Enable AMP Compatibility for GCP Model to Reduce VRAM Usage

[mahdip72]

This update allows the GCP model to be used with Automatic Mixed Precision (AMP) training without encountering dtype mismatch errors. Previously, the following error occurred during AMP training:

"proteinworkshop/models/graph_encoders/layers/gcp.py", line 271, in scalarize local_scalar_rep_i[edge_mask] = torch.matmul( RuntimeError: Index put requires the source and destination dtypes match, got Float for the destination and Half for the source.</module>

Based on my experiments using Huggingface Accelerate AMP (fp16), it reduces VRAM usage by approximately 40% during training.

[amorehead]

LGTM

[amorehead]

@mahdip72, may I ask you to also update the CHANGELOG.md?

[mahdip72]

I’ve updated CHANGELOG.md under 0.2.6 (UNRELEASED).