Skip to main content
All boileroom models return typed dataclass objects containing prediction results and metadata. Each model’s output type is documented on its own page — this page covers the shared PredictionMetadata type, cross-model comparison, and the include_fields filtering mechanism.

​
PredictionMetadata

Every output’s metadata field is a PredictionMetadata instance with timing and model information.
​
model_name
str
Name of the model (e.g., "ESMFold", "ESM-2", "Chai-1", "Boltz-2").
​
model_version
str
Version string of the model.
​
sequence_lengths
list[int] | None
Number of residues for each input sequence (excluding chain separators).
​
preprocessing_time
float | None
Time spent in preprocessing (seconds).
​
inference_time
float | None
Time spent in model inference (seconds).
​
postprocessing_time
float | None
Time spent in postprocessing (seconds).

​
Per-model output types

Each model page documents its full output dataclass:

​
Cross-model comparison

FieldESMFoldESM2Chai-1Boltz-2
metadataYesYesYesYes
atom_arrayYes—YesYes
embeddings—Yes——
plddtYes—YesYes
paeYes—YesYes
pde——YesYes
ptmYes—Yes—
iptm——Yes—
confidence———Yes
pdbYes——Yes
cifYes—YesYes
chain_indexYesYes——
residue_indexYesYes——
hidden_states—Yes——
s_s / s_zYes———

​
Filtering with include_fields

All models support an include_fields configuration key that controls which optional fields are returned. This is useful for reducing memory usage and transfer time when you only need specific outputs. 
# Only return pLDDT scores (atom_array is always included for folding models)
result = model.fold("MKTVRQERLKSIVRI", options={"include_fields": ["plddt"]})

# Return everything including PDB/CIF strings
result = model.fold("MKTVRQERLKSIVRI", options={"include_fields": ["*"]})

# Return PDB and PAE
result = model.fold("MKTVRQERLKSIVRI", options={"include_fields": ["pdb", "pae"]})
Fields not listed in include_fields are set to None. The following fields are always included regardless of the filter:
  • Folding models: metadata, atom_array
  • Embedding models: metadata, embeddings, chain_index, residue_index