failed Pipeline #30 triggered by Administrator's avatar Administrator

Apply dropout scaling to dQ and dK instead of to V (in bwd)

Theoretically this might have lower numerical error since the scaling is in
fp32 instead of fp16 (not sure, I haven't thought too carefully about it).
However, in practice, the numerical errors seem about the same.
7 jobs for 5b838a8b in 0 seconds (queued for 81 minutes and 14 seconds)
Status Name Job ID Coverage
  Build
failed build #203

 
  Test
skipped bandit-sast #206
allowed to fail
failed code_quality #205
allowed to fail

skipped flawfinder-sast #207
allowed to fail
skipped secret_detection #209
allowed to fail
skipped semgrep-sast #208
allowed to fail
failed test #204

 
Name Stage Failure
failed
build Build There has been a timeout failure or the job got stuck. Check your timeout limits or try again
No job log
failed
test Test There has been a timeout failure or the job got stuck. Check your timeout limits or try again
No job log
failed
code_quality Test There has been a timeout failure or the job got stuck. Check your timeout limits or try again
No job log

Speed up your pipelines with Needs relationships

Using the needs keyword makes jobs run before their stage is reached. Jobs run as soon as their needs relationships are met, which speeds up your pipelines.

If you add needs to jobs in your pipeline you'll be able to view the needs relationships between jobs in this tab as a Directed Acyclic Graph (DAG).

There are no test reports for this pipeline

You can configure your job to use unit test reports, and GitLab displays a report here and in the related merge request.