apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-06 17:17:14 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[feat] add metrics for yiyan adapter (#3219)

Browse Source 
* [feat] add metrics for yiyan adapter

* [fix] fix metrics num_requests_waiting and num_requests_running

* [fix] fix metrics gpu_cache_usage_perc

* [refactor] change where requests_number increases

* [chore] rename xxx_block_num as xxx_gpu_block_num, and update their values accordingly

* [chore] delete useless code
This commit is contained in:
李泳桦
2025-08-21 16:58:10 +08:00
committed by GitHub
parent 6854506533
commit d18a637a17
7 changed files with 181 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
fastdeploy/output/token_processor.py
View File

@@ -283,6 +283,15 @@ class TokenProcessor:
self.resource_manager.stop_flags[index] = True
self.resource_manager.tasks_list[index] = None
self.resource_manager._recycle_block_tables(task)
task_used_block_num = sum([len(task.block_tables) if task else 0 for task in self.resource_manager.tasks_list])
main_process_metrics.available_gpu_block_num.set(
self.resource_manager.total_block_number() - task_used_block_num
)
main_process_metrics.batch_size.set(
self.resource_manager.max_num_seqs - self.resource_manager.available_batch()
)
if task_id in self.tokens_counter:
del self.tokens_counter[task_id]
@@ -574,6 +583,7 @@ class TokenProcessor:
def _record_first_token_metrics(self, task, current_time):
"""Record metrics for first token"""
task.first_token_time = current_time
main_process_metrics.first_token_latency.set(current_time - task.inference_start_time)
main_process_metrics.time_to_first_token.observe(current_time - task.inference_start_time)
main_process_metrics.request_queue_time.observe(task.schedule_start_time - task.preprocess_end_time)
@@ -585,6 +595,7 @@ class TokenProcessor:
main_process_metrics.num_requests_running.dec(1)
main_process_metrics.request_success_total.inc()
main_process_metrics.infer_latency.set(current_time - task.inference_start_time)
main_process_metrics.request_inference_time.observe(current_time - task.inference_start_time)
main_process_metrics.request_generation_tokens.observe(self.tokens_counter[task.request_id])