Supervised fine-tuning (SFT) corpus for Large Discovery Models (LDM): a dataset that
distils an acquisition-guided, test-time search policy into a language-model proposer so
that a single forward pass emulates a full model-based optimization loop.
An LDM couples three components in a recurrent generate → select → evaluate → update
loop: an LLM that proposes candidate experiments, a probabilistic surrogate that maps
observations… See the full description on the dataset page:
https://huggingface.co/datasets/Yangtze-ailab/LDM-CoT-Acq-SFT-16K.
