Search results

wandb_gql import fails with W&B SDK >=0.27.1 - ModuleNotFoundError

Status: Open.

#3087 In PrimeIntellect-ai/prime-rl;

· TM23-sanji opened on Jul 19, 2026

Missing K8S guide

Status: Open.

#3054 In PrimeIntellect-ai/prime-rl;

· strickvl opened on Jul 16, 2026

Multi-run checkpoints can be marked stable without restartable policy weights

Status: Open.

#3037 In PrimeIntellect-ai/prime-rl;

· peter941221 opened on Jul 15, 2026

RL loss importance ratio can overflow to non-finite gradients

Status: Open.

#2972 In PrimeIntellect-ai/prime-rl;

· morluto opened on Jul 7, 2026

feat: adaptive concurrency control for rollout dispatch

enhancementNew feature or requestNew feature or request

Status: Open.

#2902 In PrimeIntellect-ai/prime-rl;

· mikasenghaas opened on Jun 29, 2026

TARGET_LAG = 1 is a silent, hardcoded config choice that I think should be surfaced?

Status: Open.

#2812 In PrimeIntellect-ai/prime-rl;

· GnarlyMshtep opened on Jun 14, 2026

Qwen 3.5 doesn't converge for SFT on Reverse-text for cp=2, fa3

Status: Open.

#2712 In PrimeIntellect-ai/prime-rl;

· Sirorezka opened on Jun 4, 2026

FFT resume fails on Qwen3.5-4B: missing optimizer state for untrained visual params

Status: Open.

#2676 In PrimeIntellect-ai/prime-rl;

· mikasenghaas opened on May 30, 2026

Unify max off-policy steps to measure trainer step − oldest inference policy version

Status: Open.

#2674 In PrimeIntellect-ai/prime-rl;

· mikasenghaas opened on May 30, 2026

Investigate step-0 KL mismatch for Qwen3.5 4B

Status: Open.

#2671 In PrimeIntellect-ai/prime-rl;

· mikasenghaas opened on May 29, 2026

How should Prime-RL disable Qwen3.5 thinking in renderer mode?

Status: Open.

#2662 In PrimeIntellect-ai/prime-rl;

· austinmw opened on May 28, 2026

feat: support top-k training

Status: Open.

#2600 In PrimeIntellect-ai/prime-rl;

· mikasenghaas opened on May 22, 2026