ifixkart-backend/swarm/providers/mock_provider.py

83 lines
3.1 KiB
Python

"""
@provider MockProvider (swarm/providers/mock_provider.py)
@purpose Deterministic offline provider for zero-cost testing and CI validation.
"""
import time
import json
from typing import Type, TypeVar, List, Dict, Tuple
from pydantic import BaseModel
from swarm.providers.base import BaseLLMProvider, TokenUsage
T = TypeVar("T", bound=BaseModel)
class MockProvider(BaseLLMProvider):
def __init__(self, latency_ms: int = 15):
self.latency_ms = latency_ms
async def generate_structured(
self,
prompt: str,
schema: Type[T],
system_prompt: str = ""
) -> Tuple[T, TokenUsage]:
start = time.perf_counter()
# Build mock data corresponding to schema fields
mock_data = {}
for field_name, field_info in schema.model_fields.items():
field_type = field_info.annotation
# Simple heuristic mapping for typical simulation schemas
if field_name == "archetype_id":
mock_data[field_name] = "ARCH-001"
elif field_name == "decisions" or "list" in str(field_type).lower():
mock_data[field_name] = []
elif field_type in (int,):
mock_data[field_name] = 42
elif field_type in (float,):
mock_data[field_name] = 0.185
elif field_type in (bool,):
mock_data[field_name] = True
elif field_type in (dict,):
mock_data[field_name] = {"mock_key": "mock_value"}
else:
mock_data[field_name] = f"Simulated observation for {field_name}"
# If schema has a custom mock generator method or validator
instance = schema.model_validate(mock_data)
elapsed_ms = int((time.perf_counter() - start) * 1000) + self.latency_ms
usage = TokenUsage(
prompt_tokens=len(prompt.split()) * 2,
completion_tokens=45,
total_tokens=len(prompt.split()) * 2 + 45,
latency_ms=elapsed_ms,
model_name="mock-deterministic",
estimated_cost_usd=0.0
)
return instance, usage
async def chat(
self,
messages: List[Dict[str, str]],
system_prompt: str = ""
) -> Tuple[str, TokenUsage]:
start = time.perf_counter()
last_user_msg = messages[-1]["content"] if messages else "Hello"
reply = (
f"[Simulated Persona Response]: Based on my budget constraints and repair priorities, "
f"regarding '{last_user_msg}': I decided to compare prices first because I value genuine OEM parts "
f"over a small discount, but I appreciate the transparent warranty option."
)
elapsed_ms = int((time.perf_counter() - start) * 1000) + self.latency_ms
usage = TokenUsage(
prompt_tokens=sum(len(m["content"].split()) for m in messages) * 2,
completion_tokens=len(reply.split()) * 2,
total_tokens=120,
latency_ms=elapsed_ms,
model_name="mock-deterministic",
estimated_cost_usd=0.0
)
return reply, usage