NVIDIA-NeMo / NVIDIA-NeMo/RL

data_processing metric is incorrect

Open
#955 1 comment 0 reactions 0 assignees View on GitHub
bug Performance research
Dominant language
Python
Stars
2k
Forks
561
Avg merge
4d 5h
Merged PRs (30d)
145

Description

**Describe the bug**

We see things like this reported by nemo-rl

========================= Step 2/100 =========================
▶ Preparing batch...
▶ Taking a training step...

📊 Training Results:
• Loss: 1.4154
• Training FLOPS: 2709.39 TFLOPS (2709.39 TFLOPS per rank)
• Training Model Floating Point Utilization: 273.81%

⏱️ Timing:
• Total step time: 158.50s
• policy_training: 157.59s (99.4%)
• data_processing: 0.90s (0.6%)

Yet further analysis reveals that there are ~50 seconds of extra data processing time that's not captured in metrics.

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.