Repository navigation

All pull requests

docs: fix typos in RULER workflows and file reader docstring
#642Avicennasis opened Oct 24, 20262 tasks done
fix(verl): normalize the engine policy loss over the global mini-batch
#641flyingdoubleG opened Oct 21, 20263 of 4 tasks
fix(jsd): make the JSD advantage follow the gradient of the JSD
#640flyingdoubleG opened Oct 21, 20263 of 4 tasks
fix(kl): apply the reward KL penalty only when ref_log_prob is present
#639flyingdoubleG opened Oct 20, 20263 of 4 tasks
fix(rec): handle groups with a single experience
#638flyingdoubleG opened Oct 20, 20263 of 4 tasks
fix(gigpo): match anchor states within each task
#637flyingdoubleG opened Oct 20, 20263 of 4 tasks
fix(sapo): differentiate the soft gate through the ratio
#636flyingdoubleG opened Oct 20, 20263 of 4 tasks
Add Docker-free multi-turn SQLite text-to-SQL workflow (BIRD)
#635flyingdoubleG opened Oct 18, 20262 of 4 tasks
fix(rewards): make DAPO zero-penalty region inclusive so cache_length=0 no longer divides by zero
#627Linxiushen opened Sep 21, 2026
fix(config): reject non-positive explorer_sync_interval / trainer_sync_interval
#626Linxiushen opened Sep 21, 2026
Add GSM8K examples for RLOO, REINFORCE++, GSPO and SAPO
#622KunyangZhang opened Sep 15, 20263 of 4 tasks
fix:_normalize_ascend_visible_devices
#618huajinguo12 opened Aug 26, 2026Contributor4 tasks done
fix: resolve Trinity RFT issue
#617sarthakkondawar2007 opened Aug 25, 2026
1
feat(algorithm): add M2PO for stale-rollout RFT
#616primorLee opened Aug 20, 20263 of 4 tasks
fix(docs): include group in checkpoint loading paths
#614CharlesXu-HQ opened Aug 16, 20264 tasks done
feat(config): add fused and foreach optimizer controls
#613CharlesXu-HQ opened Aug 15, 20263 of 4 tasks
Fix multi-turn truncation
#609pan-x-c opened Jul 17, 2026Collaborator4 tasks
4
Rollout Models support recording experience data
#591pan-x-c opened Jun 24, 2026Collaborator4 tasks done
27
feat(monitor): unify explorer/trainer wandb logging into one run
#590MengsD opened Jun 24, 2026Collaborator4 tasks
2
Add TaskDirReader for harbor datasets
#588pan-x-c opened Jun 23, 2026Collaborator4 tasks
[Experimental] Add a MultiStepPadding Operator
#511hiyuchang opened Feb 26, 2026Collaborator4 tasks
4
[Example] browse comp plus
#377garyzhang99 opened Nov 12, 2025Collaborator4 tasks
8

Navigation Menu

Repository navigation

All pull requests

docs: fix typos in RULER workflows and file reader docstring

fix(verl): normalize the engine policy loss over the global mini-batch

fix(jsd): make the JSD advantage follow the gradient of the JSD

fix(kl): apply the reward KL penalty only when ref_log_prob is present

fix(rec): handle groups with a single experience

fix(gigpo): match anchor states within each task

fix(sapo): differentiate the soft gate through the ratio

Add Docker-free multi-turn SQLite text-to-SQL workflow (BIRD)

fix(rewards): make DAPO zero-penalty region inclusive so `cache_length=0` no longer divides by zero

fix(config): reject non-positive `explorer_sync_interval` / `trainer_sync_interval`

Add GSM8K examples for RLOO, REINFORCE++, GSPO and SAPO

fix:_normalize_ascend_visible_devices

fix: resolve Trinity RFT issue

feat(algorithm): add M2PO for stale-rollout RFT

fix(docs): include group in checkpoint loading paths

feat(config): add fused and foreach optimizer controls

Fix multi-turn truncation

Rollout Models support recording experience data

feat(monitor): unify explorer/trainer wandb logging into one run

Add TaskDirReader for harbor datasets

[Experimental] Add a MultiStepPadding Operator

[Example] browse comp plus

Pull requests