Search results

Qwen3-30B-A3B LoRA训练router_aux_loss_coef设置问题

Status: Open.

#5598 In modelscope/ms-swift;

· alex-wll opened on Aug 29, 2025

训练Qwen3_32B_GRPO的时候rank0会重命名dataset_infos.json这个文件

Status: Open.

#5597 In modelscope/ms-swift;

· Zikhjm opened on Aug 29, 2025

Bug:megatron sft训练进行到eval和save的iter时退出

Status: Open.

#5594 In modelscope/ms-swift;

· Sarahtorekryo opened on Aug 28, 2025

GPT-OSS Sequence Parallelism

Status: Open.

#5593 In modelscope/ms-swift;

· cbartmann opened on Aug 28, 2025

GRPO with vLLM is not compatible with device_map. Please set NPROC_PER_NODE equal to num_processes.

Status: Open.

#5592 In modelscope/ms-swift;

· MSS444 opened on Aug 28, 2025

swift3.5.0中attn_impl='flash_attn'在哪部分生效？

Status: Open.

#5591 In modelscope/ms-swift;

· juliajulietZ opened on Aug 28, 2025

处理数据集的时候会往磁盘里写数据

Status: Open.

#5590 In modelscope/ms-swift;

· LaoWangGB opened on Aug 28, 2025

Qwen-MOE finetune results in mixed language tokens (language switching in generated answers)

Status: Open.

#5587 In modelscope/ms-swift;

· chapter544 opened on Aug 28, 2025

AWQ量化后的模型，infer的时候报错

Status: Open.

#5583 In modelscope/ms-swift;

· HouTong-s opened on Aug 28, 2025

early stop

Status: Open.

#5582 In modelscope/ms-swift;

· Awesome065 opened on Aug 28, 2025

Clarification on --loss_scale ignore_empty_think when fine-tuning Qwen3-Instruct vs Qwen3-Thinking

Status: Open.

#5581 In modelscope/ms-swift;

· imrankh46 opened on Aug 28, 2025

eval_loss is always nan during lora fine tuning qwen2.5-vl-3b-instruct

Status: Open.

#5580 In modelscope/ms-swift;

· doggnol opened on Aug 28, 2025