jetskewur/ClimAdaptLM-III-policy-nli-agent

distilroberta-base-climate-adaptation-policy-nli

This model is a fine-tuned version of climatebert/distilroberta-base-climate-f on a dataset of adaptation policy premise-hypothesis pairs.

Model description

A dataset of 47,869 premise-hypothesis pairs with binary labelling (0: neutral, 1: entailment) was used to fine-tune this model. The premises contain climate change adaptation policy elements (i.e., substrings from climate policy documents representing a goal, instrument, or output). In 16,474 pairs, the hypotheses are class descriptions of hazard and sector classes (5 hazard, 5 sector classes). Out of these, 4,708 were labelled "entailment" (1). The remaining 31,395 pairs contain hypotheses referring to goal, instrument, or output class definitions. Here, a total of 5 goal, 6 instrument, and 5 output classes are included. Per "entailment" label (1), two premise-hypothesis pairs with a "neutral" (0) label are included.

This model can be used to classify climate (adaptation) policy elements by providing a premise (the policy goal, instrument, or output) and a hypothesis containing the class definition and (optionally) corresponding themes provided between brackets.

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 3

Training results

Framework versions

• Transformers 4.56.1
• Pytorch 2.8.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.0