLE_TYPES,
|
||||
@ -98,6 +99,7 @@ __all__ = [
|
||||
"get_gns3_connector",
|
||||
"get_gns3_connector_with_llm_config",
|
||||
"get_gns3_server_host",
|
||||
"get_llm_config",
|
||||
"add_file_to_index",
|
||||
"get_file_list",
|
||||
]
|
||||
|
||||
@ -335,3 +335,60 @@ def get_gns3_server_host() -> str:
|
||||
except Exception as e:
|
||||
logger.warning("Failed to extract host from URL %s: %s, using fallback", url, e)
|
||||
return DEFAULT_GNS3_URL.split("://")[1].split(":")[0]
|
||||
|
||||
|
||||
def get_llm_config(user_id, jwt_token: str) -> Optional[dict]:
|
||||
"""
|
||||
Get LLM model configuration for a user.
|
||||
|
||||
This is a convenience function that retrieves the user's LLM configuration
|
||||
from the llm_model_configs system.
|
||||
|
||||
Args:
|
||||
user_id: User UUID (can be string or UUID object)
|
||||
jwt_token: JWT token for authentication
|
||||
|
||||
Returns:
|
||||
Dictionary with LLM configuration keys (provider, model, api_key, etc.),
|
||||
or None if not found.
|
||||
|
||||
Example:
|
||||
from gns3_copilot.gns3_client import get_llm_config
|
||||
|
||||
config = get_llm_config(user_id, jwt_token)
|
||||
if config:
|
||||
provider = config['provider']
|
||||
model = config['model']
|
||||
api_key = config['api_key']
|
||||
"""
|
||||
import asyncio
|
||||
|
||||
try:
|
||||
# Convert user_id to UUID if it's a string
|
||||
if isinstance(user_id, str):
|
||||
user_id = UUID(user_id)
|
||||
|
||||
# Detect GNS3 URL
|
||||
url = _detect_url_for_api()
|
||||
|
||||
# Import the async helper
|
||||
from gns3_copilot.utils.llm_config_helper import get_user_llm_config
|
||||
|
||||
# Run async function in sync context
|
||||
loop = asyncio.get_event_loop()
|
||||
if loop.is_running():
|
||||
# If loop is already running, run in a new thread
|
||||
import concurrent.futures
|
||||
with concurrent.futures.ThreadPoolExecutor() as executor:
|
||||
future = executor.submit(
|
||||
asyncio.run,
|
||||
get_user_llm_config(user_id, jwt_token, url)
|
||||
)
|
||||
return future.result(timeout=10)
|
||||
else:
|
||||
# No loop running, use run() directly
|
||||
return asyncio.run(get_user_llm_config(user_id, jwt_token, url))
|
||||
|
||||
except Exception as e:
|
||||
logger.error("Failed to get LLM config for user %s: %s", user_id, e)
|
||||
return None
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user