Search results

Qwen3-VL-8B推理问题

Status: Open.

#7335 In modelscope/ms-swift;

· mxw20010804 opened on Jan 8, 2026

How to do entropy regularization?

Status: Open.

#7331 In modelscope/ms-swift;

· asdfqwer2015 opened on Jan 8, 2026

四机8卡910B3昇腾环境下用swift-megatron微调deepseek_r1报错

bugSomething isn't workingSomething isn't working

npu

Status: Open.

#7330 In modelscope/ms-swift;

· smallshallot opened on Jan 8, 2026

NPU设备使用adamw_torch_npu_fused报错

bugSomething isn't workingSomething isn't working

npu

Status: Open.

#7327 In modelscope/ms-swift;

· weiliang987644015 opened on Jan 8, 2026

Question: Multi-frame VL training OOM with Megatron + GRPO (Qwen3-VL-8B)

Status: Open.

#7326 In modelscope/ms-swift;

· LinYuOu opened on Jan 8, 2026

想请教下 多个视频数据集一起训 如何为每个数据集设定不同的FPS_MAX_FRAMES

Status: Open.

#7325 In modelscope/ms-swift;

· yuchen2199 opened on Jan 8, 2026

Qwen3-VL-32B-Instruct GRPO Training Script

Status: Open.

#7324 In modelscope/ms-swift;

· Xujxyang opened on Jan 8, 2026

swift的packing是否支持attention的样本隔离

Status: Open.

#7322 In modelscope/ms-swift;

· xindaaW opened on Jan 8, 2026

rejected_response must be a str or list. rejected_response: 0.5

Status: Open.

#7319 In modelscope/ms-swift;

· kabugge opened on Jan 7, 2026

Question about freezing parameters of VL model

Status: Open.

#7318 In modelscope/ms-swift;

· Pay20Y opened on Jan 7, 2026

使用chunked cross entropy对显存没有优化

npu

Status: Open.

#7317 In modelscope/ms-swift;

· doctorMcy opened on Jan 7, 2026

NPU运行Qwen3-30B-A3B GRPO训练报错

Status: Open.

#7315 In modelscope/ms-swift;

· yingzhao27 opened on Jan 7, 2026