Qingfu Wen

> New solution, use docker directly.This is nignx with nginx-gridfs plugin for mongo access file > https://hub.docker.com/r/zzshcool/nginx-mongo What mongo version does this image suppport?

@jcupitt I applied `fetch` method to 4 types of WSI for 2048*2048 patch， mrxs, tiff, ndpi, svs, they both use JPEG compression schema. their fetch time for one single patch...

@jcupitt ok, thanks a lot!

I got this error too when useing pytorch 1.13.0, colossalai 0.2.8 with transformers 4.28.1 and 4.24.0

@binmakeswell it works fine using official docker colossalai:0.2.7 which is pytorch 1.12.1

I also trained qwen3-8b eagle3 using SpecForge on a Chinese dataset(Chinese-DeepSeek-R1-Distill-data-110k)(generated answer using Qwen3-8B), but after 10 epoch only got (Position 0) train acc 0.7. and test using vllm with...

> > I also trained qwen3-8b eagle3 using SpecForge on a Chinese dataset(Chinese-DeepSeek-R1-Distill-data-110k)(generated answer using Qwen3-8B), but after 10 epoch only got (Position 0) train acc 0.7. and test using...

same issue on V100, any update to support V100?