mirror of
https://github.com/huggingface/lerobot.git
synced 2026-07-24 10:16:09 +00:00
rtc: exclude cold-start inference from latency tracker
This commit is contained in:
@@ -252,6 +252,8 @@ class RTCInferenceEngine(InferenceEngine):
|
|||||||
policy_device = torch.device(self._device)
|
policy_device = torch.device(self._device)
|
||||||
|
|
||||||
warmup_required = max(1, self._compile_warmup_inferences) if self._use_torch_compile else 0
|
warmup_required = max(1, self._compile_warmup_inferences) if self._use_torch_compile else 0
|
||||||
|
# exclude the first N inferences from the latency tracker to avoid cold-start spikes
|
||||||
|
latency_warmup_required = max(1, warmup_required)
|
||||||
inference_count = 0
|
inference_count = 0
|
||||||
consecutive_errors = 0
|
consecutive_errors = 0
|
||||||
|
|
||||||
@@ -316,7 +318,8 @@ class RTCInferenceEngine(InferenceEngine):
|
|||||||
inference_count += 1
|
inference_count += 1
|
||||||
consecutive_errors = 0
|
consecutive_errors = 0
|
||||||
is_warmup = self._use_torch_compile and inference_count <= warmup_required
|
is_warmup = self._use_torch_compile and inference_count <= warmup_required
|
||||||
if is_warmup:
|
# Ignore the first N inferences for latency tracking to avoid cold-start spikes
|
||||||
|
if inference_count <= latency_warmup_required:
|
||||||
latency_tracker.reset()
|
latency_tracker.reset()
|
||||||
else:
|
else:
|
||||||
latency_tracker.add(new_latency)
|
latency_tracker.add(new_latency)
|
||||||
|
|||||||
Reference in New Issue
Block a user