Search results

Qwen3-8B-VL使用GRPO-chord训练，loss一直是0，使用LLM训练，loss看起来是正常的。

Status: Open.

#7128 In modelscope/ms-swift;

· white-wolf-tech opened on Dec 19, 2025

HunyuanOCR微调loss异常

Status: Open.

#7127 In modelscope/ms-swift;

· wusijie123 opened on Dec 19, 2025

Qwen3-235B-A22B GRPO OOM

Status: Open.

#7125 In modelscope/ms-swift;

· qingyuanxingsi opened on Dec 19, 2025

How can i merge lora that finetuned form megatron and save adapter as safetensors

Status: Open.

#7124 In modelscope/ms-swift;

· Kaiizx opened on Dec 19, 2025

swift deply 部署模型，生产的日志不包含集体的chat id，以及时间戳这些

Status: Open.

#7123 In modelscope/ms-swift;

· leileicheng854 opened on Dec 19, 2025

on_policy_distillation训练不稳定

Status: Open.

#7121 In modelscope/ms-swift;

· Jim2016713 opened on Dec 19, 2025

megatron 微调 qwen3vl-235b-a22b OOM

Status: Open.

#7119 In modelscope/ms-swift;

· gengenkai opened on Dec 19, 2025

How to train with a fixed frame using qwen2.5-omni？

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