zydmtaichi

> @avivbrokman, can you try adding `torch_adam: true` into the optimizer section of your ds_config? As described [here](https://www.deepspeed.ai/docs/config-json/#optimizer-parameters), this will enable `torch.optim.Adam` instead of DeepSpeed's cpu_adam, and should avoid the...

> 如果是v100 其并不支持bf16 你好，可是我用V100，在lora模式下微调了zero3-offload的glm4，没有发现报错，这是什么情况？

> I'd like to add a vote to @BradKML's suggestion on a re-ranker feature in the Ollama framework. This will help keep any RAG pipeline within Ollama end-to-end. +1

> This is because there are only 500 learned positional encodings and if you try to infer an image much higher than the default model resolution, then the number of...