Search results

[tinker] Enable multi-lora training for async RL

Status: Open.

#1647 In NovaSky-AI/SkyRL;

· SumanthRH opened on May 11, 2026

[tinker] Support additional Tinker LoraConfig parameters

Status: Open.

#1632 In NovaSky-AI/SkyRL;

· erictang000 opened on May 7, 2026

[vllm] Investigate whether NCCL_CUMEM_ENABLE should be set to 0

Status: Open.

#1630 In NovaSky-AI/SkyRL;

· erictang000 opened on May 7, 2026

[models] Qwen3.5 Issue Tracker

Status: Open.

#1626 In NovaSky-AI/SkyRL;

· erictang000 opened on May 6, 2026

[bug] OOM with new inference during weight sync

Status: Open.

#1623 In NovaSky-AI/SkyRL;

· erictang000 opened on May 5, 2026

SkyRLGymGenerator crashes whole training step when one rollout fails

Status: Open.

#1613 In NovaSky-AI/SkyRL;

· jamesbraza opened on May 2, 2026

RayPPOTrainer._remove_tail_data silently truncates a training batch to empty and throws ValueError

Status: Open.

#1609 In NovaSky-AI/SkyRL;

· jamesbraza opened on May 1, 2026

[bug] Race condition in new inference weight sync: start_weight_update called while already active

Status: Open.

#1605 In NovaSky-AI/SkyRL;

· pbokc opened on May 1, 2026

Flaky test_qwen3_moe_layer_lora, replace np.allclose?

Status: Open.

#1604 In NovaSky-AI/SkyRL;

· agolajko opened on Apr 30, 2026

New inference codepath gap (_SKYRL_USE_NEW_INFERENCE=1)

Status: Open.

#1591 In NovaSky-AI/SkyRL;

· CharlieFRuan opened on Apr 29, 2026

vllm==0.20.0 (and thus torch==2.11.0) support

Status: Open.

#1590 In NovaSky-AI/SkyRL;

· jamesbraza opened on Apr 29, 2026

AssertionError crash on eval when val-set size is not a multiple of train_batch_size

Status: Open.

#1583 In NovaSky-AI/SkyRL;

· jamesbraza opened on Apr 28, 2026