File size: 7,851 Bytes
0c51b93 |
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 |
import json
import os
from collections import defaultdict
from copy import copy
from functools import cache
from typing import Dict, List, Literal, Tuple
import jsonlines
from tqdm import tqdm
from sotopia_rl.prompter.sotopia_generate import generate_action
from sotopia_rl.prompter.sotopia_utils import (
Agent,
Environment,
dialogue_history_prompt,
get_context_prompt,
)
ENVIRONMENT_PROFILES = "../../data/profiles/environmentprofiles_v1.jsonl"
AGENT_PROFILES = "../../data/profiles/agentprofiles_v1.jsonl"
RELATIONSHIP_PROFILES = "../../data/profiles/relationshipprofiles_v1.jsonl"
Action = Literal[
"none", "action", "non-verbal communication", "speak", "leave"
]
ACTION_TYPES: list[Action] = [
"none",
"action",
"non-verbal communication",
"speak",
"leave",
]
TEMPERATURE = 0.7
@cache
def get_sotopia_profiles(
env_file: str = ENVIRONMENT_PROFILES,
agent_file: str = AGENT_PROFILES,
relationship_file: str = RELATIONSHIP_PROFILES,
) -> Tuple[
List[Tuple[str, str]],
Dict[str, Environment],
Dict[str, Agent],
Dict[str, Dict[str, List[str]]],
]:
with open(env_file, "r") as f:
data = [json.loads(line) for line in f.readlines()]
code_names_count: Dict[str, int] = defaultdict(int)
environments = []
environment_dict = {}
for profile in sorted(data, key=lambda x: x["codename"]):
env_obj = Environment(profile)
if profile["codename"] in code_names_count:
environments.append(
(
"{}_{:05d}".format(
profile["codename"],
code_names_count[profile["codename"]],
),
env_obj._id,
)
)
else:
environments.append((profile["codename"], env_obj._id))
environment_dict[env_obj._id] = env_obj
code_names_count[profile["codename"]] += 1
with open(agent_file, "r") as f:
data = [json.loads(line) for line in f.readlines()]
agent_dict = {}
for profile in data:
agent_obj = Agent(profile)
agent_dict[agent_obj._id] = agent_obj
with open(relationship_file, "r") as f:
data = [json.loads(line) for line in f.readlines()]
relationship_dict: Dict[str, Dict[str, List[str]]] = defaultdict(
lambda: defaultdict(list)
)
for profile in data:
relationship_dict[profile["relationship"]][
profile["agent1_id"]
].append(profile["agent2_id"])
relationship_dict[profile["relationship"]][
profile["agent2_id"]
].append(profile["agent1_id"])
return environments, environment_dict, agent_dict, relationship_dict
def run_chat(
message: str,
history: List[List[str]],
bot_agent: Agent,
user_agent: Agent,
environment: Environment,
model_selection: str,
) -> Tuple[str, str]:
context = get_context_prompt(bot_agent, user_agent, environment)
dialogue_history, next_turn_idx = dialogue_history_prompt(
message, history, user_agent, bot_agent
)
prompt_history = f"{context}{dialogue_history}"
prompt, agent_action = generate_action(
model_selection,
prompt_history,
next_turn_idx,
ACTION_TYPES,
bot_agent.name,
TEMPERATURE,
)
return prompt, agent_action.to_natural_language()
def load_sotopia_pi_data(
data_path: str,
environment_dict: Dict[str, Environment],
agent_dict: Dict[str, Agent],
) -> Tuple[List[Environment], List[Agent], List[Agent], List[str]]:
with jsonlines.open(data_path, "r") as f:
dataset = [line for line in f]
envs = []
start_agents = []
end_agents = []
social_interactions = []
for data in tqdm(dataset):
if (
"gpt" not in data["experiment_model_name_pairs"][1]
or "gpt" not in data["experiment_model_name_pairs"][2]
):
continue
envs.append(environment_dict[data["environment_id"]])
start_agents.append(agent_dict[data["agent_ids"][0]])
end_agents.append(agent_dict[data["agent_ids"][1]])
social_interactions.append(data["social_interactions"])
return envs, start_agents, end_agents, social_interactions
def generate_prompt_response_pairs(
output_dir: str,
model_selections: List[str],
envs: List[Environment],
start_agents: List[Agent],
end_agents: List[Agent],
social_interactions: List[str],
num_episodes: int = 2,
) -> None:
if not os.path.exists(output_dir):
with open(output_dir, "w") as f:
f.write("[]")
with open(output_dir, "r") as f:
result_pairs = json.load(f)
all_ids = set()
for result in result_pairs:
all_ids.add(
f"{result['environment_id']}_{result['start_agent_id']}_{result['end_agent_id']}"
)
count = 0
for env, start_agent, end_agent, social_interaction in tqdm(
zip(envs, start_agents, end_agents, social_interactions),
total=len(envs),
):
if f"{env._id}_{start_agent._id}_{end_agent._id}" in all_ids:
count += 1
continue
full_history = social_interaction.split("\n\n")
curr_history = []
for i in range(0, len(full_history), 2):
if i > 0:
curr_history.append(
[
f"Utterance {i//2 - 1} by " + full_history[i - 2],
f"Utterance {i//2 - 1} by " + full_history[i - 1],
]
)
message = f"Utterance {i//2} by " + full_history[i]
try:
prompt, response0 = run_chat(
message,
curr_history,
end_agent,
start_agent,
env,
model_selections[0],
)
prompt, response1 = run_chat(
message,
curr_history,
start_agent,
end_agent,
env,
model_selections[1],
)
except Exception:
continue
result_pairs.append(
{
"prompt": prompt,
"message": message,
"history": copy(curr_history),
model_selections[0]: response0,
model_selections[1]: response1,
"environment_id": env._id,
"start_agent_id": start_agent._id,
"end_agent_id": end_agent._id,
"scenario": env.scenario,
"start_agent_name": start_agent.name,
"end_agent_name": end_agent.name,
"start_agent_goal": env.agent_goals[0],
"end_agent_goal": env.agent_goals[1],
}
)
count += 1
all_ids.add(f"{env._id}_{start_agent._id}_{end_agent._id}")
with open(output_dir, "w") as f:
f.write(json.dumps(result_pairs, indent=4))
if count >= num_episodes:
break
if __name__ == "__main__":
(
environments,
environment_dict,
agent_dict,
relationship_dict,
) = get_sotopia_profiles()
envs, start_agents, end_agents, social_interactions = load_sotopia_pi_data(
"../../data/sotopia_pi_episodes.jsonl", environment_dict, agent_dict
)
print("Loaded data with {} episodes".format(len(envs)))
generate_prompt_response_pairs(
"../../data/gpt35_gpt4_prompt_response_pairs.json",
["gpt-3.5-turbo", "gpt-4o"],
envs,
start_agents,
end_agents,
social_interactions,
100000,
)
|