File size: 4,092 Bytes
301a0ac |
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 |
import asyncio
from python.helpers.extension import Extension
from python.helpers.memory import Memory
from agent import LoopData
DATA_NAME_TASK = "_recall_solutions_task"
class RecallSolutions(Extension):
INTERVAL = 3
HISTORY = 10000
SOLUTIONS_COUNT = 2
INSTRUMENTS_COUNT = 2
THRESHOLD = 0.6
async def execute(self, loop_data: LoopData = LoopData(), **kwargs):
# every 3 iterations (or the first one) recall memories
if loop_data.iteration % RecallSolutions.INTERVAL == 0:
task = asyncio.create_task(self.search_solutions(loop_data=loop_data, **kwargs))
else:
task = None
# set to agent to be able to wait for it
self.agent.set_data(DATA_NAME_TASK, task)
async def search_solutions(self, loop_data: LoopData, **kwargs):
#cleanup
extras = loop_data.extras_persistent
if "solutions" in extras:
del extras["solutions"]
# try:
# show temp info message
self.agent.context.log.log(
type="info", content="Searching memory for solutions...", temp=True
)
# show full util message, this will hide temp message immediately if turned on
log_item = self.agent.context.log.log(
type="util",
heading="Searching memory for solutions...",
)
# get system message and chat history for util llm
# msgs_text = self.agent.concat_messages(
# self.agent.history[-RecallSolutions.HISTORY :]
# ) # only last X messages
# msgs_text = self.agent.history.current.output_text()
msgs_text = self.agent.history.output_text()[-RecallSolutions.HISTORY:]
system = self.agent.read_prompt(
"memory.solutions_query.sys.md", history=msgs_text
)
# log query streamed by LLM
async def log_callback(content):
log_item.stream(query=content)
# call util llm to summarize conversation
query = await self.agent.call_utility_model(
system=system, message=loop_data.user_message.output_text() if loop_data.user_message else "", callback=log_callback
)
# get solutions database
db = await Memory.get(self.agent)
solutions = await db.search_similarity_threshold(
query=query,
limit=RecallSolutions.SOLUTIONS_COUNT,
threshold=RecallSolutions.THRESHOLD,
filter=f"area == '{Memory.Area.SOLUTIONS.value}'",
)
instruments = await db.search_similarity_threshold(
query=query,
limit=RecallSolutions.INSTRUMENTS_COUNT,
threshold=RecallSolutions.THRESHOLD,
filter=f"area == '{Memory.Area.INSTRUMENTS.value}'",
)
log_item.update(
heading=f"{len(instruments)} instruments, {len(solutions)} solutions found",
)
if instruments:
instruments_text = ""
for instrument in instruments:
instruments_text += instrument.page_content + "\n\n"
instruments_text = instruments_text.strip()
log_item.update(instruments=instruments_text)
instruments_prompt = self.agent.read_prompt(
"agent.system.instruments.md", instruments=instruments_text
)
loop_data.system.append(instruments_prompt)
if solutions:
solutions_text = ""
for solution in solutions:
solutions_text += solution.page_content + "\n\n"
solutions_text = solutions_text.strip()
log_item.update(solutions=solutions_text)
solutions_prompt = self.agent.parse_prompt(
"agent.system.solutions.md", solutions=solutions_text
)
# append to prompt
extras["solutions"] = solutions_prompt
# except Exception as e:
# err = errors.format_error(e)
# self.agent.context.log.log(
# type="error", heading="Recall solutions extension error:", content=err
# )
|