Search results

bash examples/qwen3_vl_4b_geo3k_grpo_cus.sh error

Status: Open.

#598 In hiyouga/EasyR1;

· cqray1990 opened on Jan 7, 2026

Discarding of generated samples in _make_batch_data in DAPO

Status: Open.

#594 In hiyouga/EasyR1;

· HalvesChen opened on Jan 1, 2026

Any plan to support CHORD, balance of SFT and RL

Status: Open.

#593 In hiyouga/EasyR1;

· yantijin opened on Dec 31, 2025

TypeError: 'NoneType' object is not subscriptable

Status: Open.

#591 In hiyouga/EasyR1;

· liuchaohu opened on Dec 30, 2025

请问有考虑集成 VAPO 吗？

Status: Open.

#590 In hiyouga/EasyR1;

· Taskii-Lei opened on Dec 29, 2025

Qwen3-VL F.conv3d Error

Status: Open.

#589 In hiyouga/EasyR1;

· Kun-Xiang opened on Dec 28, 2025

关于DAPO training step的疑惑

Status: Open.

#588 In hiyouga/EasyR1;

· MGDDestiny opened on Dec 26, 2025

BUG: model_inputs['input_ids'] not calculating correctly for videos.

Status: Open.

#584 In hiyouga/EasyR1;

· tqosu opened on Dec 12, 2025

rollout的vllm引擎默认是bf16精度，那么使用fp16训练时这里有精度的偏差么？

Status: Open.

#583 In hiyouga/EasyR1;

· madorism opened on Dec 12, 2025

<code>Qwen2VLImageProcessorFast</code> works only with image inputs and doesn't process videos anymore.

Status: Open.

#582 In hiyouga/EasyR1;

· tqosu opened on Dec 11, 2025

[Question] Cuda failure 217 (peer access is not supported).

Status: Open.

#579 In hiyouga/EasyR1;

· liuchongming1999 opened on Dec 5, 2025

bugfix: <code>log_importance_ratio = negative_approx_kl_in_seq * response_mask</code>. tensor shape mismatch when loss_type='gspo'

Status: Open.

#577 In hiyouga/EasyR1;

· weichenma15 opened on Dec 2, 2025