Tagalog homograph multiple choice

Full fine-tuning of FacebookAI/roberta-base. Input: bare sentence paired with Target: {homograph}. Candidate: {choice}. Select the candidate with highest logit. Use the exported tokenizer and mcq_spec.json.

Smoke-test model: False. Best checkpoint selected by validation accuracy. Training and validation data hashes are in training_config.json. The benchmark is not used for training or checkpoint selection.

Downloads last month
40
Safetensors
Model size
0.1B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for tdr-research-team/roberta-base-mcq-tdr-finetuned

Finetuned
(2399)
this model