Search results

Feature Request: Support LoRA Fine-Tuning for DeepSeek-V3 in RL Training

Status: Open.

#3279 In volcengine/verl;

· zhenzhenzhizhi opened on Aug 30, 2025

How to load the fine-tuned LORA model?

Status: Open.

#3278 In volcengine/verl;

· zhuhrun opened on Aug 30, 2025

训练qwen2.5-1.5B在gsm8k上grpo训练时间多久正常？

Status: Open.

#3277 In volcengine/verl;

· digbangbang opened on Aug 30, 2025

use_kl_loss vs use_kl_in_reward

Status: Open.

#3276 In volcengine/verl;

· QiyaoWei opened on Aug 30, 2025

SFT VLM multimodal examples?

Status: Open.

#3275 In volcengine/verl;

· dragen1860 opened on Aug 30, 2025

[Bug] verl 0.5.0 is incompatible with vLLM v1 API when using LoRA

Status: Open.

#3271 In volcengine/verl;

· llz3724 opened on Aug 29, 2025

Clarification: Does VERL support CPU-only training

Status: Open.

#3266 In volcengine/verl;

· kahlun opened on Aug 29, 2025

Should we add <code>log_prob_max_token_len_per_gpu</code> feature back? (which is remove in PR 2524)

Status: Open.

#3265 In volcengine/verl;

· binary-sky opened on Aug 29, 2025

[one-step off-policy]communication group optimize suggestion

Status: Open.

#3264 In volcengine/verl;

· github-my opened on Aug 29, 2025

How to Check Time Consumed in 1. Rollout and 2. Trainer for Async

Status: Open.

#3263 In volcengine/verl;

· Zhichao-Jimmy970201 opened on Aug 29, 2025

Does Verl plan to support reinforcement learning for the DIT model in the future?

Status: Open.

#3262 In volcengine/verl;

· razor1991 opened on Aug 29, 2025

[BUG] Qwen3 MoE FSDP2 meets <code>torch.utils.checkpoint.CheckpointError</code> when <code>offload_policy=True</code>

Status: Open.

#3258 In volcengine/verl;

· chenyushuo opened on Aug 29, 2025