Using distributed or parallel set-up in script?: yes
Who can help?
@ArthurZucker @stas00
Hi, I am performing inference using GPT-NeoX 20B model using greedy search. Without deepspeed the text generation works fine. However, when I use deepspeed for inference, I am getting the following error
transformers isn't involved with deepspeed's inference engine, other than being used by it indirectly, so please refile at https://github.com/microsoft/DeepSpeed/issues. Thank you.
System Info
transformers
version: 4.28.0.dev0Who can help?
@ArthurZucker @stas00
Hi, I am performing inference using
GPT-NeoX 20B
model using greedy search. Without deepspeed the text generation works fine. However, when I use deepspeed for inference, I am getting the following errorThis is how I am wrapping deepspeed around the model
Information
Tasks
examples
folder (such as GLUE/SQuAD, ...)Reproduction
Expected behavior
There should be no difference whether I wrap
deepspeed
around the model or not.