Nicolas Rabault 838ab9e234 feat(train): resume training from a Hub checkpoint ...

Allow --config_path to be a Hub repo id when resuming, not only a local path.
The latest checkpoint under checkpoints/<step>/ is downloaded into a fresh local
run dir and resumed from there (optimizer, scheduler, RNG and data order
restored as for a local resume). TrainPipelineConfig.from_pretrained falls back
to the latest checkpoint's train_config.json when a repo has no root config
(an interrupted run that only pushed checkpoints). The download is skipped when
dispatching remotely so the executor (local machine or HF Jobs pod) performs it.

- add find_latest_hub_checkpoint (utils/hub) and resolve_resume_checkpoint
  (common/train_utils), the symmetric download counterpart to
  push_checkpoint_to_hub
- unit tests for both helpers and the from_pretrained fallback

2026-06-25 21:50:39 +02:00

..

test_collate.py

Add extensive language support (#3467)

2026-05-19 14:46:11 +02:00

test_encoding_utils.py

chore(utils): move encoding utils and process to their respective modules (#2029)

2025-09-24 16:47:37 +02:00

test_hub.py

feat(train): resume training from a Hub checkpoint

2026-06-25 21:50:39 +02:00

test_io_utils.py

chore(docs): add missing license headers (#2140)

2025-10-08 14:27:52 +02:00

test_logging_utils.py

feat(training): bump accelerate + use reduction types for tracked metrics in a multi rank setup (#3773)

2026-06-11 19:07:28 +02:00

test_process.py

RL stack refactoring (#3075)

2026-05-12 15:49:54 +02:00

test_random_utils.py

chore(docs): add missing license headers (#2140)

2025-10-08 14:27:52 +02:00

test_replay_buffer.py

RL stack refactoring (#3075)

2026-05-12 15:49:54 +02:00

test_sample_weighting.py

Reward models refactor (#3142)

2026-04-28 17:56:24 +02:00

test_train_utils.py

feat(train): resume training from a Hub checkpoint

2026-06-25 21:50:39 +02:00

test_visualization_utils.py

feat(dependencies): minimal default tag install (#3362)

2026-04-12 20:03:04 +02:00