admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-05-27 14:39:43 +00:00
Code Issues Packages Projects Releases Wiki Activity

wandb: flip training-example logging defaults to on (every 5000 steps)

Browse Source 
The training-example wandb.Table dump (camera images + text fields +
GT/predicted action chunk endpoints) was opt-in. Flip defaults so any
run with --wandb.enable=true gets visual training observability for free.

  log_examples_freq:           0     -> 5000   (push table every 5k steps)
  log_examples_n:              4     -> 4      (unchanged)
  log_examples_predict_actions: False -> True   (extra forward in eval mode)

Runs without --wandb.enable=true are unaffected (the training loop gate
checks wandb_logger is not None first). Set log_examples_freq=0 to opt
out of the dump even with wandb enabled; set log_examples_predict_actions
=false to skip the extra inference forward pass.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Pepijn
2026-05-25 18:00:04 +02:00
parent b1e83f556c
commit 057c794ffe
1 changed files with 10 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/lerobot/configs/default.py
+10 -8
View File
@@ -67,16 +67,18 @@ class WandBConfig:
# with one row per sampled batch element containing each camera view # with one row per sampled batch element containing each camera view
# (rendered as ``wandb.Image``), any text fields present in the batch # (rendered as ``wandb.Image``), any text fields present in the batch
# (``task`` / ``subtask`` / ``memory`` / ``instruction``), and the # (``task`` / ``subtask`` / ``memory`` / ``instruction``), and the
# ground-truth action chunk's first + last frames. 0 disables — recommended # ground-truth action chunk's first + last frames. Defaults to 5000 — set
# starting value is 5000 for long runs. # to 0 to disable. Only fires when ``enable=True``, so runs without wandb
log_examples_freq: int = 0 # are unaffected.
log_examples_freq: int = 5000
# Number of batch elements to include in each example dump. # Number of batch elements to include in each example dump.
log_examples_n: int = 4 log_examples_n: int = 4
# If True, also run ``policy.predict_action_chunk`` on the logged samples # If True (default), also run ``policy.predict_action_chunk`` on the logged
# (in eval mode, no_grad) and add predicted vs ground-truth action columns # samples (in eval mode, no_grad) and add predicted vs ground-truth action
# to the table. Costs one extra forward pass per dump — negligible at # columns to the table. Costs one extra forward pass per dump — negligible
# 5k-step cadence. # at the 5k-step default cadence. Set to ``False`` if your policy doesn't
log_examples_predict_actions: bool = False # implement ``predict_action_chunk`` or you want to skip the extra forward.
log_examples_predict_actions: bool = True
@dataclass @dataclass