Skip to main content

​
Loguru

We log to console and files using loguru using a global logger instance. Each entrypoint should call setup_logger exactly once at the beginning of execution. Afterwards, all components can log using the global logger instance. For more details on loguru, see the documentation. All logs are written into {output_dir}/logs and for RL training we recommend viewing logs by streaming the file logs into tmux panes, as set up by the tmux.sh script.

​
Torchrun

For multi-node training, we use torchrun to set up distributed training. Because torchrun is SPMD, all ranks are logging to console and file at the same time. To only view the logs from the master rank, you can use the --local-ranks-filter flag. For example, to only view the logs from the master rank when training on a full node 
uv run torchrun \
  --local-ranks-filter 0 \
  --nproc-per-node 8 \
  ...
In addition to the loguru file logs you can also use the --log-dir and --redirects and --tee flags to redirect the console logs to files. 
uv run torchrun \
  --local-ranks-filter 0 \
  --nproc-per-node 8 \
  --log-dir outputs/torchrun \
  --redirects 3 \
  --tee 3 \
  ...
This will redirect the console logs to outputs/torchrun/{rdzv_id}/attempt_0/{rank}/{stdout,stderr}.log.

​
W&B

For most runs we recommend logging to W&B. Before enabling W&B, make sure that you have an account and are logged in. 
uv run wandb login
# Or set `export WANDB_API_KEY=...`

​
SFT

Logging to W&B is disabled by default. Enable the default configuration with --wandb 
uv run sft ... --wandb
This will log to the prime-rl project with a random run name. You can specify which project and name to log to 
uv run sft ... --wandb.project my-project --wandb.name my-run
The same settings also work for multi-node training with torchrun. Note, that we only log global metrics from the master rank (e.g. the all-reduced loss) 
uv run torchrun --nproc-per-node 8 ...  --wandb

​
RL

For RL training, both the trainer and orchestrator log to W&B as separate runs. Again, logging to W&B is disabled by default. Enable the default configuration with --wandb 
uv run rl ... --wandb
This will log to the prime-rl project with a random run name. The trainer run is suffixed with -trainer and the orchestrator run is suffixed with -orchestrator. You can specify which project and name to log to using the same flags as for SFT. 
uv run rl ... --wandb.project my-project --wandb.name my-run
For the RL trainer, we support logging samples (e.g. prompt, completion, reward, advantage for selected rollouts) and distributions (e.g. reward, advantage, entropy distributions) as W&B tables using the wandb.log-extras subconfig. If W&B is setup, this is enabled by default and will log for the RL trainer and orchestrator every 10 steps. You can configure this on the trainer and orchestrator separately. For example, to only log samples on the orchestrator every 50 steps, but not distribution on either 
uv run rl  ... \
  --no-trainer.wandb.log-extras.distributions \
  --orchestrator.wandb.log-extras.interval 50