Repository navigation
Fix rollout lifecycle, GPU colocation, and Claude Code integration - #2
Merged
Merged
Conversation
Forward colocate placement to Slime, release sealed trace payloads, and block retries and restarts while cleanup is unresolved. Reuse the embedded rollout loop, drain pending tasks on shutdown, and export bootstrap environment variables. Add focused regressions and lifecycle documentation. Validation: 293 tests passed with one environment-dependent skip; 11 Runner-environment tests passed. Real GPU acceptance completed three rollout rounds each for split and colocate, plus Docker cleanup and restart checks.
Share ordered task gathering and sibling cancellation between group and batch rollout. Consolidate cleanup marker publication while preserving the first cause and recording Runner identity. Extend cancellation and remote Runner regressions. Validation: 295 tests passed, one environment-dependent module skipped; git diff --check passed.
AltmanD
pushed a commit
to AltmanD/HarborRL-dev
that referenced
this pull request
Nov 11, 2026
Merge lwm_guide_zh.md, lwm_offline_verify_design_zh.md and lwm_offline_verify_report_zh.md into a single self-contained docs/algorithms/lwm_offline_zh.md, restructured to a Qwen-AgentWorld-style standard: abstract, formalized transition model and terminology, numbered figures/tables with in-text references, metric definitions with acceptance gates, and an honest analysis/interpretation-boundary section. Content purification: - drop all PR references (PR datamllab#2/#16) and local-commit citations; the replay push/sample contract is now described inline - drop branch development history and process notes (split fix, DNS unavailability); only current behavior and runnable commands remain - unify terminology (LWM=Latent World Model with an explicit disambiguation from Qwen-AgentWorld's Language World Model; LeWM; JEPA; tb2.1; ATIF) - replace site-specific absolute paths with placeholders and mark the WM_TRAJECTORIES default as must-override - correct stale output names (replay_buffer.pt, mpc_plan.json) and add the online_learner / tb21 env converter entry points Update docs/README.md, tools/README.md, slime/slime/world_model/README.md and examples/training/world_model/README.md to point at the merged doc. References use stable resources only (arXiv, official project pages, upstream GitHub repos); all links verified reachable.
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.
修复训练与任务执行中的以下问题：
验证：8×H200 上 split 和 colocate 均完成三轮真实推理与训练。