This is a SetFit model that can be used for Text Classification. This SetFit model uses Qwen/Qwen3-Embedding-0.6B 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 Type: SetFit
Sentence Transformer body: Qwen/Qwen3-Embedding-0.6B
Classification head: a LogisticRegression instance
Maximum Sequence Length: 32768 tokens
Number of Classes: 5 classes

Model Sources

Repository: SetFit on GitHub
Paper: Efficient Few-Shot Learning Without Prompts
Blogpost: SetFit: Efficient Few-Shot Learning Without Prompts

Model Labels

Label	Examples
Moderate negative	'1:21PM Growth downgrade due to slowdown in Eurozone\n\nGrowth is projected to slow in advanced economies to 2.0% in 2019 and 1.7% in 2020, and tick down to 4.5% in 2019 before improving to 4.9% in emerging markets and developing economies.' 'Informally, they say any likely credit impact is manageable, and that they are constructive on what the new finance minister and central bank governor can do in combination with the IMF, though they mention the IMF option should have been explored much sooner.' '22 Aug 2012 09:22 EDT DJ \nIMF Sees Greek Recession Continuing to 2015 - Sources\n\n22 Aug 2012 09:22 EDT DJ \nIMF Sees Greek Economic Recovery in 2016 - Sources\n\n22'
Strong negative	'Over time, it has become a crucial source of financial support for Argentina when it faces cash shortages.' 'The Soviet leadership was forced to liberalize to rescue the economy.' 'Going forward, the exchange rate adjustment rate will be carefully calibrated to facilitate the achievement of reserve accumulation and inflation reduction objectives, while real interest rates will remain at appropriately positive levels to continue to support demand for peso assets," said the head of the Fund.'
Moderate positive	'To recap, we suspect Trump’s proposed curbs on immigration and new tariffs would be stagflationary.' '"\n\nShe said the government's primary focus under Prime Minister Narendra Modi is achieving 'Viksit Bharat' by 2047.' 'Debt trades at stress values with high implied default probabilities.'
Neutral	'[ID:nLDE74Q11E]\n\nJuncker on Thursday warned European countries could not be counted on to step up and fill the gap left by the IMF if it decided against releasing its portion of the June aid tranche.' 'Thursday afternoon, the euro was at $1.3558 from $1.3703 late Wednesday, according to EBS via CQG.' 'She said good governance must be embedded into the legal system of Malaysia because that would be the best way to make sure it would survive the passing of time and the change of authorities over the course of time.'
Strong positive	'It is a huge success, because when the discussions first started four years ago, the amount was around $200 million, and now we have managed to raise it to $500 million.' "More than fear, the economic policy of today's government inspires curiosity and criticism of its recharged populism." 'Dar said the government would generate additional resources to meet the flood expenditure requirements before seeking help from the international community.'

Evaluation

Metrics

Label	Accuracy
all	0.3025

Uses

Direct Use for Inference

First install the SetFit library:

pip install setfit

Then you can load this model and run inference.

python

from setfit import SetFitModel

# Download from the 🤗 Hub
model = SetFitModel.from_pretrained("beethogedeon/reform-sent-Qwen3-0.6B")
# Run inference
preds = model("The IMF needs \"to remain very attentive\" over trade, says Ms Lagarde.")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	3	24.2946	116

Label	Training Sample Count
Moderate negative	100
Moderate positive	99
Neutral	100
Strong negative	100
Strong positive	100

Training Hyperparameters

batch_size: (8, 2)
num_epochs: (1, 16)
max_steps: 500
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: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.008	1	0.2758	-
0.4	50	0.2592	-
0.8	100	0.2532	-
1.2	150	0.2179	-
1.6	200	0.1754	-
2.0	250	0.14	-
2.4	300	0.069	-
2.8	350	0.028	-
3.2	400	0.0034	-
3.6	450	0.0013	-
4.0	500	0.0005	-

Framework Versions

Python: 3.11.5
SetFit: 1.1.3
Sentence Transformers: 5.1.2
Transformers: 4.53.3
PyTorch: 2.7.0
Datasets: 4.1.0
Tokenizers: 0.21.1

Citation

BibTeX

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}
}