-
Hi,
I use layernorm and rmsnorm in my training pipeline on an A100 and observed via the pytorch profiler that these functions were quite slow.
E.g. I measured via time.time() just for the rmsnorm:
…
-
**Describe the bug**
For ZeRO-3, i'm noticing an increase in training times on g5.48xlarge nodes with torch >= 2.3.1 and CUDA 12.1. I can reproduce this with small and large models, and in some cases…
-
### Describe the bug
Hi, Andrei
We recently encountered a crash. After analysis, I think this problem may occur on OpenJDK 8, 11, and 17. It will not occur on OpenJDK 18 and above because [JDK-8…
-
Hello,
I am CPU profiling 32bit c++ server application on x64 machine running Centos 6.4. I linked my main module (-lprofiler) with libprofiler 0.4.8 which is the latest one if I am not mistaken. …
-
**System information**
- Google Pixel 7 / Android 14 / Google Tensor G2
- TensorFlow 2.16.1
Running a simple model with a Add + Mul on both GPU and CPU gives vastly different results (on device C…
-
The cudf.pandas profiler provides insight into which operations ran on the GPU and which ran on the CPU (and how many times), but it doesn't record the number of CPU/GPU transfers (and how long they t…
-
cli.TestTenantZip [failed](https://tanzanite.cluster.engflow.com/invocations/default/49698918-f71e-4c77-8c2c-abf16967832c?testReportRun=1&testReportShard=13&testReportAttempt=1#targets-Ly9wa2cvY2xpOmN…
-
Aug 23 19:48:42 loris systemd[1]: Finished Record System Boot/Shutdown in UTMP.
Aug 23 19:48:42 loris systemd[1]: Started Network Time Synchronization.
Aug 23 19:48:42 loris systemd[1]: Started User…
-
let _p_guard = pprof::ProfilerGuardBuilder::default()
.frequency(1000)
.blocklist(&["libc", "libgcc", "pthread", "vdso"])
.build()
.unwrap();
Execute th…
-
After deploying Qwen-4B, Qwen-1.8B, and Qwen-0.5B to an Android phone using mlc-llm, I used the following commands to check the system memory usage:
```
adb shell ps | findstr "mlc"
adb shell dmab…