README.md
20.6 KB · 184 lines · markdown Raw
1 ---
2 language:
3 - en
4 tags:
5 - text-classification
6 - zero-shot-classification
7 base_model: microsoft/deberta-v3-large
8 pipeline_tag: zero-shot-classification
9 library_name: transformers
10 license: mit
11 ---
12
13 # Model description: deberta-v3-large-zeroshot-v2.0
14
15 ## zeroshot-v2.0 series of models
16 Models in this series are designed for efficient zeroshot classification with the Hugging Face pipeline.
17 These models can do classification without training data and run on both GPUs and CPUs.
18 An overview of the latest zeroshot classifiers is available in my [Zeroshot Classifier Collection](https://huggingface.co/collections/MoritzLaurer/zeroshot-classifiers-6548b4ff407bb19ff5c3ad6f).
19
20 The main update of this `zeroshot-v2.0` series of models is that several models are trained on fully commercially-friendly data for users with strict license requirements.
21
22 These models can do one universal classification task: determine whether a hypothesis is "true" or "not true" given a text
23 (`entailment` vs. `not_entailment`).
24 This task format is based on the Natural Language Inference task (NLI).
25 The task is so universal that any classification task can be reformulated into this task by the Hugging Face pipeline.
26
27
28 ## Training data
29 Models with a "`-c`" in the name are trained on two types of fully commercially-friendly data:
30 1. Synthetic data generated with [Mixtral-8x7B-Instruct-v0.1](https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1).
31 I first created a list of 500+ diverse text classification tasks for 25 professions in conversations with Mistral-large. The data was manually curated.
32 I then used this as seed data to generate several hundred thousand texts for these tasks with Mixtral-8x7B-Instruct-v0.1.
33 The final dataset used is available in the [synthetic_zeroshot_mixtral_v0.1](https://huggingface.co/datasets/MoritzLaurer/synthetic_zeroshot_mixtral_v0.1) dataset
34 in the subset `mixtral_written_text_for_tasks_v4`. Data curation was done in multiple iterations and will be improved in future iterations.
35 2. Two commercially-friendly NLI datasets: ([MNLI](https://huggingface.co/datasets/nyu-mll/multi_nli), [FEVER-NLI](https://huggingface.co/datasets/fever)).
36 These datasets were added to increase generalization.
37 3. Models without a "`-c`" in the name also included a broader mix of training data with a broader mix of licenses: ANLI, WANLI, LingNLI,
38 and all datasets in [this list](https://github.com/MoritzLaurer/zeroshot-classifier/blob/7f82e4ab88d7aa82a4776f161b368cc9fa778001/v1_human_data/datasets_overview.csv)
39 where `used_in_v1.1==True`.
40
41
42 ## How to use the models
43 ```python
44 #!pip install transformers[sentencepiece]
45 from transformers import pipeline
46 text = "Angela Merkel is a politician in Germany and leader of the CDU"
47 hypothesis_template = "This text is about {}"
48 classes_verbalized = ["politics", "economy", "entertainment", "environment"]
49 zeroshot_classifier = pipeline("zero-shot-classification", model="MoritzLaurer/deberta-v3-large-zeroshot-v2.0") # change the model identifier here
50 output = zeroshot_classifier(text, classes_verbalized, hypothesis_template=hypothesis_template, multi_label=False)
51 print(output)
52 ```
53
54 `multi_label=False` forces the model to decide on only one class. `multi_label=True` enables the model to choose multiple classes.
55
56
57 ## Metrics
58
59 The models were evaluated on 28 different text classification tasks with the [f1_macro](https://scikit-learn.org/stable/modules/generated/sklearn.metrics.f1_score.html) metric.
60 The main reference point is `facebook/bart-large-mnli` which is, at the time of writing (03.04.24), the most used commercially-friendly 0-shot classifier.
61
62 ![results_aggreg_v2.0](https://raw.githubusercontent.com/MoritzLaurer/zeroshot-classifier/main/v2_synthetic_data/results/zeroshot-v2.0-aggreg.png)
63
64
65 | | facebook/bart-large-mnli | roberta-base-zeroshot-v2.0-c | roberta-large-zeroshot-v2.0-c | deberta-v3-base-zeroshot-v2.0-c | deberta-v3-base-zeroshot-v2.0 (fewshot) | deberta-v3-large-zeroshot-v2.0-c | deberta-v3-large-zeroshot-v2.0 (fewshot) | bge-m3-zeroshot-v2.0-c | bge-m3-zeroshot-v2.0 (fewshot) |
66 |:---------------------------|---------------------------:|-----------------------------:|------------------------------:|--------------------------------:|-----------------------------------:|---------------------------------:|------------------------------------:|-----------------------:|--------------------------:|
67 | all datasets mean | 0.497 | 0.587 | 0.622 | 0.619 | 0.643 (0.834) | 0.676 | 0.673 (0.846) | 0.59 | (0.803) |
68 | amazonpolarity (2) | 0.937 | 0.924 | 0.951 | 0.937 | 0.943 (0.961) | 0.952 | 0.956 (0.968) | 0.942 | (0.951) |
69 | imdb (2) | 0.892 | 0.871 | 0.904 | 0.893 | 0.899 (0.936) | 0.923 | 0.918 (0.958) | 0.873 | (0.917) |
70 | appreviews (2) | 0.934 | 0.913 | 0.937 | 0.938 | 0.945 (0.948) | 0.943 | 0.949 (0.962) | 0.932 | (0.954) |
71 | yelpreviews (2) | 0.948 | 0.953 | 0.977 | 0.979 | 0.975 (0.989) | 0.988 | 0.985 (0.994) | 0.973 | (0.978) |
72 | rottentomatoes (2) | 0.83 | 0.802 | 0.841 | 0.84 | 0.86 (0.902) | 0.869 | 0.868 (0.908) | 0.813 | (0.866) |
73 | emotiondair (6) | 0.455 | 0.482 | 0.486 | 0.459 | 0.495 (0.748) | 0.499 | 0.484 (0.688) | 0.453 | (0.697) |
74 | emocontext (4) | 0.497 | 0.555 | 0.63 | 0.59 | 0.592 (0.799) | 0.699 | 0.676 (0.81) | 0.61 | (0.798) |
75 | empathetic (32) | 0.371 | 0.374 | 0.404 | 0.378 | 0.405 (0.53) | 0.447 | 0.478 (0.555) | 0.387 | (0.455) |
76 | financialphrasebank (3) | 0.465 | 0.562 | 0.455 | 0.714 | 0.669 (0.906) | 0.691 | 0.582 (0.913) | 0.504 | (0.895) |
77 | banking77 (72) | 0.312 | 0.124 | 0.29 | 0.421 | 0.446 (0.751) | 0.513 | 0.567 (0.766) | 0.387 | (0.715) |
78 | massive (59) | 0.43 | 0.428 | 0.543 | 0.512 | 0.52 (0.755) | 0.526 | 0.518 (0.789) | 0.414 | (0.692) |
79 | wikitoxic_toxicaggreg (2) | 0.547 | 0.751 | 0.766 | 0.751 | 0.769 (0.904) | 0.741 | 0.787 (0.911) | 0.736 | (0.9) |
80 | wikitoxic_obscene (2) | 0.713 | 0.817 | 0.854 | 0.853 | 0.869 (0.922) | 0.883 | 0.893 (0.933) | 0.783 | (0.914) |
81 | wikitoxic_threat (2) | 0.295 | 0.71 | 0.817 | 0.813 | 0.87 (0.946) | 0.827 | 0.879 (0.952) | 0.68 | (0.947) |
82 | wikitoxic_insult (2) | 0.372 | 0.724 | 0.798 | 0.759 | 0.811 (0.912) | 0.77 | 0.779 (0.924) | 0.783 | (0.915) |
83 | wikitoxic_identityhate (2) | 0.473 | 0.774 | 0.798 | 0.774 | 0.765 (0.938) | 0.797 | 0.806 (0.948) | 0.761 | (0.931) |
84 | hateoffensive (3) | 0.161 | 0.352 | 0.29 | 0.315 | 0.371 (0.862) | 0.47 | 0.461 (0.847) | 0.291 | (0.823) |
85 | hatexplain (3) | 0.239 | 0.396 | 0.314 | 0.376 | 0.369 (0.765) | 0.378 | 0.389 (0.764) | 0.29 | (0.729) |
86 | biasframes_offensive (2) | 0.336 | 0.571 | 0.583 | 0.544 | 0.601 (0.867) | 0.644 | 0.656 (0.883) | 0.541 | (0.855) |
87 | biasframes_sex (2) | 0.263 | 0.617 | 0.835 | 0.741 | 0.809 (0.922) | 0.846 | 0.815 (0.946) | 0.748 | (0.905) |
88 | biasframes_intent (2) | 0.616 | 0.531 | 0.635 | 0.554 | 0.61 (0.881) | 0.696 | 0.687 (0.891) | 0.467 | (0.868) |
89 | agnews (4) | 0.703 | 0.758 | 0.745 | 0.68 | 0.742 (0.898) | 0.819 | 0.771 (0.898) | 0.687 | (0.892) |
90 | yahootopics (10) | 0.299 | 0.543 | 0.62 | 0.578 | 0.564 (0.722) | 0.621 | 0.613 (0.738) | 0.587 | (0.711) |
91 | trueteacher (2) | 0.491 | 0.469 | 0.402 | 0.431 | 0.479 (0.82) | 0.459 | 0.538 (0.846) | 0.471 | (0.518) |
92 | spam (2) | 0.505 | 0.528 | 0.504 | 0.507 | 0.464 (0.973) | 0.74 | 0.597 (0.983) | 0.441 | (0.978) |
93 | wellformedquery (2) | 0.407 | 0.333 | 0.333 | 0.335 | 0.491 (0.769) | 0.334 | 0.429 (0.815) | 0.361 | (0.718) |
94 | manifesto (56) | 0.084 | 0.102 | 0.182 | 0.17 | 0.187 (0.376) | 0.258 | 0.256 (0.408) | 0.147 | (0.331) |
95 | capsotu (21) | 0.34 | 0.479 | 0.523 | 0.502 | 0.477 (0.664) | 0.603 | 0.502 (0.686) | 0.472 | (0.644) |
96
97
98 These numbers indicate zeroshot performance, as no data from these datasets was added in the training mix.
99 Note that models without a "`-c`" in the title were evaluated twice: one run without any data from these 28 datasets to test pure zeroshot performance (the first number in the respective column) and
100 the final run including up to 500 training data points per class from each of the 28 datasets (the second number in brackets in the column, "fewshot"). No model was trained on test data.
101
102 Details on the different datasets are available here: https://github.com/MoritzLaurer/zeroshot-classifier/blob/main/v1_human_data/datasets_overview.csv
103
104
105 ## When to use which model
106
107 - **deberta-v3-zeroshot vs. roberta-zeroshot**: deberta-v3 performs clearly better than roberta, but it is a bit slower.
108 roberta is directly compatible with Hugging Face's production inference TEI containers and flash attention.
109 These containers are a good choice for production use-cases. tl;dr: For accuracy, use a deberta-v3 model.
110 If production inference speed is a concern, you can consider a roberta model (e.g. in a TEI container and [HF Inference Endpoints](https://ui.endpoints.huggingface.co/catalog)).
111 - **commercial use-cases**: models with "`-c`" in the title are guaranteed to be trained on only commercially-friendly data.
112 Models without a "`-c`" were trained on more data and perform better, but include data with non-commercial licenses.
113 Legal opinions diverge if this training data affects the license of the trained model. For users with strict legal requirements,
114 the models with "`-c`" in the title are recommended.
115 - **Multilingual/non-English use-cases**: use [bge-m3-zeroshot-v2.0](https://huggingface.co/MoritzLaurer/bge-m3-zeroshot-v2.0) or [bge-m3-zeroshot-v2.0-c](https://huggingface.co/MoritzLaurer/bge-m3-zeroshot-v2.0-c).
116 Note that multilingual models perform worse than English-only models. You can therefore also first machine translate your texts to English with libraries like [EasyNMT](https://github.com/UKPLab/EasyNMT)
117 and then apply any English-only model to the translated data. Machine translation also facilitates validation in case your team does not speak all languages in the data.
118 - **context window**: The `bge-m3` models can process up to 8192 tokens. The other models can process up to 512. Note that longer text inputs both make the
119 mode slower and decrease performance, so if you're only working with texts of up to 400~ words / 1 page, use e.g. a deberta model for better performance.
120 - The latest updates on new models are always available in the [Zeroshot Classifier Collection](https://huggingface.co/collections/MoritzLaurer/zeroshot-classifiers-6548b4ff407bb19ff5c3ad6f).
121
122
123
124
125 ## Reproduction
126
127 Reproduction code is available in the `v2_synthetic_data` directory here: https://github.com/MoritzLaurer/zeroshot-classifier/tree/main
128
129
130 ## Limitations and bias
131 The model can only do text classification tasks.
132
133 Biases can come from the underlying foundation model, the human NLI training data and the synthetic data generated by Mixtral.
134
135
136
137 ## License
138 The foundation model was published under the MIT license.
139 The licenses of the training data vary depending on the model, see above.
140
141
142 ## Citation
143
144 This model is an extension of the research described in this [paper](https://arxiv.org/pdf/2312.17543.pdf).
145
146 If you use this model academically, please cite:
147 ```
148 @misc{laurer_building_2023,
149 title = {Building {Efficient} {Universal} {Classifiers} with {Natural} {Language} {Inference}},
150 url = {http://arxiv.org/abs/2312.17543},
151 doi = {10.48550/arXiv.2312.17543},
152 abstract = {Generative Large Language Models (LLMs) have become the mainstream choice for fewshot and zeroshot learning thanks to the universality of text generation. Many users, however, do not need the broad capabilities of generative LLMs when they only want to automate a classification task. Smaller BERT-like models can also learn universal tasks, which allow them to do any text classification task without requiring fine-tuning (zeroshot classification) or to learn new tasks with only a few examples (fewshot), while being significantly more efficient than generative LLMs. This paper (1) explains how Natural Language Inference (NLI) can be used as a universal classification task that follows similar principles as instruction fine-tuning of generative LLMs, (2) provides a step-by-step guide with reusable Jupyter notebooks for building a universal classifier, and (3) shares the resulting universal classifier that is trained on 33 datasets with 389 diverse classes. Parts of the code we share has been used to train our older zeroshot classifiers that have been downloaded more than 55 million times via the Hugging Face Hub as of December 2023. Our new classifier improves zeroshot performance by 9.4\%.},
153 urldate = {2024-01-05},
154 publisher = {arXiv},
155 author = {Laurer, Moritz and van Atteveldt, Wouter and Casas, Andreu and Welbers, Kasper},
156 month = dec,
157 year = {2023},
158 note = {arXiv:2312.17543 [cs]},
159 keywords = {Computer Science - Artificial Intelligence, Computer Science - Computation and Language},
160 }
161 ```
162
163 ### Ideas for cooperation or questions?
164 If you have questions or ideas for cooperation, contact me at moritz{at}huggingface{dot}co or [LinkedIn](https://www.linkedin.com/in/moritz-laurer/)
165
166
167 ### Flexible usage and "prompting"
168 You can formulate your own hypotheses by changing the `hypothesis_template` of the zeroshot pipeline.
169 Similar to "prompt engineering" for LLMs, you can test different formulations of your `hypothesis_template` and verbalized classes to improve performance.
170
171 ```python
172 from transformers import pipeline
173 text = "Angela Merkel is a politician in Germany and leader of the CDU"
174 # formulation 1
175 hypothesis_template = "This text is about {}"
176 classes_verbalized = ["politics", "economy", "entertainment", "environment"]
177 # formulation 2 depending on your use-case
178 hypothesis_template = "The topic of this text is {}"
179 classes_verbalized = ["political activities", "economic policy", "entertainment or music", "environmental protection"]
180 # test different formulations
181 zeroshot_classifier = pipeline("zero-shot-classification", model="MoritzLaurer/deberta-v3-large-zeroshot-v2.0") # change the model identifier here
182 output = zeroshot_classifier(text, classes_verbalized, hypothesis_template=hypothesis_template, multi_label=False)
183 print(output)
184 ```