Module pretrained 

ESM3 pretrained weight loading.

Downloads weights from EvolutionaryScale/esm3-sm-open-v1 on HuggingFace (gated access — the user must accept the Cambrian Non-Commercial license and run huggingface-cli login).

Weight layout

The .pth file is a plain PyTorch state dict saved via torch.save(model.state_dict(), ...). No wrapping key is used; PthTensors::new(path, None) loads it directly.

Model attribute (Python)	VarBuilder prefix
encoder.*	encoder.*
transformer.*	transformer.*
output_heads.*	output_heads.*

The structure encoder uses a separate checkpoint (esm3_structure_encoder_v0.pth):

Python attribute	VarBuilder prefix
encoder.* (blocks)	encoder.*
pre_vq_proj.weight	pre_vq_proj.*
codebook.embeddings	codebook.*

Structs

ESM3Runner

Wraps a loaded ESM3 model for sequence embedding and multi-track inference.

StructureEncoderRunner

Wraps the ESM3 structure token encoder for converting backbone coordinates to tokens.

Enums

ESM3Models

Available ESM3 model variants.