Spaces:
Running on CPU Upgrade
Running on CPU Upgrade
File size: 10,803 Bytes
61d29fc | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 | """
MLflow-based agent foundation for Databricks Agent Bricks.
Provides:
- MLflow Pyfunc model wrappers for agents
- Unity Catalog integration
- Automatic tracing and observability
- Model serving compatibility
"""
from typing import Any, Dict, List, Optional, Union
from abc import ABC, abstractmethod
import mlflow
from mlflow.pyfunc import PythonModel
from mlflow.models import infer_signature
from mlflow.tracking import MlflowClient
import pandas as pd
from datetime import datetime
from loguru import logger
from agents.base import AgentRole, AgentMessage, AgentStatus
from config import settings
class MLflowAgentBase(PythonModel, ABC):
"""
Base class for agents that can be deployed via MLflow Model Serving.
Integrates with:
- Unity Catalog for governance
- MLflow Tracking for experimentation
- Databricks Model Serving for deployment
- Mosaic AI Agent Framework for evaluation
"""
def __init__(self, agent_id: str, role: AgentRole):
"""
Initialize MLflow agent.
Args:
agent_id: Unique identifier for this agent
role: Agent role in the pipeline
"""
super().__init__()
self.agent_id = agent_id
self.role = role
self.status = AgentStatus.IDLE
self.client = MlflowClient()
def load_context(self, context):
"""
Load agent context from MLflow (called during model loading).
Args:
context: MLflow context with model artifacts
"""
logger.info(f"Loading {self.role.value} agent from MLflow context")
# Load any model artifacts, configs, etc.
pass
@abstractmethod
def _process_request(self, request: Dict[str, Any]) -> Dict[str, Any]:
"""
Process a single agent request.
Args:
request: Input request dictionary
Returns:
Response dictionary
"""
pass
def predict(
self,
context,
model_input: Union[pd.DataFrame, Dict[str, Any], List[Dict[str, Any]]]
) -> Union[pd.DataFrame, List[Dict[str, Any]]]:
"""
MLflow Pyfunc predict interface.
This is the main entry point when the agent is deployed as a Model Serving endpoint.
Args:
context: MLflow context
model_input: Input data (DataFrame, dict, or list of dicts)
Returns:
Predictions in same format as input
"""
# Enable MLflow tracing for observability
with mlflow.start_span(name=f"{self.role.value}_agent") as span:
span.set_attribute("agent_id", self.agent_id)
span.set_attribute("agent_role", self.role.value)
try:
# Convert input to standard format
if isinstance(model_input, pd.DataFrame):
requests = model_input.to_dict('records')
return_df = True
elif isinstance(model_input, dict):
requests = [model_input]
return_df = False
else:
requests = model_input
return_df = False
# Process each request with tracing
results = []
for idx, request in enumerate(requests):
with mlflow.start_span(name=f"process_request_{idx}") as req_span:
req_span.set_attribute("request_id", request.get("request_id", f"req_{idx}"))
try:
result = self._process_request(request)
result["status"] = "success"
result["agent_id"] = self.agent_id
result["timestamp"] = datetime.utcnow().isoformat()
results.append(result)
req_span.set_attribute("status", "success")
except Exception as e:
error_result = {
"status": "error",
"error": str(e),
"agent_id": self.agent_id,
"timestamp": datetime.utcnow().isoformat()
}
results.append(error_result)
req_span.set_attribute("status", "error")
req_span.set_attribute("error", str(e))
logger.error(f"Error processing request {idx}: {e}")
# Return in requested format
if return_df:
return pd.DataFrame(results)
elif len(results) == 1 and not isinstance(model_input, list):
return results[0]
else:
return results
except Exception as e:
span.set_attribute("status", "error")
span.set_attribute("error", str(e))
logger.error(f"Error in {self.role.value} agent: {e}")
raise
def log_to_mlflow(
self,
model_name: str,
artifact_path: str = "agent",
registered_model_name: Optional[str] = None,
**kwargs
):
"""
Log this agent to MLflow.
Args:
model_name: Name for the MLflow run
artifact_path: Path within the run to store the model
registered_model_name: Unity Catalog model name (e.g., "main.agents.scraper")
**kwargs: Additional MLflow logging parameters
"""
with mlflow.start_run(run_name=model_name) as run:
# Log agent metadata
mlflow.log_param("agent_id", self.agent_id)
mlflow.log_param("agent_role", self.role.value)
mlflow.log_param("framework", "databricks-agent-bricks")
# Create example input/output for signature
example_input = self._get_example_input()
example_output = self.predict(None, example_input)
signature = infer_signature(example_input, example_output)
# Log the model
mlflow.pyfunc.log_model(
artifact_path=artifact_path,
python_model=self,
signature=signature,
registered_model_name=registered_model_name,
**kwargs
)
logger.info(f"Logged {self.role.value} agent to MLflow run {run.info.run_id}")
if registered_model_name:
logger.info(f"Registered model as {registered_model_name}")
return run.info.run_id
@abstractmethod
def _get_example_input(self) -> Union[pd.DataFrame, Dict[str, Any]]:
"""
Get example input for MLflow signature inference.
Returns:
Example input data
"""
pass
def deploy_to_model_serving(
self,
model_name: str,
endpoint_name: str,
workload_size: str = "Small",
scale_to_zero: bool = True
) -> str:
"""
Deploy this agent to Databricks Model Serving.
Args:
model_name: Registered model name in Unity Catalog
endpoint_name: Name for the serving endpoint
workload_size: Endpoint size (Small, Medium, Large)
scale_to_zero: Whether to scale to zero when idle
Returns:
Endpoint URL
"""
from databricks.sdk import WorkspaceClient
from databricks.sdk.service.serving import ServedEntityInput, EndpointCoreConfigInput
w = WorkspaceClient(
host=settings.databricks_host,
token=settings.databricks_token
)
# Get latest model version
latest_version = self.client.get_latest_versions(model_name, stages=["None"])[0].version
# Create or update endpoint
endpoint_config = EndpointCoreConfigInput(
name=endpoint_name,
served_entities=[
ServedEntityInput(
entity_name=model_name,
entity_version=latest_version,
workload_size=workload_size,
scale_to_zero_enabled=scale_to_zero
)
]
)
try:
endpoint = w.serving_endpoints.create_and_wait(
name=endpoint_name,
config=endpoint_config
)
logger.info(f"Created endpoint: {endpoint_name}")
except Exception as e:
if "already exists" in str(e):
endpoint = w.serving_endpoints.update_config_and_wait(
name=endpoint_name,
served_entities=endpoint_config.served_entities
)
logger.info(f"Updated endpoint: {endpoint_name}")
else:
raise
endpoint_url = f"{settings.databricks_host}/serving-endpoints/{endpoint_name}/invocations"
return endpoint_url
class MLflowChainAgent(MLflowAgentBase):
"""
Agent that uses LangChain with MLflow tracing.
Provides integration with:
- LangChain agents and chains
- Automatic prompt logging
- LLM call tracing
- Tool usage tracking
"""
def __init__(self, agent_id: str, role: AgentRole):
"""Initialize LangChain-based agent."""
super().__init__(agent_id, role)
self.chain = None
def _setup_langchain_tracing(self):
"""Enable MLflow tracing for LangChain."""
mlflow.langchain.autolog()
@abstractmethod
def _build_chain(self):
"""
Build the LangChain chain for this agent.
Returns:
LangChain chain or agent
"""
pass
def _process_request(self, request: Dict[str, Any]) -> Dict[str, Any]:
"""Process request through LangChain."""
if self.chain is None:
self.chain = self._build_chain()
with mlflow.start_span(name="langchain_invoke") as span:
result = self.chain.invoke(request)
# Log relevant metrics
if hasattr(result, "llm_output"):
span.set_attribute("tokens_used", result.llm_output.get("token_usage", {}).get("total_tokens", 0))
return result
|