Thank you for releasing MiniFold! I noticed that running the LM in half precision is currently disabled: https://github.com/jwohlwend/minifold/blob/63db8b91f5bb64535f20810f014e69df66b0faad/minifold/model/model.py#L276 Can you say more about what the inference issues are with autocast on the LM?
Thank you for releasing MiniFold!
I noticed that running the LM in half precision is currently disabled:
minifold/minifold/model/model.py
Line 276 in 63db8b9
Can you say more about what the inference issues are with autocast on the LM?