github/MemGPT
7
0
Fork 0
mirror of https://github.com/cpacker/MemGPT.git synced 2025-06-03 04:30:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

merge conflict 2

Browse Source
This commit is contained in:
Caren Thomas 2025-05-21 16:28:54 -07:00
parent 4d8a3eacdf
commit 06f7e5e741
1 changed files with 0 additions and 40 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

40
letta/agents/letta_agent.py
View File

@ -202,10 +202,6 @@ class LettaAgent(BaseAgent):
3. Fetches a response from the LLM 3. Fetches a response from the LLM
4. Processes the response 4. Processes the response
""" """
<<<<<<< HEAD
=======
agent_state = await self.agent_manager.get_agent_by_id_async(self.agent_id, actor=self.actor)
>>>>>>> main
current_in_context_messages, new_in_context_messages = await _prepare_in_context_messages_async( current_in_context_messages, new_in_context_messages = await _prepare_in_context_messages_async(
input_messages, agent_state, self.message_manager, self.actor input_messages, agent_state, self.message_manager, self.actor
) )
@ -216,10 +212,6 @@ class LettaAgent(BaseAgent):
actor=self.actor, actor=self.actor,
) )
usage = LettaUsageStatistics() usage = LettaUsageStatistics()
<<<<<<< HEAD
=======
>>>>>>> main
for _ in range(max_steps): for _ in range(max_steps):
step_id = generate_step_id() step_id = generate_step_id()
@ -238,7 +230,6 @@ class LettaAgent(BaseAgent):
tool_rules_solver=tool_rules_solver, tool_rules_solver=tool_rules_solver,
# TODO: pass in reasoning content # TODO: pass in reasoning content
) )
<<<<<<< HEAD
log_event("agent.step.llm_request.created") # [2^] log_event("agent.step.llm_request.created") # [2^]
try: try:
@ -352,8 +343,6 @@ class LettaAgent(BaseAgent):
raise llm_client.handle_llm_error(e) raise llm_client.handle_llm_error(e)
log_event("agent.stream.llm_response.received") # [3^] log_event("agent.stream.llm_response.received") # [3^]
=======
>>>>>>> main
# TODO: THIS IS INCREDIBLY UGLY # TODO: THIS IS INCREDIBLY UGLY
# TODO: THERE ARE MULTIPLE COPIES OF THE LLM_CONFIG EVERYWHERE THAT ARE GETTING MANIPULATED # TODO: THERE ARE MULTIPLE COPIES OF THE LLM_CONFIG EVERYWHERE THAT ARE GETTING MANIPULATED
if agent_state.llm_config.model_endpoint_type == "anthropic": if agent_state.llm_config.model_endpoint_type == "anthropic":
@ -366,7 +355,6 @@ class LettaAgent(BaseAgent):
use_assistant_message=use_assistant_message, use_assistant_message=use_assistant_message,
put_inner_thoughts_in_kwarg=agent_state.llm_config.put_inner_thoughts_in_kwargs, put_inner_thoughts_in_kwarg=agent_state.llm_config.put_inner_thoughts_in_kwargs,
) )
<<<<<<< HEAD
else: else:
raise ValueError(f"Streaming not supported for {agent_state.llm_config}") raise ValueError(f"Streaming not supported for {agent_state.llm_config}")
@ -375,8 +363,6 @@ class LettaAgent(BaseAgent):
ttft_span = tracer.start_span("time_to_first_token", start_time=request_start_timestamp_ns) ttft_span = tracer.start_span("time_to_first_token", start_time=request_start_timestamp_ns)
ttft_span.set_attributes({f"llm_config.{k}": v for k, v in agent_state.llm_config.model_dump().items() if v is not None}) ttft_span.set_attributes({f"llm_config.{k}": v for k, v in agent_state.llm_config.model_dump().items() if v is not None})
=======
>>>>>>> main
async for chunk in interface.process(stream): async for chunk in interface.process(stream):
# Measure time to first token # Measure time to first token
if first_chunk and ttft_span is not None: if first_chunk and ttft_span is not None:
@ -414,7 +400,6 @@ class LettaAgent(BaseAgent):
self.response_messages.extend(persisted_messages) self.response_messages.extend(persisted_messages)
new_in_context_messages.extend(persisted_messages) new_in_context_messages.extend(persisted_messages)
<<<<<<< HEAD
# TODO (cliandy): the stream POST request span has ended at this point, we should tie this to the stream # TODO (cliandy): the stream POST request span has ended at this point, we should tie this to the stream
# log_event("agent.stream.llm_response.processed") # [4^] # log_event("agent.stream.llm_response.processed") # [4^]
@ -442,8 +427,6 @@ class LettaAgent(BaseAgent):
), ),
) )
=======
>>>>>>> main
if not use_assistant_message or should_continue: if not use_assistant_message or should_continue:
tool_return = [msg for msg in persisted_messages if msg.role == "tool"][-1].to_letta_messages()[0] tool_return = [msg for msg in persisted_messages if msg.role == "tool"][-1].to_letta_messages()[0]
yield f"data: {tool_return.model_dump_json()}\n\n" yield f"data: {tool_return.model_dump_json()}\n\n"
@ -473,7 +456,6 @@ class LettaAgent(BaseAgent):
agent_state: AgentState, agent_state: AgentState,
tool_rules_solver: ToolRulesSolver, tool_rules_solver: ToolRulesSolver,
) -> ChatCompletion | AsyncStream[ChatCompletionChunk]: ) -> ChatCompletion | AsyncStream[ChatCompletionChunk]:
<<<<<<< HEAD
self.num_messages = self.num_messages or (await self.message_manager.size_async(actor=self.actor, agent_id=agent_state.id)) self.num_messages = self.num_messages or (await self.message_manager.size_async(actor=self.actor, agent_id=agent_state.id))
self.num_archival_memories = self.num_archival_memories or ( self.num_archival_memories = self.num_archival_memories or (
await self.passage_manager.size_async(actor=self.actor, agent_id=agent_state.id) await self.passage_manager.size_async(actor=self.actor, agent_id=agent_state.id)
@ -481,21 +463,6 @@ class LettaAgent(BaseAgent):
in_context_messages = await self._rebuild_memory_async( in_context_messages = await self._rebuild_memory_async(
in_context_messages, agent_state, num_messages=self.num_messages, num_archival_memories=self.num_archival_memories in_context_messages, agent_state, num_messages=self.num_messages, num_archival_memories=self.num_archival_memories
) )
=======
if settings.experimental_enable_async_db_engine:
self.num_messages = self.num_messages or (await self.message_manager.size_async(actor=self.actor, agent_id=agent_state.id))
self.num_archival_memories = self.num_archival_memories or (
await self.passage_manager.size_async(actor=self.actor, agent_id=agent_state.id)
)
in_context_messages = await self._rebuild_memory_async(
in_context_messages, agent_state, num_messages=self.num_messages, num_archival_memories=self.num_archival_memories
)
else:
if settings.experimental_skip_rebuild_memory and agent_state.llm_config.model_endpoint_type == "google_vertex":
logger.info("Skipping memory rebuild")
else:
in_context_messages = self._rebuild_memory(in_context_messages, agent_state)
>>>>>>> main
tools = [ tools = [
t t
@ -625,10 +592,6 @@ class LettaAgent(BaseAgent):
pre_computed_tool_message_id=pre_computed_tool_message_id, pre_computed_tool_message_id=pre_computed_tool_message_id,
step_id=logged_step.id if logged_step else None, # TODO (cliandy): eventually move over other agent loops step_id=logged_step.id if logged_step else None, # TODO (cliandy): eventually move over other agent loops
) )
<<<<<<< HEAD
=======
>>>>>>> main
persisted_messages = await self.message_manager.create_many_messages_async(tool_call_messages, actor=self.actor) persisted_messages = await self.message_manager.create_many_messages_async(tool_call_messages, actor=self.actor)
self.last_function_response = function_response self.last_function_response = function_response
@ -716,10 +679,7 @@ class LettaAgent(BaseAgent):
results = await asyncio.gather(*tasks) results = await asyncio.gather(*tasks)
return results return results
<<<<<<< HEAD
@trace_method @trace_method
=======
>>>>>>> main
async def _load_last_function_response_async(self): async def _load_last_function_response_async(self):
"""Load the last function response from message history""" """Load the last function response from message history"""
in_context_messages = await self.agent_manager.get_in_context_messages_async(agent_id=self.agent_id, actor=self.actor) in_context_messages = await self.agent_manager.get_in_context_messages_async(agent_id=self.agent_id, actor=self.actor)