Viacheslav Kukushkin
Viacheslav Kukushkin
Fixes #610. ## Proposed Changes * [x] **[base PR]** replaced `gandlf_*` scripts in favor of entrypoints. Thus, user is still able to run `gandlf_constructCSV`, but now it can be done...
### Is your feature request related to a problem? Please describe. Now running all the tests takes around ~30 mins. I understand we run a few training loops (on CPU)...
### Is your feature request related to a problem? Please describe. Tensorboard allows to monitor training in a handy way - user-friendly web UI shows loss / metrics / confusion...
If gandlf config cannot be parsed properly (typos, wrong values, etc), now the error is just suppressed without any meaningful error and `gandlf_config` variable is just broken. Fix ensures 1....
Here is restored old file about leader selection. It needs review and comments to be changed accordingly to other documentation style
An ancestor of https://github.com/mlcommons/medperf/pull/550 (branch was renamed) ==== The goal is to make medperf client installable from pypi. This PR is the first part that allows installing package not from...
- now config.yaml location can be defined with custom value (env var) - integration tests use a new /tmp/ folder storage with every run - After every command db is...
[In some places](https://github.com/mlcommons/medperf/blob/main/cli/medperf/entities/dataset.py#L93) we are using `{}` as default function param values. However this is error-prone as these default values are mutable and can be modified unexpectedly during the call:...
Replacement of https://github.com/hasan7n/medperf/pull/3 as that base branch was already merged into main While reading output from spawned mlcube process, stop&display not only when a full line (ending with \n) is...
If you are a benchmark owner, you have to implement three cubes: data_prep, model and evaluation. The issue is that you cannot run (or implement) them one-by-one: if the benchmark...