Skip to content

Pull requests: PrimeIntellect-ai/prime-rl

New pull request New
165 Open 2,834 Closed
165 Open 2,834 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

fix: restore standalone GPU trainer launch
#3362 opened Aug 23, 2026 by LeonEricsson Loading…
fix(nccl): preserve checkpoint chunk boundaries during vLLM reload
#3361 opened Aug 23, 2026 by rasdani Contributor Loading…
POC: stop-gradient through context tokens (row-sparse backbone backward)
#3360 opened Aug 23, 2026 by samsja Member • Draft
feat: add local run dashboard
#3351 opened Aug 23, 2026 by mikasenghaas Member • Draft
1
fix(trainer): distribute routing replay padding across experts
#3350 opened Aug 23, 2026 by EazyReal Contributor Loading…
1
feat: add dcp-to-hf and fp8 conversion scripts
#3348 opened Aug 22, 2026 by mikasenghaas Member Loading…
16
perf(nixl): pipeline reusable weight pulls
#3341 opened Aug 22, 2026 by S1ro1 Collaborator • 2/2 • Draft
feat(slurm): launch ModelExpress for NIXL transfers
#3340 opened Aug 22, 2026 by S1ro1 Collaborator • 1/2 • Draft
docs(algorithms): document renderer template controls on the RL path
#3339 opened Aug 22, 2026 by HBX814 • Draft
1
feat(transports): add mx_refit weight-broadcast transport
#3336 opened Aug 22, 2026 by tanushriya910 • Draft
Don't compute the logprobs and their gradients for tokens no loss com…
#3326 opened Aug 21, 2026 by MarioSieg Member • Draft
feat: add deepseek v4
#3325 opened Aug 21, 2026 by garrett361 Contributor • Draft
fix(slurm): don't kill the final eval when the trainer finishes
#3324 opened Aug 21, 2026 by mikasenghaas Member • Draft
Train from inline multimodal rollouts
#3320 opened Aug 20, 2026 by eligotts Contributor • Draft
feat: add Qwen3.8-27B (dense) support
#3319 opened Aug 20, 2026 by samsja Member Loading…
2
docs: launch standalone RL trainer check with torchrun
#3311 opened Aug 20, 2026 by CooperSheroy Loading…
1
fix: guard None raw tokens in logprobs detokenization
#3304 opened Aug 19, 2026 by mikasenghaas Member • Draft
fix: cap attn='auto' at FA2 when the model impl resolves to HF
#3300 opened Aug 18, 2026 by mikasenghaas Member • Draft
feat(inference): Dynamo integration
#3296 opened Aug 18, 2026 by biswapanda Loading…
feat: add KPop loss type
#3283 opened Aug 17, 2026 by mikasenghaas Member • Draft
feat(orchestrator): turn-credit shaping algorithm (turn_credit)
#3282 opened Aug 16, 2026 by snimu Collaborator • Draft
feat(orchestrator): record the scalar advantage on trace records
#3277 opened Aug 15, 2026 by Mantissagithub Loading…
1
2
chore: add test for parallelism (#1432)
#3276 opened Aug 15, 2026 by ParamThakkar123 Loading…
1
Enable training with streaming harnesses
#3273 opened Aug 14, 2026 by hallerite Member • Draft
refactor(inference): use native vLLM weight reload and NCCL transfer
#3260 opened Aug 13, 2026 by samsja Member • Draft
ProTip! Type g p on any issue or pull request to go back to the pull request listing page.