Skip to content

Pull requests: PrimeIntellect-ai/prime-rl

New pull request New
152 Open 2,750 Closed
152 Open 2,750 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

perf: keep scalar optimizer step counters on CPU in state offload
#3258 opened Aug 12, 2026 by samsja Member • Draft
1
chore: swap vllm-router fork for upstream on the router stack
#3257 opened Aug 12, 2026 by mikasenghaas Member • Draft
feat: online evals in the SFT entrypoint
#3256 opened Aug 12, 2026 by mikasenghaas Member • Draft
feat: hardcode rollout filters and zero-advantage drop
#3254 opened Aug 12, 2026 by mikasenghaas Member • Draft
fix: never fast-forward an infinite env's generator on resume
#3253 opened Aug 12, 2026 by parkerpettit • Draft
fix: keep zero-element dummy params out of the optimizer (NemotronH MoE resume)
#3252 opened Aug 12, 2026 by parkerpettit • Draft
feat(deps): add opt-in flashinfer-aot group for air-gapped installs
#3251 opened Aug 12, 2026 by samsja Member Loading…
chore: replace vllm-router fork with upstream release
#3250 opened Aug 12, 2026 by mikasenghaas Member • Draft
perf!: remove the zero-gradient-ratio metric
#3249 opened Aug 12, 2026 by samsja Member • Draft
feat: dynamic pass-through vLLM inference metrics to W&B
#3240 opened Aug 11, 2026 by mikasenghaas Member • Draft
feat: top-p/top-k train sampling with sampling replay
#3235 opened Aug 11, 2026 by mikasenghaas Member • Draft
feat: overlap CPU optimizer with backward
#3234 opened Aug 11, 2026 by samsja Member • Draft
feat: SFT from a HF dataset through the RL stack (dataset_sft)
#3233 opened Aug 11, 2026 by mikasenghaas Member • Draft
feat(monitor): upload native v1 episodes without orchestration refactor
#3232 opened Aug 10, 2026 by mrmoxon Loading…
7
fix: preserve non-RL signals in zero-advantage rollouts
#3223 opened Aug 8, 2026 by JayZenith Loading…
fix(nixl): size and allocate transfer arenas per accelerator
#3221 opened Aug 8, 2026 by yafshar • Draft
refactor: make orchestration verifiers v1-native
#3220 opened Aug 8, 2026 by hallerite Member • 2/2 • Draft
chore!: drop verifiers v0 environment support
#3219 opened Aug 8, 2026 by hallerite Member • 1/2 Loading…
fix(nixl): select UCX copy transport by accelerator
#3215 opened Aug 7, 2026 by yafshar Loading…
feat(trainer): recurrent GDN forward matching vLLM decode for Qwen3.5
#3214 opened Aug 7, 2026 by hubert-marek Contributor • Draft
2 tasks
feat(dynamo): add dynamo extra, local example, and fix renderers version constraint
#3213 opened Aug 6, 2026 by samsja Member • Draft
feat(orchestrator)!: make the episode the unit the orchestrator passes around
#3206 opened Aug 6, 2026 by mikasenghaas Member • Draft
fix: keep GLM-5.2 weight broadcast within GPU memory at 4-node scale
#3199 opened Aug 5, 2026 by samsja Member Loading…
fix(sft): drop rows with phantom media tokens; fail fast on fatal rank errors
#3198 opened Aug 5, 2026 by hubert-marek Contributor • Draft
1
feat: stream FSDP gradients through CPU offload
#3196 opened Aug 4, 2026 by samsja Member • Draft
3
ProTip! Type g p on any issue or pull request to go back to the pull request listing page.