abcalan

Results 9 comments of abcalan

Does the image have to be aligned?

Does the image have to be aligned @auroua

@songwsx 请问学习率是怎么调整的?

我的是pytorch模型转成mnn的,我看您这边是tf模型转成mnn,这2种模型在使用nativeGetSessionOutput这个api获取推理结果的时候是不是有所不同?

hello, Have you finished your work? @HannaRiver

大家后来怎么解决的,我稳定复现这个报错,我的命令脚本如下: CUDA_VISIBLE_DEVICES=0,1,2,3 \ NPROC_PER_NODE=4 \ MAX_PIXELS=50176 \ swift rlhf \ --rlhf_type grpo \ --model Qwen2.5-VL-32B-Instruct \ --external_plugins swift/qwen_fine_tuning/plugin/plugin.py \ --train_type lora \ --dataset 'lmms-lab/multimodal-open-r1-8k-verified' \ --torch_dtype bfloat16 \ --num_train_epochs 3...

问题解决了吗,不需要重新训练能推理了吗