Search results

How to train with a fixed frame using qwenomni2.5？

Status: Open.

#7114 In modelscope/ms-swift;

· wwxs123W opened on Dec 18, 2025

使用sglang推理加速Qwen2.5-vl模型报错

Status: Open.

#7113 In modelscope/ms-swift;

· chenyili502 opened on Dec 18, 2025

GRPO训练qwen3vl8B rollout一直为乱码导致无法训练

Status: Open.

#7112 In modelscope/ms-swift;

· lingfengren opened on Dec 18, 2025

如何将自定义指标保存至tensorboard和swanlab

Status: Open.

#7111 In modelscope/ms-swift;

· weiliang987644015 opened on Dec 18, 2025

qwen3-vl微调grounding任务，数据可以包含负样本吗

Status: Open.

#7110 In modelscope/ms-swift;

· fenfenwang520 opened on Dec 18, 2025

best模型的保存

Status: Open.

#7109 In modelscope/ms-swift;

· ALONGNEU opened on Dec 18, 2025

Qwen-3-Omni SFT and RL training slowdown with Megatron after upgrading to vLLM ≥ 0.12 (audio-in-video enabled)

Status: Open.

#7108 In modelscope/ms-swift;

· callmeluoq-art opened on Dec 18, 2025

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:1 and cuda:0!

Status: Open.

#7107 In modelscope/ms-swift;

· AppalachianWine opened on Dec 18, 2025

可以支持Qwen3Guard的SFT吗？

Status: Open.

#7106 In modelscope/ms-swift;

· SongDark opened on Dec 18, 2025

使用多卡开启rollout通信报错？

Status: Open.

#7104 In modelscope/ms-swift;

· yyqqh098 opened on Dec 18, 2025

[QWEN3_OMNI] The Training of Qwen3-Omni-30B-A3B-Instruct is extremely slow

Status: Open.

#7103 In modelscope/ms-swift;

· Respaired opened on Dec 18, 2025

用megatron-swift训练qwen3 omni的时候，completions.jsonl中记录的completion token mean len和真实的不一致。

Status: Open.

#7101 In modelscope/ms-swift;

· WeixuanXiong opened on Dec 17, 2025