Fimeg/letta-server
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

chore: move ttft logging for new agent loop (#2465)

Browse Source
This commit is contained in:
cthomas
2025-05-27 13:50:41 -07:00
committed by GitHub
parent cc702025bc
commit b17b2dc31e
1 changed files with 4 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
letta/agents/letta_agent.py
View File

@@ -314,6 +314,10 @@ class LettaAgent(BaseAgent):
actor=self.actor,
)
usage = LettaUsageStatistics()
first_chunk, ttft_span = True, None
if request_start_timestamp_ns is not None:
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})
for _ in range(max_steps):
step_id = generate_step_id()
@@ -354,11 +358,6 @@ class LettaAgent(BaseAgent):
else:
raise ValueError(f"Streaming not supported for {agent_state.llm_config}")
first_chunk, ttft_span = True, None
if request_start_timestamp_ns is not None:
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})
async for chunk in interface.process(stream):
# Measure time to first token
if first_chunk and ttft_span is not None: