Search results

[RFC] Save LoRA-only weights during training for inference usage

Status: Open.

#285 In verl-project/verl-omni;

· guozhihao-224 opened on Jul 20, 2026

[RFC] Defer prompt ID and mask tensorization until encode_prompt() for prompt embedding cache

Status: Open.

#283 In verl-project/verl-omni;

· Sky-Trigger opened on Jul 20, 2026

[RFC] Rollout-train log-prob consistency: measurements + default monitoring

Status: Open.

#280 In verl-project/verl-omni;

· cr-gao opened on Jul 17, 2026

RFC: L3 Nightly Regression for Qwen-Image + FlowGRPO

Status: Open.

#275 In verl-project/verl-omni;

· wtomin opened on Jul 16, 2026

[RFC] Qwen3-Omni Offline DPO Training Framework on FSDP

Status: Open.

#266 In verl-project/verl-omni;

· wtomin opened on Jul 15, 2026

Model support of BOOGU-Image

help wantedExtra attention is neededExtra attention is needed

Status: Open.

#265 In verl-project/verl-omni;

· mzhang367 opened on Jul 14, 2026

[RFC] Refactor experimental step execution support by reusing vLLM-Omni interfaces

Status: Open.

#251 In verl-project/verl-omni;

· Sky-Trigger opened on Jul 9, 2026

[RFC] Support trainer v1 for verl omni Diffusion model

Status: Open.

#249 In verl-project/verl-omni;

· knlnguyen1802 opened on Jul 9, 2026

[vllm_omni, rollout] BAGEL LoRA rollout generation ~28% slower than non-LoRA due to DiffusionBaseLinearLayerWithLoRA overhead

bugSomething isn't workingSomething isn't working

Status: Open.

#248 In verl-project/verl-omni;

· zhtmike opened on Jul 9, 2026

[RFC] Layered L1-L4 CI/CD Plan for VeRL-Omni

Status: Open.

#247 In verl-project/verl-omni;

· wtomin opened on Jul 9, 2026

[RFC]: Qwen3-Omni Full-Modal RL (Thinker + Talker) on Huawei Ascend 910C with verl-omni + vllm-omni

Status: Open.

#245 In verl-project/verl-omni;

· zyxzyxzyx opened on Jul 8, 2026

[RFC] Image-to-Image (I2I) Diffusion Model Integration for FlowGRPO

good first issueGood for newcomersGood for newcomers

Status: Open.

#237 In verl-project/verl-omni;

· NancyFyong opened on Jul 5, 2026