Task description: "Collect all loss values into CSV files from WandB and -- if needed -- log files".
The most important file is pythia_runs.tsv in which I manually collect the run references to use in the analysis. Since I do not have visibility on how the model was trained, I am best guessing which runs are the good ones, from the many present on Wandb, for each model size and seed.
Task description: "Collect all loss values into CSV files from WandB and -- if needed -- log files".
The most important file is
pythia_runs.tsv
in which I manually collect the run references to use in the analysis. Since I do not have visibility on how the model was trained, I am best guessing which runs are the good ones, from the many present on Wandb, for each model size and seed.CC @oskarvanderwal @haileyschoelkopf