File size: 10,724 Bytes
3d3d712 |
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 json import JSONDecodeError
from typing import List, Optional
from injector import inject
from taskweaver.config.module_config import ModuleConfig
from taskweaver.llm import LLMApi
from taskweaver.llm.util import ChatMessageType, format_chat_message
from taskweaver.logging import TelemetryLogger
from taskweaver.memory import Attachment, Conversation, Memory, Post, Round, RoundCompressor
from taskweaver.memory.attachment import AttachmentType
from taskweaver.memory.plugin import PluginRegistry
from taskweaver.misc.example import load_examples
from taskweaver.role import PostTranslator, Role
from taskweaver.utils import read_yaml
class PlannerConfig(ModuleConfig):
def _configure(self) -> None:
self._set_name("planner")
app_dir = self.src.app_base_path
self.use_example = self._get_bool("use_example", True)
self.prompt_file_path = self._get_path(
"prompt_file_path",
os.path.join(
os.path.dirname(os.path.abspath(__file__)),
"planner_prompt.yaml",
),
)
self.example_base_path = self._get_path(
"example_base_path",
os.path.join(
app_dir,
"planner_examples",
),
)
self.prompt_compression = self._get_bool("prompt_compression", False)
self.compression_prompt_path = self._get_path(
"compression_prompt_path",
os.path.join(
os.path.dirname(os.path.abspath(__file__)),
"compression_prompt.yaml",
),
)
self.skip_planning = self._get_bool("skip_planning", False)
with open(
os.path.join(
os.path.dirname(os.path.abspath(__file__)),
"dummy_plan.json",
),
"r",
) as f:
self.dummy_plan = json.load(f)
class Planner(Role):
conversation_delimiter_message: str = "Let's start the new conversation!"
ROLE_NAME: str = "Planner"
@inject
def __init__(
self,
config: PlannerConfig,
logger: TelemetryLogger,
llm_api: LLMApi,
plugin_registry: PluginRegistry,
round_compressor: Optional[RoundCompressor] = None,
plugin_only: bool = False,
):
self.config = config
self.logger = logger
self.llm_api = llm_api
if plugin_only:
self.available_plugins = [p for p in plugin_registry.get_list() if p.plugin_only is True]
else:
self.available_plugins = plugin_registry.get_list()
self.planner_post_translator = PostTranslator(logger)
self.prompt_data = read_yaml(self.config.prompt_file_path)
if self.config.use_example:
self.examples = self.get_examples()
if len(self.available_plugins) == 0:
self.logger.warning("No plugin is loaded for Planner.")
self.plugin_description = "No plugin functions loaded."
else:
self.plugin_description = "\t" + "\n\t".join(
[f"- {plugin.name}: " + f"{plugin.spec.description}" for plugin in self.available_plugins],
)
self.instruction_template = self.prompt_data["instruction_template"]
self.code_interpreter_introduction = self.prompt_data["code_interpreter_introduction"].format(
plugin_description=self.plugin_description,
)
self.response_schema = self.prompt_data["planner_response_schema"]
self.instruction = self.instruction_template.format(
planner_response_schema=self.response_schema,
CI_introduction=self.code_interpreter_introduction,
)
self.ask_self_cnt = 0
self.max_self_ask_num = 3
self.round_compressor = round_compressor
self.compression_template = read_yaml(self.config.compression_prompt_path)["content"]
self.logger.info("Planner initialized successfully")
def compose_conversation_for_prompt(
self,
conv_rounds: List[Round],
summary: Optional[str] = None,
) -> List[ChatMessageType]:
conversation: List[ChatMessageType] = []
for rnd_idx, chat_round in enumerate(conv_rounds):
conv_init_message = None
if rnd_idx == 0:
conv_init_message = Planner.conversation_delimiter_message
if summary is not None:
self.logger.debug(f"Summary: {summary}")
summary_message = (
f"\nThe context summary of the Planner's previous rounds" f" can refer to:\n{summary}\n\n"
)
conv_init_message += "\n" + summary_message
for post in chat_round.post_list:
if post.send_from == "Planner":
if post.send_to == "User" or post.send_to == "CodeInterpreter":
planner_message = self.planner_post_translator.post_to_raw_text(
post=post,
)
conversation.append(
format_chat_message(
role="assistant",
message=planner_message,
),
)
elif (
post.send_to == "Planner"
): # self correction for planner response, e.g., format error/field check error
conversation.append(
format_chat_message(
role="assistant",
message=post.get_attachment(type=AttachmentType.invalid_response)[0],
),
) # append the invalid response to chat history
conversation.append(
format_chat_message(role="user", message="User: " + post.message),
) # append the self correction instruction message to chat history
else:
if conv_init_message is not None:
message = post.send_from + ": " + conv_init_message + "\n" + post.message
conversation.append(
format_chat_message(role="user", message=message),
)
conv_init_message = None
else:
conversation.append(
format_chat_message(role="user", message=post.send_from + ": " + post.message),
)
return conversation
def compose_prompt(self, rounds: List[Round]) -> List[ChatMessageType]:
chat_history = [format_chat_message(role="system", message=self.instruction)]
if self.config.use_example and len(self.examples) != 0:
for conv_example in self.examples:
conv_example_in_prompt = self.compose_conversation_for_prompt(conv_example.rounds)
chat_history += conv_example_in_prompt
summary = None
if self.config.prompt_compression and self.round_compressor is not None:
summary, rounds = self.round_compressor.compress_rounds(
rounds,
rounds_formatter=lambda _rounds: str(self.compose_conversation_for_prompt(_rounds)),
use_back_up_engine=True,
prompt_template=self.compression_template,
)
chat_history.extend(
self.compose_conversation_for_prompt(
rounds,
summary=summary,
),
)
return chat_history
def reply(
self,
memory: Memory,
event_handler,
prompt_log_path: Optional[str] = None,
use_back_up_engine: bool = False,
) -> Post:
rounds = memory.get_role_rounds(role="Planner")
assert len(rounds) != 0, "No chat rounds found for planner"
chat_history = self.compose_prompt(rounds)
def check_post_validity(post: Post):
assert post.send_to is not None, "send_to field is None"
assert post.send_to != "Planner", "send_to field should not be Planner"
assert post.message is not None, "message field is None"
assert post.attachment_list[0].type == AttachmentType.init_plan, "attachment type is not init_plan"
assert post.attachment_list[1].type == AttachmentType.plan, "attachment type is not plan"
assert (
post.attachment_list[2].type == AttachmentType.current_plan_step
), "attachment type is not current_plan_step"
if self.config.skip_planning and rounds[-1].post_list[-1].send_from == "User":
self.config.dummy_plan["response"][0]["content"] += rounds[-1].post_list[-1].message
llm_output = json.dumps(self.config.dummy_plan)
else:
llm_output = self.llm_api.chat_completion(chat_history, use_backup_engine=use_back_up_engine)["content"]
try:
response_post = self.planner_post_translator.raw_text_to_post(
llm_output=llm_output,
send_from="Planner",
event_handler=event_handler,
validation_func=check_post_validity,
)
if response_post.send_to == "User":
event_handler("final_reply_message", response_post.message)
except (JSONDecodeError, AssertionError) as e:
self.logger.error(f"Failed to parse LLM output due to {str(e)}")
response_post = Post.create(
message=f"Failed to parse Planner output due to {str(e)}."
f"The output format should follow the below format:"
f"{self.prompt_data['planner_response_schema']}"
"Please try to regenerate the output.",
send_to="Planner",
send_from="Planner",
attachment_list=[Attachment.create(type=AttachmentType.invalid_response, content=llm_output)],
)
self.ask_self_cnt += 1
if self.ask_self_cnt > self.max_self_ask_num: # if ask self too many times, return error message
self.ask_self_cnt = 0
raise Exception(f"Planner failed to generate response because {str(e)}")
if prompt_log_path is not None:
self.logger.dump_log_file(chat_history, prompt_log_path)
return response_post
def get_examples(self) -> List[Conversation]:
example_conv_list = load_examples(self.config.example_base_path)
return example_conv_list
|