biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Misc] Remove experimental dep from tracing.py (#12007)

Browse Source 
Signed-off-by: Adrian Cole <adrian.cole@elastic.co>
This commit is contained in:
Adrian Cole
2025-01-21 11:51:55 -08:00
committed by GitHub
parent 18fd4a8331
commit 347eeebe3b
3 changed files with 66 additions and 60 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
vllm/engine/llm_engine.py
View File

@@ -1857,46 +1857,44 @@ class LLMEngine:
metrics = seq_group.metrics
ttft = metrics.first_token_time - metrics.arrival_time
e2e_time = metrics.finished_time - metrics.arrival_time
# attribute names are based on
# https://github.com/open-telemetry/semantic-conventions/blob/main/docs/gen-ai/llm-spans.md
seq_span.set_attribute(SpanAttributes.LLM_RESPONSE_MODEL,
seq_span.set_attribute(SpanAttributes.GEN_AI_RESPONSE_MODEL,
self.model_config.model)
seq_span.set_attribute(SpanAttributes.LLM_REQUEST_ID,
seq_span.set_attribute(SpanAttributes.GEN_AI_REQUEST_ID,
seq_group.request_id)
seq_span.set_attribute(SpanAttributes.LLM_REQUEST_TEMPERATURE,
seq_span.set_attribute(SpanAttributes.GEN_AI_REQUEST_TEMPERATURE,
seq_group.sampling_params.temperature)
seq_span.set_attribute(SpanAttributes.LLM_REQUEST_TOP_P,
seq_span.set_attribute(SpanAttributes.GEN_AI_REQUEST_TOP_P,
seq_group.sampling_params.top_p)
seq_span.set_attribute(SpanAttributes.LLM_REQUEST_MAX_TOKENS,
seq_span.set_attribute(SpanAttributes.GEN_AI_REQUEST_MAX_TOKENS,
seq_group.sampling_params.max_tokens)
seq_span.set_attribute(SpanAttributes.LLM_REQUEST_N,
seq_span.set_attribute(SpanAttributes.GEN_AI_REQUEST_N,
seq_group.sampling_params.n)
seq_span.set_attribute(SpanAttributes.LLM_USAGE_NUM_SEQUENCES,
seq_span.set_attribute(SpanAttributes.GEN_AI_USAGE_NUM_SEQUENCES,
seq_group.num_seqs())
seq_span.set_attribute(SpanAttributes.LLM_USAGE_PROMPT_TOKENS,
seq_span.set_attribute(SpanAttributes.GEN_AI_USAGE_PROMPT_TOKENS,
len(seq_group.prompt_token_ids))
seq_span.set_attribute(
SpanAttributes.LLM_USAGE_COMPLETION_TOKENS,
SpanAttributes.GEN_AI_USAGE_COMPLETION_TOKENS,
sum([
seq.get_output_len()
for seq in seq_group.get_finished_seqs()
]))
seq_span.set_attribute(SpanAttributes.LLM_LATENCY_TIME_IN_QUEUE,
seq_span.set_attribute(SpanAttributes.GEN_AI_LATENCY_TIME_IN_QUEUE,
metrics.time_in_queue)
seq_span.set_attribute(
SpanAttributes.LLM_LATENCY_TIME_TO_FIRST_TOKEN, ttft)
seq_span.set_attribute(SpanAttributes.LLM_LATENCY_E2E, e2e_time)
SpanAttributes.GEN_AI_LATENCY_TIME_TO_FIRST_TOKEN, ttft)
seq_span.set_attribute(SpanAttributes.GEN_AI_LATENCY_E2E, e2e_time)
if metrics.scheduler_time is not None:
seq_span.set_attribute(
SpanAttributes.LLM_LATENCY_TIME_IN_SCHEDULER,
SpanAttributes.GEN_AI_LATENCY_TIME_IN_SCHEDULER,
metrics.scheduler_time)
if metrics.model_forward_time is not None:
seq_span.set_attribute(
SpanAttributes.LLM_LATENCY_TIME_IN_MODEL_FORWARD,
SpanAttributes.GEN_AI_LATENCY_TIME_IN_MODEL_FORWARD,
metrics.model_forward_time / 1000.0)
if metrics.model_execute_time is not None:
seq_span.set_attribute(
SpanAttributes.LLM_LATENCY_TIME_IN_MODEL_EXECUTE,
SpanAttributes.GEN_AI_LATENCY_TIME_IN_MODEL_EXECUTE,
metrics.model_execute_time)
def _validate_model_inputs(self, inputs: ProcessorInputs,