xylcbd comments

Results 12 comments of


                                            xylcbd

If there any method to merge oriented small box into oriented line?

nice work @dajiangxiaoyan

Failed to run demo

@civilman628 thanks, your answer helped me!

How to pass 2 images through the same pipeline (with same decisions)

@Oktai15 @mdbloice do you solved the issue with ground-truth on the fly? or any suggestion?

what is the purpose of delay_load for vision tower?

me too.

> Hi, I guess you can try this > > ``` > FINETUNE_NAME=Mini-Gemini-7B > STAGE3_NAME=Your_prefer_name > AUX_SIZE=768 > deepspeed minigemini/train/train_mem.py \ > --deepspeed ./scripts/zero2_offload.json \ > --model_name_or_path ./work_dirs/$FINETUNE_NAME \ >...

Failed to continous sft for yi-34B with 8x CUDA graphics card! (deepspeed zero3)

@yanwei-li

请问有推荐的部署框架吗?

> @Halflifefa 打个广告： > > LMDeploy 目前支持 InternVL之前发布的6个VL模型。1.5会在这周支持。 > > 这是相关文档： https://github.com/InternLM/lmdeploy/blob/main/docs/zh_cn/inference/vl_pipeline.md https://github.com/InternLM/lmdeploy/blob/main/docs/zh_cn/serving/api_server_vl.md 这个相比直接用pytorch推理，速度和显存有啥变化吗？

a small model such as Mobilenet v2 for pre-training

@keyu-tian Can I use swinv2-base as the backbone for pre-training?

xylcbd