# Load model directly
from transformers import AutoTokenizer, RobertaForMultiLabelSequenceClassification
tokenizer = AutoTokenizer.from_pretrained("AMUseBot/roberta-base-cookdial-v1_1")
model = RobertaForMultiLabelSequenceClassification.from_pretrained("AMUseBot/roberta-base-cookdial-v1_1")Quick Links
- Baseline NLU model for the "AMUseBot" cooking taskbot prototype. Updated version with more robust req_ingredient intent recognition thanks to finetuning with extra synthetic data.
roberta-basemodel finetuned with default hyperparameters for 7 epochs on intents from the CookDial (https://github.com/YiweiJiang2015/CookDial) dataset with an extra choose_recipe intent added. Thesimpletransformerslibrary was used for fine-tuning.- Intent mapping: {"0": "affirm", "1": "choose_recipe", "2": "confirm", "3": "goodbye", "4": "greeting", "5": "negate", "6": "other", "7": "req_amount", "8": "req_duration", "9": "req_ingredient", "10": "req_ingredient_list", "11": "req_ingredient_list_ends", "12": "req_ingredient_list_length", "13": "req_instruction", "14": "req_is_recipe_finished", "15": "req_is_recipe_ongoing", "16": "req_parallel_action", "17": "req_repeat", "18": "req_start", "19": "req_substitute", "20": "req_temperature", "21": "req_title", "22": "req_tool", "23": "req_use_all", "24": "thank"}.
- Downloads last month
- 6
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="AMUseBot/roberta-base-cookdial-v1_1")