SetFit with sentence-transformers/paraphrase-mpnet-base-v2
This is a SetFit model that can be used for Text Classification. This SetFit model uses sentence-transformers/paraphrase-mpnet-base-v2 as the Sentence Transformer embedding model. A LogisticRegression instance is used for classification.
The model has been trained using an efficient few-shot learning technique that involves:
- Fine-tuning a Sentence Transformer with contrastive learning.
- Training a classification head with features from the fine-tuned Sentence Transformer.
Model Details
Model Description
Model Sources
Model Labels
Label |
Examples |
1 |
- ' due diligence and weak monitoring control, surveillance and enforcement systems by coastal, flag, and port states," it noted.Human rights abuses are also driven by an array of other factors, it added. The Thai fishing industry is structurally dependent on unskilled workers, a result of a failure to invest in technology to increase labour productivity as well as an abundance of cheap migrant workers from the less-developed neighbours.At the same time, vessel operators face a chronic shortage of workers -- a deficit estimated by the National Fishing Association of Thailand (NFAT) to be as high as 50,000."Combined with economic pressure arising from the degradation of marine resources in the Thai exclusive economic zone (EEZ), these factors shape the prevalence of labour abuses and the use of trafficking, forced and bonded labour in the Thai fishing industry."As migrant workers from Myanmar, Cambodia, Laos and undeveloped rural regions of Thailand, particularly the Northeast, are trafficked through Indonesia, both countries must collaborate further to address abuses, said Mark Dia, Greenpeace's Regional Oceans Campaign coordinator for Southeast Asia.Greenpeace has worked with the Indonesian government on the matter while the Thai military government has already taken action to address chronic problems facing the fishing industry for many years'
- ' for all forms of trafficking, including forced and bonded labour, respecting due process.Forced labour constitutes India's largest trafficking problem; men, women, and children in debt bondage- sometimes inherited from previous generations- are forced to work in brick kilns, rice mills, agriculture, and embroidery units, it said.The majority of India's trafficking problem is internal, and those from the most disadvantaged social strata- Dalits, members of tribal communities, religious minorities, and women and girls from excluded groups- are most vulnerable, it added."Within India, some are subjected to forced labour in sectors such as construction, steel, and textile industries; wire manufacturing for underground cables; biscuit factories; pickling; floriculture; fish farms; and ship breaking," said the State Department.Thousands of unregulated work placement agencies reportedly lure adults and children under false promises of employment for sex trafficking or forced labour, including domestic servitude.In addition to bonded labour, some children are subjected to forced labour as factory and agricultural workers, domestic servants, and beggars. Begging ringleaders sometimes maim children to earn more money."Some NGOs and media report girls are sold and forced to conceive and deliver babies for sale. Conditions amounting to'
- " supplies.\xa0Americans buying Hawaiian seafood are almost certainly eating fish caught by one of these workers.'We want the same standards as the other workers in America, but we are just small people working there,' said fisherman Syamsul Maarif, who didn't get paid for four months.\xa0He was sent back to his Indonesian village after nearly dying at sea when his Hawaiian boat sank earlier this year.Because they have no visas, the men can't fly into Hawaii, so they're brought by boat.\xa0And since they are not technically in the country, they're at the mercy of their American captains on American-flagged, American-owned vessels, catching choice swordfish and ahi tuna that can fetch more than $1,000 apiece.\xa0The entire system contradicts other state and federal laws, yet operates with the blessing of U.S. officials and law enforcement.'People say these fishermen can't leave their boats, they're like captives,' said U.S. Attorney Florence Nakakuni in Hawaii.\xa0'But they don't have visas, so they can't leave their boat, really.'Each of the roughly 140 boats in the fleet docks about once every three weeks, occasionally at ports"
|
0 |
- ' forced marriage and other such exploitation. In a DW interview, Fiona David, executive director of global research at the Walk Free Foundation, calls on the countries in the region to step up their efforts to combat the problem and put in place the mechanisms that would require businesses to focus on the issues of slavery and forced labor throughout their supply chains. DW: According to the 2016 Global Slavery Index, nearly two-thirds of modern slaves are living in Asian countries. What are the reasons behind the high prevalence of this phenomenon in the region? Fiona David: The Asia-Pacific is the most populous region in the world, and it is also well integrated into the global supply chains. We do estimate that about two-thirds of the nearly 46 million people trapped in slavery are in Asia. And we see all forms of modern slavery in the region, such as forced labor in brick kilns, child beggars in Afghanistan and India, bonded labor in the agricultural as well as garment sectors. Given its population size and integration into global value chains, the Asia-Pacific is a region where a lot of low cost labor is made available to produce the goods and services that we all consume. What kind of living and working conditions do these people find themselves in? They experience miserable'
- " going to cook those up for\xa0\xa0\xa0\xa0\xa0 dinner?' I said, no, that's bait for fishing'. He thought we were\xa0\xa0\xa0\xa0\xa0\xa0\xa0\xa0\xa0\xa0\xa0 going to cook and eat those frozen prawns!\xa0\xa0\xa0\xa0\xa0\xa0\xa0\xa0\xa0\xa0\xa0\xa0\xa0Signs to look for when fish are biting, as well as learning about how to catch and name fish, are among other activities, as is counting and comparing the yield, talking about how many fish should be cooked on the spot, or taken home to family members. In addition, there is the preparation and cooking of the fish once caught: gutting and singeing it when required, making a coal fire, building a bush oven, timing the bake, turning the fish, and finding leaves and bark from local trees to plate cooked fish (Toussaint 2010; Toussaint et al. 2005; Yu 2006). The eating and sharing of fish, enjoying its tastiness, or comparing the quality of one fish to another, marks the culmination of a good family day, a point I develop below. All of these activities, whether seen on their own as individual parts, or brought together as a whole, reveal their experiential importance to those families who are directly involved, as well"
- "8 million in modern slaveryPakistan2,000,000 - 2,200,000Bonded labour affects men, women and children largely from rural areas who travel to cities to find work, and has been reported in many industries, primarily brick kilns, but also in agriculture, fisheries and mining.Ethiopia620,000 - 680,000Domestic workers travelling under illegal private employment agencies are particularly vulnerable as are girls who can be subjected to child marriage.Nigeria670,000 - 740,000An estimated 15.88% of the estimated total 29.6 million people in modern slavery are in Sub-Saharan Africa.Bangladesh330,000 - 360,000Large numbers of women and girls are reportedly trafficked to India and Pakistan annually and children, including boys, are exploited and trafficked for sex and labour.Democratic Republic of Congo440,000 - 490,000One of the world's poorest countries, despite a wealth of resources; 90% of men working in mines in eastern DRC are trapped by debt bondage.India13,300,000 - 14,700,000Men, women and children, many enslaved"
|
Evaluation
Metrics
Uses
Direct Use for Inference
First install the SetFit library:
pip install setfit
Then you can load this model and run inference.
from setfit import SetFitModel
model = SetFitModel.from_pretrained("JoshELambert/forced-labor")
preds = model(" in their own villages by debt bondage or born ito slavery, work in construction, textiles, brick-making, mines, fish and prawn processing and hospitality.Russia490,000 - 540,000Migrant workers endure extortion and physical abuse; anecdotal evidence suggests that forced labour camps still operate in Siberia.China2,800,000 - 3,100,000Severe forced labour in brick kilns in the north; forced labour in modern industries including fashion and computer supply chains.Myanmar360,000 - 400,000Slavery includes reports of deceptive recruitment of women for sale as brides in China, forced labour of adults on plantations and in industry and forced labour of children in tea shops, home industries and as beggars.Thailand450,000 - 500,000An explosion in global demand for seafood has led to an increased need for cheap migrant labour, including on fishing boats. High numbers of children are exploited, particulary those from ethnic minorities and hill tribes.SOURCE: THE GLOBAL SLAVERY INDEX 2013")
Training Details
Training Set Metrics
Training set |
Min |
Median |
Max |
Word count |
50 |
189.8442 |
221 |
Label |
Training Sample Count |
0 |
8 |
1 |
69 |
Training Hyperparameters
- batch_size: (16, 16)
- num_epochs: (4, 4)
- max_steps: -1
- sampling_strategy: oversampling
- body_learning_rate: (2e-05, 1e-05)
- head_learning_rate: 0.01
- loss: CosineSimilarityLoss
- distance_metric: cosine_distance
- margin: 0.25
- end_to_end: False
- use_amp: False
- warmup_proportion: 0.1
- l2_weight: 0.01
- seed: 42
- eval_max_steps: -1
- load_best_model_at_end: True
Training Results
Epoch |
Step |
Training Loss |
Validation Loss |
0.0033 |
1 |
0.1808 |
- |
0.1629 |
50 |
0.1363 |
- |
0.3257 |
100 |
0.0103 |
- |
0.4886 |
150 |
0.0019 |
- |
0.6515 |
200 |
0.0005 |
- |
0.8143 |
250 |
0.0001 |
- |
0.9772 |
300 |
0.0 |
- |
1.0 |
307 |
- |
0.0407 |
1.1401 |
350 |
0.0001 |
- |
1.3029 |
400 |
0.0 |
- |
1.4658 |
450 |
0.0 |
- |
1.6287 |
500 |
0.0 |
- |
1.7915 |
550 |
0.0 |
- |
1.9544 |
600 |
0.0 |
- |
2.0 |
614 |
- |
0.0272 |
2.1173 |
650 |
0.0 |
- |
2.2801 |
700 |
0.0 |
- |
2.4430 |
750 |
0.0 |
- |
2.6059 |
800 |
0.0 |
- |
2.7687 |
850 |
0.0 |
- |
2.9316 |
900 |
0.0 |
- |
3.0 |
921 |
- |
0.0238 |
3.0945 |
950 |
0.0 |
- |
3.2573 |
1000 |
0.0 |
- |
3.4202 |
1050 |
0.0 |
- |
3.5831 |
1100 |
0.0 |
- |
3.7459 |
1150 |
0.0 |
- |
3.9088 |
1200 |
0.0 |
- |
4.0 |
1228 |
- |
0.0227 |
Framework Versions
- Python: 3.10.12
- SetFit: 1.1.0
- Sentence Transformers: 3.3.1
- Transformers: 4.42.2
- PyTorch: 2.5.1+cu121
- Datasets: 3.2.0
- Tokenizers: 0.19.1
Citation
BibTeX
@article{https://doi.org/10.48550/arxiv.2209.11055,
doi = {10.48550/ARXIV.2209.11055},
url = {https://arxiv.org/abs/2209.11055},
author = {Tunstall, Lewis and Reimers, Nils and Jo, Unso Eun Seo and Bates, Luke and Korat, Daniel and Wasserblat, Moshe and Pereg, Oren},
keywords = {Computation and Language (cs.CL), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {Efficient Few-Shot Learning Without Prompts},
publisher = {arXiv},
year = {2022},
copyright = {Creative Commons Attribution 4.0 International}
}