Issues

• Enable CodeQL code scanning

  openrl-dev/openrl#449 · Filter by author hodaaaaaaaaaa opened on Nov 6, 2026

• delta sync: a sampler's delta read is latency-bound on NFS; parallel reads help there and hurt on local disk

  openrl-dev/openrl#443 · Filter by author droot opened on Nov 6, 2026

• Run inspection API: design and rollout plan

  openrl-dev/openrl#390 · Filter by author ShubyM opened on Oct 29, 2026
  1

• Clean up old vLLM compile caches on the shared volume

  openrl-dev/openrl#376 · Filter by author hodaaaaaaaaaa opened on Oct 23, 2026

• sampling: size vLLM's batch the same way for in-process and vllm serve samplers

  openrl-dev/openrl#353 · Filter by author ShubyM opened on Oct 21, 2026

• Complete CNCF onboarding

  openrl-dev/openrl#349 · Filter by author droot opened on Oct 21, 2026
  

• Trainer precision: Automodel LoRA and FFT update bf16-resident weights (no fp32 master), unlike HF LoRA

  openrl-dev/openrl#347 · Filter by author sunnythecode opened on Oct 21, 2026

• Weight-sync format contract: self-describing metadata, trainer↔sampler dtype agreement, and cast detection

  openrl-dev/openrl#327 · Filter by author sunnythecode opened on Oct 12, 2026
  

• Choose trainer_gpus and trainer_cp automatically

  openrl-dev/openrl#317 · Filter by author ShubyM opened on Oct 7, 2026

• PoC: full fine-tuning on TPU

  openrl-dev/openrl#315 · Filter by author hodaaaaaaaaaa opened on Oct 6, 2026
  

• PoC: LoRA on TPU

  openrl-dev/openrl#314 · Filter by author hodaaaaaaaaaa opened on Oct 6, 2026
  

• Support TPUs in OpenRL

  openrl-dev/openrl#313 · Filter by author hodaaaaaaaaaa opened on Oct 6, 2026
  

• Download trained weights and serve them with vLLM

  priority:P1
  openrl-dev/openrl#291 · Filter by author droot opened on Oct 2, 2026
  

• Consistent session configuration: one mechanism for OpenRL settings

  cleanuppriority:P1
  openrl-dev/openrl#290 · Filter by author droot opened on Oct 2, 2026
  5

  

• Exclusive sessions: dedicated accelerators for hero runs

  priority:P0
  openrl-dev/openrl#289 · Filter by author droot opened on Oct 2, 2026
  

• Bring your own trainer and sampler images

  priority:P0
  openrl-dev/openrl#280 · Filter by author droot opened on Sep 30, 2026
  1

• Per-Role Worker Images

  priority:P0
  openrl-dev/openrl#273 · Filter by author hodaaaaaaaaaa opened on Sep 29, 2026

• Batch depandabot PRs into a single PR

  openrl-dev/openrl#269 · Filter by author barney-s opened on Sep 25, 2026

• Docs: tinker-cookbook guidance is out of date

  openrl-dev/openrl#249 · Filter by author droot opened on Sep 18, 2026

• get_info reports hardcoded lora_rank: 16 and arch: unknown

  openrl-dev/openrl#248 · Filter by author droot opened on Sep 18, 2026

• Worker teardown ~2 min after heartbeats stop; docs say Workloads persist

  openrl-dev/openrl#247 · Filter by author droot opened on Sep 18, 2026

• Deleting a Workload under a live session strands the client

  openrl-dev/openrl#246 · Filter by author droot opened on Sep 18, 2026

• Harvey Labs Recipe against Cluster

  openrl-dev/openrl#225 · Filter by author ShubyM opened on Sep 4, 2026

• Multi-gpu support

  openrl-dev/openrl#221 · Filter by author ShubyM opened on Sep 3, 2026

• OpenRL recipes should demonstrate use of sandbox RL environments

  priority:P1
  openrl-dev/openrl#181 · Filter by author droot opened on Aug 18, 2026