Search results

[tx] Implement heuristic to determine micro batch sizes automatically

tx

Status: Open.

#1048 In NovaSky-AI/SkyRL;

· pcmoritz opened on Feb 7, 2026

[tx] Add end-to-end tests for multi-node in CI

tx

Status: Open.

#1036 In NovaSky-AI/SkyRL;

· pcmoritz opened on Feb 6, 2026

[algorithm] Add MaxRL mean normalization over advantages

Status: Open.

#1030 In NovaSky-AI/SkyRL;

· erictang000 opened on Feb 5, 2026

[algorithm] Add DPPO

Status: Open.

#1028 In NovaSky-AI/SkyRL;

· erictang000 opened on Feb 5, 2026

Support Google Cloud TPUs

Status: Open.

#1024 In NovaSky-AI/SkyRL;

· andrewsykim opened on Feb 4, 2026

[megatron][perf] Integrate megatron dynamic context parallelism

skyrl-train

Status: Open.

#1019 In NovaSky-AI/SkyRL;

· erictang000 opened on Feb 4, 2026

[skyrl][inference] Rollout plan for the new inference backend

Status: Open.

#1014 In NovaSky-AI/SkyRL;

· kouroshHakha opened on Feb 3, 2026

[tinkerification] Remove scheduler from inside the workers and default to use set_lr() and get_lr() in training loop

skyrl-train

Status: Open.

#998 In NovaSky-AI/SkyRL;

· erictang000 opened on Jan 31, 2026

[train][asyncRL] Add a flag to optionally reset KV cache for in-flight-weight update

skyrl-train

Status: Open.

#979 In NovaSky-AI/SkyRL;

· CharlieFRuan opened on Jan 28, 2026

[skyrl-train] Integrate Ray Core RDT for weight syncing

enhancementNew feature or requestNew feature or request

skyrl-train

Status: Open.

#977 In NovaSky-AI/SkyRL;

· erictang000 opened on Jan 27, 2026

[tx] Enable LoRAMixin to also accept flattened states (token_dim, hidden_dim) instead of only (batch, seq_len, hidden_dim)

tx

Status: Open.

#975 In NovaSky-AI/SkyRL;

· pcmoritz opened on Jan 27, 2026

[trainer][rollout] /health implementation for router when inference refactoring lands, and do not route to dead engines

Status: Open.

#958 In NovaSky-AI/SkyRL;

· CharlieFRuan opened on Jan 25, 2026