[codex] prune raw train rollout payloads#2839
Draft
samsja wants to merge 1 commit into
Draft
Conversation
3455a57 to
1182bcc
Compare
bd1b10a to
82e4eff
Compare
1182bcc to
11b33d5
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Prunes duplicate raw train rollout payloads after trainer samples are built:
TrainingSampleconstructionTrainingSamplepayloads intactAblation
Same fast no-inference routed-payload run used for the memory stack:
5 steps,batch=64,inflight=128,seq_len=32768,turns=50,max_off_policy=4, delay mean0s, std3.33s.4.378, 6.056, 7.571, 9.283, 9.7419.74115.8381m33s4.099, 5.746, 6.729, 8.377, 9.2769.27612.5081m35sTakeaway: raw pruning reduces external peak RSS by about
3.33 GiB; post-step retained RSS drops by about0.47 GiB.Validation
uv run ruff check src/prime_rl/orchestrator/trajectories.py src/prime_rl/orchestrator/filters.py src/prime_rl/orchestrator/train_sink.py src/prime_rl/orchestrator/types.py src/prime_rl/orchestrator/utils.py src/prime_rl/orchestrator/metrics.py src/prime_rl/orchestrator/orchestrator.py src/prime_rl/utils/monitor/prime.py src/prime_rl/utils/monitor/wandb.py tests/unit/orchestrator/test_trajectories.py tests/unit/orchestrator/test_filters.py tests/unit/orchestrator/test_advantage.pyuv run ruff format --check src/prime_rl/orchestrator/trajectories.py src/prime_rl/orchestrator/filters.py src/prime_rl/orchestrator/train_sink.py src/prime_rl/orchestrator/types.py src/prime_rl/orchestrator/utils.py src/prime_rl/orchestrator/metrics.py src/prime_rl/orchestrator/orchestrator.py src/prime_rl/utils/monitor/prime.py src/prime_rl/utils/monitor/wandb.py tests/unit/orchestrator/test_trajectories.py tests/unit/orchestrator/test_filters.py tests/unit/orchestrator/test_advantage.pyuv run pytest tests/unit/orchestrator/test_trajectories.py tests/unit/orchestrator/test_filters.py tests/unit/orchestrator/test_advantage.py tests/unit/orchestrator/test_batch.py tests/unit/train/rl/test_packer.pygit diff --check