TheoLvs/wsl-prescreening-multi-v0.0

SetFit with BAAI/bge-small-en-v1.5

This is a SetFit model that can be used for Text Classification. This SetFit model uses BAAI/bge-small-en-v1.5 as the Sentence Transformer embedding model. A MultiOutputClassifier instance is used for classification.

The model has been trained using an efficient few-shot learning technique that involves:

1. Fine-tuning a Sentence Transformer with contrastive learning.
2. Training a classification head with features from the fine-tuned Sentence Transformer.

Model Details

Model Description

• Model Type: SetFit
• Sentence Transformer body: BAAI/bge-small-en-v1.5
• Classification head: a MultiOutputClassifier instance
• Maximum Sequence Length: 512 tokens

Model Sources

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

Evaluation

Metrics

Label	Accuracy
all	0.8485

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

# Download from the 🤗 Hub
model = SetFitModel.from_pretrained("TheoLvs/wsl-prescreening-multi-v0.0")
# Run inference
preds = model("Wetland carbon sequestration capacity shows non-linear response to restoration technique and hydrological regime This study examines carbon sequestration outcomes from 124 wetland restoration projects across North America, Europe, and Asia over a 15-year monitoring period. Using standardized carbon flux measurements and sediment coring, we quantified how restoration approach and hydrological management influence carbon accumulation rates. Results demonstrate that restoration technique explained 53% of variance in carbon sequestration outcomes, with significant interaction effects between technique and hydroperiod. Projects restoring natural hydrological fluctuations achieved 2.7 times higher carbon accumulation rates than those maintaining static water levels. Vegetation community composition emerged as a significant mediating variable, with diverse native assemblages sequestering 34% more carbon than simplified or non-native communities. Our findings indicate that wetland restoration prioritizing hydrological dynamism and diverse vegetation delivers superior climate mitigation benefits while simultaneously enhancing habitat value and water quality functions.")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	90	191.8561	348

Training Hyperparameters

• batch_size: (8, 8)
• num_epochs: (5, 5)
• max_steps: 5000
• 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
0.0006	1	0.158	-
0.0288	50	0.2511	-
0.0575	100	0.215	-
0.0863	150	0.1883	-
0.1151	200	0.165	-
0.1438	250	0.1274	-
0.1726	300	0.0801	-
0.2014	350	0.0635	-
0.2301	400	0.0427	-
0.2589	450	0.0355	-
0.2877	500	0.0337	-
0.3165	550	0.0271	-
0.3452	600	0.0069	-
0.3740	650	0.0032	-
0.4028	700	0.0033	-
0.4315	750	0.0027	-
0.4603	800	0.0022	-
0.4891	850	0.002	-
0.5178	900	0.0019	-
0.5466	950	0.0017	-
0.5754	1000	0.0017	-
0.6041	1050	0.0015	-
0.6329	1100	0.0015	-
0.6617	1150	0.0013	-
0.6904	1200	0.0013	-
0.7192	1250	0.0014	-
0.7480	1300	0.0012	-
0.7768	1350	0.0012	-
0.8055	1400	0.0011	-
0.8343	1450	0.0012	-
0.8631	1500	0.0011	-
0.8918	1550	0.0011	-
0.9206	1600	0.0011	-
0.9494	1650	0.001	-
0.9781	1700	0.001	-
1.0069	1750	0.001	-
1.0357	1800	0.001	-
1.0644	1850	0.0009	-
1.0932	1900	0.0009	-
1.1220	1950	0.0009	-
1.1507	2000	0.0009	-
1.1795	2050	0.0009	-
1.2083	2100	0.0009	-
1.2371	2150	0.0008	-
1.2658	2200	0.0009	-
1.2946	2250	0.0008	-
1.3234	2300	0.0008	-
1.3521	2350	0.0008	-
1.3809	2400	0.0008	-
1.4097	2450	0.0008	-
1.4384	2500	0.0008	-
1.4672	2550	0.0007	-
1.4960	2600	0.0007	-
1.5247	2650	0.0007	-
1.5535	2700	0.0007	-
1.5823	2750	0.0007	-
1.6110	2800	0.0007	-
1.6398	2850	0.0007	-
1.6686	2900	0.0007	-
1.6974	2950	0.0007	-
1.7261	3000	0.0006	-
1.7549	3050	0.0007	-
1.7837	3100	0.0007	-
1.8124	3150	0.0007	-
1.8412	3200	0.0007	-
1.8700	3250	0.0007	-
1.8987	3300	0.0006	-
1.9275	3350	0.0006	-
1.9563	3400	0.0006	-
1.9850	3450	0.0006	-
2.0138	3500	0.0006	-
2.0426	3550	0.0006	-
2.0713	3600	0.0006	-
2.1001	3650	0.0006	-
2.1289	3700	0.0006	-
2.1577	3750	0.0006	-
2.1864	3800	0.0006	-
2.2152	3850	0.0006	-
2.2440	3900	0.0006	-
2.2727	3950	0.0006	-
2.3015	4000	0.0006	-
2.3303	4050	0.0006	-
2.3590	4100	0.0006	-
2.3878	4150	0.0006	-
2.4166	4200	0.0005	-
2.4453	4250	0.0006	-
2.4741	4300	0.0005	-
2.5029	4350	0.0006	-
2.5316	4400	0.0006	-
2.5604	4450	0.0005	-
2.5892	4500	0.0005	-
2.6180	4550	0.0005	-
2.6467	4600	0.0005	-
2.6755	4650	0.0005	-
2.7043	4700	0.0005	-
2.7330	4750	0.0005	-
2.7618	4800	0.0005	-
2.7906	4850	0.0005	-
2.8193	4900	0.0005	-
2.8481	4950	0.0005	-
2.8769	5000	0.0005	-

Framework Versions

• Python: 3.11.12
• SetFit: 1.1.2
• Sentence Transformers: 4.1.0
• Transformers: 4.45.2
• PyTorch: 2.6.0+cu124
• Datasets: 3.6.0
• Tokenizers: 0.20.3

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}
}