Update experiment_issues.md

Author: mahdis-saeedi

docs/e2e_journal_experiment/experiment_issues.md

@@ -1,7 +1,7 @@
 Experiments on different [variations of fnn](https://docs.google.com/spreadsheets/d/1jt4Pvdz58qs0LyAnSjYr0MTpzy5ZfNonw5bBtfbIvKY/edit?gid=212563191#gid=212563191) with mhot vector as an input:
 **training phase only**
 - We initially ran 3 models on cuda:1 and 3 models on cuda:3, resulting in a training time of ~3 hours per epoch.
-- Running 6 models in parallel led to CPU usage exceeding 97% across all 224 cores.
+- Running 6 models in parallel led to CPU usage [exceeding 97%](https://github.com/mahdis-saeedi/OpeNTF/blob/main/docs/e2e_journal_experiment/cpu%26gpu_usage/cpu_6models_2gpus.txt) across all 224 cores.
 - We terminated 3 runs on cuda:3, which reduced CPU usage to ~7% per core, but it was not stable and increased to more than 95% again.
 - We then launched a new run on cuda:3, increasing the batch size from 1,000 to 10,000, and started monitoring GPU and CPU utilization.
 As a result: