Search results

Identical completions in multi-node GRPO training

Status: Open.

#7219 In modelscope/ms-swift;

· dmit-vuk opened on Dec 26, 2025

多轮对话grpo训练出现问题

Status: Open.

#7214 In modelscope/ms-swift;

· warriorandroyal opened on Dec 26, 2025

使用swift训练qwen3_next，效率很低是否是正常现象？

enhancementNew feature or requestNew feature or request

Status: Open.

#7210 In modelscope/ms-swift;

· alanayu opened on Dec 25, 2025

GRPO数据集，提示词部分是否需要提示模型 <Think>格式，如果使用format作为reward的话

Status: Open.

#7209 In modelscope/ms-swift;

· OKC13 opened on Dec 25, 2025

能正常推理的情况下，使用官方命令尝试微调时，下载数据集时显示获取数据集文件列表失败，无论使用modelscope还是huggingface的数据集都存在这个问题

bugSomething isn't workingSomething isn't working

Status: Open.

#7206 In modelscope/ms-swift;

· 124sds opened on Dec 25, 2025

多机多卡 sft Qwen3-VL-8B时同时开启sequence_parallel_size和enable_channel_loss时报错

bugSomething isn't workingSomething isn't working

Status: Open.

#7204 In modelscope/ms-swift;

· pingpingpangpang opened on Dec 25, 2025

gme训练多batchsize使用自己设置的negative负样本报错，直接使用batch内采样正常

bugSomething isn't workingSomething isn't working

Status: Open.

#7202 In modelscope/ms-swift;

· dickobe opened on Dec 24, 2025

ms-swift + vLLM 0.11.0,grpo无法运行

Status: Open.

#7200 In modelscope/ms-swift;

· LIjia555 opened on Dec 24, 2025

sample建议支持多个orm

enhancementNew feature or requestNew feature or request

Status: Open.

#7196 In modelscope/ms-swift;

· PiggerZZM opened on Dec 24, 2025

Support Fun-Audio-Chat web reasoning (client and server).

Status: Open.

#7193 In modelscope/ms-swift;

· pruidong opened on Dec 24, 2025

[MS-Swift][vLLM][Mistral] MS-Swift vLLM inference output corrupted by GPT-2 BPE artifacts (Ċ, Ġ) with Ministral-3-14B-Instruct-2512

Status: Open.

#7185 In modelscope/ms-swift;

· eunryonghan opened on Dec 24, 2025

多轮对话DPO训练集构建

Status: Open.

#7184 In modelscope/ms-swift;

· codecat0 opened on Dec 24, 2025