tasksource/deberta-small-long-nli

Model Card for DeBERTa-v3-small-tasksource-nli

DeBERTa-v3-small with context length of 1680 tokens fine-tuned on tasksource for 250k steps. I oversampled long NLI tasks (ConTRoL, doc-nli). Training data include HelpSteer v1/v2, logical reasoning tasks (FOLIO, FOL-nli, LogicNLI...), OASST, hh/rlhf, linguistics oriented NLI tasks, tasksource-dpo, fact verification tasks.

This model is suitable for long context NLI or as a backbone for reward models or classifiers fine-tuning.

This checkpoint has strong zero-shot validation performance on many tasks (e.g. 70% on WNLI), and can be used for:

• Zero-shot entailment-based classification for arbitrary labels [ZS].
• Natural language inference [NLI]
• Further fine-tuning on a new task or tasksource task (classification, token classification or multiple-choice) [FT].

test_name	accuracy
anli/a1	57.2
anli/a2	46.1
anli/a3	47.2
nli_fever	71.7
FOLIO	47.1
ConTRoL-nli	52.2
cladder	52.8
zero-shot-label-nli	70.0
chatbot_arena_conversations	67.8
oasst2_pairwise_rlhf_reward	75.6
doc-nli	75.0

Zero-shot GPT-4 scores 61% on FOLIO (logical reasoning), 62% on cladder (probabilistic reasoning) and 56.4% on ConTRoL (long context NLI).

[ZS] Zero-shot classification pipeline

from transformers import pipeline
classifier = pipeline("zero-shot-classification",model="tasksource/deberta-small-long-nli")

text = "one day I will see the world"
candidate_labels = ['travel', 'cooking', 'dancing']
classifier(text, candidate_labels)

NLI training data of this model includes label-nli, a NLI dataset specially constructed to improve this kind of zero-shot classification.

[NLI] Natural language inference pipeline

from transformers import pipeline
pipe = pipeline("text-classification",model="tasksource/deberta-small-long-nli")
pipe([dict(text='there is a cat',
  text_pair='there is a black cat')]) #list of (premise,hypothesis)
# [{'label': 'neutral', 'score': 0.9952911138534546}]

[FT] Tasknet: 3 lines fine-tuning

# !pip install tasknet
import tasknet as tn
hparams=dict(model_name='tasksource/deberta-small-long-nli', learning_rate=2e-5)
model, trainer = tn.Model_Trainer([tn.AutoTask("glue/rte")], hparams)
trainer.train()

Software and training details

The model was trained on 600 tasks for 250k steps with a batch size of 384 and a peak learning rate of 2e-5. Training took 14 days on Nvidia A30 24GB gpu. This is the shared model with the MNLI classifier on top. Each task had a specific CLS embedding, which is dropped 10% of the time to facilitate model use without it. All multiple-choice model used the same classification layers. For classification tasks, models shared weights if their labels matched.

https://github.com/sileod/tasksource/
https://github.com/sileod/tasknet/
Training code: https://colab.research.google.com/drive/1iB4Oxl9_B5W3ZDzXoWJN-olUbqLBxgQS?usp=sharing

Citation

More details on this article:

@inproceedings{sileo-2024-tasksource,
    title = "tasksource: A Large Collection of {NLP} tasks with a Structured Dataset Preprocessing Framework",
    author = "Sileo, Damien",
    booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)",
    month = may,
    year = "2024",
    address = "Torino, Italia",
    publisher = "ELRA and ICCL",
    url = "https://aclanthology.org/2024.lrec-main.1361",
    pages = "15655--15684",
}

Model Card Contact

damien.sileo@inria.fr