How to obtain useful outputs from existing models through conditioning, feature extraction, sampling, guidance, and ensembling.
How to obtain useful outputs from existing models through conditioning, feature extraction, sampling, guidance, and ensembling.
How runtime inputs such as MSAs, homolog prompts, templates, ligands, and initial structures affect predictions. Active reward-driven changes to conditioning also belong under inference/guidance.
15 items with this tag.
Combining predictions, scores, prompts, or samples from multiple model runs. Weight merging is documented separately under training/fine-tuning.
6 items with this tag.
Using learned embeddings, attention-derived features, pooling, or probes for downstream prediction and retrieval. A frozen base model may supply features to a separately trained head.
29 items with this tag.
Steering a fixed model with rewards, constraints, gradients, activation edits, or importance reweighting. Backpropagation to an input is not model fine-tuning.
26 items with this tag.
Exploring candidate sequences, structures, or designs through sampling, recycles, decoding order, search algorithms, and increased inference compute. Physical MD sampling stays under biophysics/molecular-simulation.
14 items with this tag.