Search results

Performance of <code>enable_activation_offloading</code>

Status: Open.

#2486 In volcengine/verl;

· vadimkantorov opened on Jul 11, 2025

Running GRPO training reports an error：CUDA error: device-side assert triggered

Status: Open.

#2483 In volcengine/verl;

· HelloWorld506 opened on Jul 11, 2025

[Feature Request] Add system prompt in SFTDataset

Status: Open.

#2482 In volcengine/verl;

· imShZh opened on Jul 11, 2025

Inconsistent usage of req.max_model_len and config.max_model_len when using sglang multi-turn training

Status: Open.

#2479 In volcengine/verl;

· bigbigQI opened on Jul 11, 2025

fsdp识别不到多卡，无论trainer.n_gpus_per_node设计为几，都会警告world_size为1

Status: Open.

#2478 In volcengine/verl;

· leesabak opened on Jul 11, 2025

在多轮对话强化学习训练中reward计算问题

Status: Open.

#2471 In volcengine/verl;

· jackwwy opened on Jul 11, 2025

[BUG] 用 fsdp2 训 72B 模型会出现 OOM，fsdp1 就不会

Status: Open.

#2464 In volcengine/verl;

· mlpod opened on Jul 10, 2025

想请教一下有没有代码RL有经验的大佬，算reward的时候平均时耗如何呢？Seeking advice from those experienced with CodeRL. What is a typical average time per reward computation?

Status: Open.

#2458 In volcengine/verl;

· F-Yuan303 opened on Jul 10, 2025

多机多卡训练，wandb不起作用，checkout会保存在镜像的/tmp/ray路径下，而不是训练参数指定的路径

Status: Open.

#2457 In volcengine/verl;

· shayxurui opened on Jul 10, 2025

Asynchronous mode of vLLM fails to converge.

Status: Open.

#2453 In volcengine/verl;

· gyr66 opened on Jul 10, 2025

Qwen/Qwen2.5-VL-3B-Instruct on GRPO LORA OOM 千问vl3B 上OOM问题

Status: Open.

#2450 In volcengine/verl;

· syyuan2021 opened on Jul 10, 2025

entropy checkpointing can be optimized when using "entropy_from_logits_with_chunking" function

Status: Open.

#2448 In volcengine/verl;

· hailuo97 opened on Jul 10, 2025