Skip to content

Pull requests: PrimeIntellect-ai/prime-rl

New pull request New
217 Open 2,939 Closed
217 Open 2,939 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

chore(trainer): unify context-parallel setup across models
#3533 opened Sep 11, 2026 by garrett361 Contributor Loading…
5
fix(inference): preserve configured worker extension
#3531 opened Sep 11, 2026 by samjiawng Loading…
feat(algo): add optional SWE-2 length-weighted group baseline
#3528 opened Sep 10, 2026 by samsja Member • Draft
test: remove HF reference suites and organize native model coverage
#3527 opened Sep 10, 2026 by S1ro1 Collaborator • Draft
feat(trainer): add optional IcePop loss
#3526 opened Sep 10, 2026 by samsja Member Loading…
6
Persist per-token policy mask decisions
#3525 opened Sep 10, 2026 by samsja Member Loading…
2
Add Total Router Recall for MoE training
#3523 opened Sep 10, 2026 by samsja Member • Draft
Reclaim consumed CPU gradient pages during full offload
#3522 opened Sep 9, 2026 by samsja Member • Draft
Reproduce exact-zero train/inference mismatch for Qwen3 and GLM-4.5-Air
#3520 opened Sep 9, 2026 by samsja Member • Draft
fix(orchestrator): restore zero-advantage filtering flag
#3515 opened Sep 9, 2026 by S1ro1 Collaborator • 2/4 • Draft
chore(inference): update vLLM and use its native FP32 head
#3513 opened Sep 9, 2026 by S1ro1 Collaborator • 3/4 • Draft
feat: add NVFP4 MoE training with optional 4over6 quantization
#3511 opened Sep 9, 2026 by S1ro1 Collaborator • 4/4 • Draft
feat(rl): add experimental Total Router Recall
#3509 opened Sep 9, 2026 by faresobeid Contributor • Draft
chore: sync agentic syn-gen stack with latest main
#3507 opened Sep 8, 2026 by samsja Member • Draft
Feat/dequant triton kernel
#3506 opened Sep 8, 2026 by dhanyamd • Draft
fix: propagate engine startup timeout to vLLM workers
#3500 opened Sep 7, 2026 by samsja Member • Draft
feat: accept max reasoning effort for evaluation
#3499 opened Sep 7, 2026 by samsja Member • Draft
feat: support DeepEP v2 with vLLM 0.28 compatibility
#3498 opened Sep 7, 2026 by samsja Member • Draft
exp: pin completion and routing dependency stack for OpenSeeker RL
#3496 opened Sep 6, 2026 by hallerite Member • Draft
fix: configure router request timeout for long agentic generations
#3495 opened Sep 6, 2026 by hallerite Member • Draft
fix(sft): preserve tool definitions in nullable columns
#3494 opened Sep 6, 2026 by feng1201 • Draft
feat(inference): launch managed Dynamo workers
#3493 opened Sep 6, 2026 by biswapanda Contributor Loading…
fix(inference): support vLLM skip-gather logits
#3492 opened Sep 5, 2026 by biswapanda Contributor Loading…
fix(wandb): use compatible GraphQL transport
#3489 opened Sep 4, 2026 by LOGO127 Loading…
1
feat(transports): add mx_refit ModelExpress weight-broadcast transport
#3487 opened Sep 4, 2026 by KavinKrishnan Loading…
ProTip! Follow long discussions with comments:>50.