McRogueFace/tests/vllm_demo/3_multi_turn_demo.py

319 lines
9.0 KiB
Python

#!/usr/bin/env python3
"""
Multi-Turn Simulation Demo
==========================
Runs multiple turns of agent interaction with full logging.
This is the Phase 1 implementation from issue #154.
Two agents start in separate rooms and can move, observe,
and (in future versions) communicate to solve puzzles.
"""
import sys
import os
# Add the vllm_demo directory to path for imports
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
import mcrfpy
from mcrfpy import automation
import requests
import base64
from world_graph import (
WorldGraph, Room, Door, WorldObject, Direction, AgentInfo,
create_two_room_scenario, create_button_door_scenario
)
from action_parser import parse_action
from action_executor import ActionExecutor
from turn_orchestrator import TurnOrchestrator, SimulationLog
# Configuration
VLLM_URL = "http://192.168.1.100:8100/v1/chat/completions"
SCREENSHOT_DIR = "/tmp/vllm_multi_turn"
LOG_PATH = "/tmp/vllm_multi_turn/simulation_log.json"
MAX_TURNS = 5
# Sprites
FLOOR_TILE = 0
WALL_TILE = 40
WIZARD_SPRITE = 84
KNIGHT_SPRITE = 96
class Agent:
"""Agent with WorldGraph integration."""
def __init__(self, name: str, display_name: str, entity, world: WorldGraph):
self.name = name
self.display_name = display_name
self.entity = entity
self.world = world
self.message_history = []
@property
def pos(self) -> tuple:
return (int(self.entity.pos[0]), int(self.entity.pos[1]))
@property
def current_room(self) -> str:
room = self.world.room_at(*self.pos)
return room.name if room else None
def get_context(self, visible_agents: list) -> dict:
"""Build context for LLM query."""
room_name = self.current_room
agent_infos = [
AgentInfo(
name=a.name,
display_name=a.display_name,
position=a.pos,
is_player=(a.name == self.name)
)
for a in visible_agents
]
return {
"location": self.world.describe_room(room_name, agent_infos, self.name),
"available_actions": self.world.get_available_actions(room_name),
"recent_messages": self.message_history[-5:],
}
def file_to_base64(path: str) -> str:
"""Convert file to base64 string."""
with open(path, 'rb') as f:
return base64.b64encode(f.read()).decode('utf-8')
def llm_query(agent, screenshot_path: str, context: dict) -> str:
"""
Query VLLM for agent action.
This function is passed to TurnOrchestrator as the LLM query callback.
"""
system_prompt = f"""You are {agent.display_name} exploring a dungeon.
You receive visual and text information about your surroundings.
Your goal is to explore, find items, and interact with the environment.
Always end your response with: Action: <YOUR_ACTION>"""
actions_str = ", ".join(context["available_actions"])
user_prompt = f"""{context["location"]}
Available actions: {actions_str}
[Screenshot attached showing your current view - dark areas are outside your vision]
What do you do? Brief reasoning (1-2 sentences), then Action: <action>"""
messages = [
{"role": "system", "content": system_prompt},
{
"role": "user",
"content": [
{"type": "text", "text": user_prompt},
{"type": "image_url", "image_url": {
"url": "data:image/png;base64," + file_to_base64(screenshot_path)
}}
]
}
]
try:
resp = requests.post(VLLM_URL, json={'messages': messages}, timeout=60)
data = resp.json()
if "error" in data:
return f"[VLLM Error: {data['error']}]"
return data.get('choices', [{}])[0].get('message', {}).get('content', 'No response')
except Exception as e:
return f"[Connection Error: {e}]"
def setup_scene(world: WorldGraph):
"""Create McRogueFace scene from WorldGraph."""
mcrfpy.createScene("multi_turn")
mcrfpy.setScene("multi_turn")
ui = mcrfpy.sceneUI("multi_turn")
texture = mcrfpy.Texture("assets/kenney_TD_MR_IP.png", 16, 16)
grid = mcrfpy.Grid(
grid_size=(25, 15),
texture=texture,
pos=(5, 5),
size=(1014, 700)
)
grid.fill_color = mcrfpy.Color(20, 20, 30)
grid.zoom = 2.0
ui.append(grid)
# Initialize all as walls
for x in range(25):
for y in range(15):
p = grid.at(x, y)
p.tilesprite = WALL_TILE
p.walkable = False
p.transparent = False
# Carve rooms from WorldGraph
for room in world.rooms.values():
for rx in range(room.x, room.x + room.width):
for ry in range(room.y, room.y + room.height):
if 0 <= rx < 25 and 0 <= ry < 15:
p = grid.at(rx, ry)
p.tilesprite = FLOOR_TILE
p.walkable = True
p.transparent = True
# Place doors
for door in world.doors:
dx, dy = door.position
if 0 <= dx < 25 and 0 <= dy < 15:
p = grid.at(dx, dy)
p.tilesprite = FLOOR_TILE
p.walkable = not door.locked
p.transparent = True
# FOV layer
fov_layer = grid.add_layer('color', z_index=10)
fov_layer.fill(mcrfpy.Color(0, 0, 0, 255))
return grid, fov_layer, texture
def create_agents(grid, world: WorldGraph, texture) -> list:
"""Create agents in their starting rooms."""
agents = []
# Wizard in guard_room (left)
room_a = world.rooms["guard_room"]
wizard = mcrfpy.Entity(
grid_pos=room_a.center,
texture=texture,
sprite_index=WIZARD_SPRITE
)
grid.entities.append(wizard)
agents.append(Agent("Wizard", "a wizard", wizard, world))
# Knight in armory (right)
room_b = world.rooms["armory"]
knight = mcrfpy.Entity(
grid_pos=room_b.center,
texture=texture,
sprite_index=KNIGHT_SPRITE
)
grid.entities.append(knight)
agents.append(Agent("Knight", "a knight", knight, world))
return agents
def run_demo():
"""Run multi-turn simulation."""
print("=" * 70)
print("Multi-Turn Simulation Demo")
print(f"Running up to {MAX_TURNS} turns with 2 agents")
print("=" * 70)
os.makedirs(SCREENSHOT_DIR, exist_ok=True)
# Create world
print("\nCreating world...")
world = create_two_room_scenario()
print(f" Rooms: {list(world.rooms.keys())}")
print(f" Objects: {list(world.objects.keys())}")
# Setup scene
print("\nSetting up scene...")
grid, fov_layer, texture = setup_scene(world)
# Create agents
print("\nCreating agents...")
agents = create_agents(grid, world, texture)
for agent in agents:
print(f" {agent.name} at {agent.pos} in {agent.current_room}")
# Create orchestrator
orchestrator = TurnOrchestrator(
grid=grid,
fov_layer=fov_layer,
world=world,
agents=agents,
screenshot_dir=SCREENSHOT_DIR,
llm_query_fn=llm_query
)
# Optional: Define a stop condition
def agents_met(orch):
"""Stop when agents are in the same room."""
return orch.agents_in_same_room()
# Run simulation
log = orchestrator.run_simulation(
max_turns=MAX_TURNS,
stop_condition=None # Or use agents_met for early stopping
)
# Save log
log.save(LOG_PATH)
# Print summary
print("\n" + "=" * 70)
print(log.summary())
print("=" * 70)
# Show final positions
print("\nFinal Agent Positions:")
for agent in agents:
print(f" {agent.name}: {agent.pos} in {agent.current_room}")
print(f"\nScreenshots saved to: {SCREENSHOT_DIR}/")
print(f"Simulation log saved to: {LOG_PATH}")
return True
def replay_log(log_path: str):
"""
Replay a simulation from a log file.
This is a utility function for reviewing past simulations.
"""
print(f"Loading simulation from: {log_path}")
log = SimulationLog.load(log_path)
print("\n" + log.summary())
print("\nTurn-by-Turn Replay:")
print("-" * 50)
current_turn = 0
for step in log.steps:
if step.turn != current_turn:
current_turn = step.turn
print(f"\n=== Turn {current_turn} ===")
status = "OK" if step.result_success else "FAIL"
print(f" {step.agent_id}: {step.parsed_action_type} {step.parsed_action_args}")
print(f" {status}: {step.result_message}")
if step.new_position:
print(f" Moved to: {step.new_position}")
if __name__ == "__main__":
# Check for replay mode
if len(sys.argv) > 1 and sys.argv[1] == "--replay":
log_file = sys.argv[2] if len(sys.argv) > 2 else LOG_PATH
replay_log(log_file)
sys.exit(0)
# Normal execution
try:
success = run_demo()
print("\nPASS" if success else "\nFAIL")
sys.exit(0 if success else 1)
except Exception as e:
import traceback
traceback.print_exc()
sys.exit(1)