Mazen619/hope-setfit-finance

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 LogisticRegression 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 LogisticRegression instance
• Maximum Sequence Length: 512 tokens
• Number of Classes: 4 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
liquidity_agent	'Show me working capital by division for Q1 2025.' "What's our cash burn rate this quarter?" 'What is the cash conversion cycle for Mutual Fund Series 1?'
product_agent	'Set up a new product bundle including Insurance Plan Alpha.' 'What is the product code for Insurance Plan Alpha?' 'Create a limited-time pricing offer for the Student Savings Account.'
knowledge_agent	'Hey there!' 'Can you explain what NSFR means in banking?' 'What is a reserve requirement, generally speaking?'
profitability_agent	'Calculate the interest margin on the mortgage book for the last 2 months.' 'Confirm our EBITDA for last quarter.' 'What was the yield on earning assets for FY2025?'

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("setfit_model_id")
# Run inference
preds = model("What was the yield on earning assets for FY2025?")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	1	8.4	17

Label	Training Sample Count
profitability_agent	80
liquidity_agent	80
product_agent	80
knowledge_agent	80

Training Hyperparameters

• batch_size: (16, 16)
• num_epochs: (3, 3)
• 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: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0002	1	0.1857	-
0.0104	50	0.2271	-
0.0208	100	0.2321	-
0.0312	150	0.2251	-
0.0417	200	0.217	-
0.0521	250	0.2034	-
0.0625	300	0.1863	-
0.0729	350	0.1624	-
0.0833	400	0.1316	-
0.0938	450	0.0843	-
0.1042	500	0.0878	-
0.1146	550	0.0779	-
0.125	600	0.0626	-
0.1354	650	0.0538	-
0.1458	700	0.039	-
0.1562	750	0.025	-
0.1667	800	0.0096	-
0.1771	850	0.0061	-
0.1875	900	0.0043	-
0.1979	950	0.0036	-
0.2083	1000	0.0031	-
0.2188	1050	0.0029	-
0.2292	1100	0.0024	-
0.2396	1150	0.0022	-
0.25	1200	0.002	-
0.2604	1250	0.0018	-
0.2708	1300	0.0016	-
0.2812	1350	0.0015	-
0.2917	1400	0.0014	-
0.3021	1450	0.0014	-
0.3125	1500	0.0013	-
0.3229	1550	0.0012	-
0.3333	1600	0.0012	-
0.3438	1650	0.0011	-
0.3542	1700	0.001	-
0.3646	1750	0.001	-
0.375	1800	0.001	-
0.3854	1850	0.0009	-
0.3958	1900	0.001	-
0.4062	1950	0.0009	-
0.4167	2000	0.0008	-
0.4271	2050	0.0008	-
0.4375	2100	0.0008	-
0.4479	2150	0.0008	-
0.4583	2200	0.0008	-
0.4688	2250	0.0008	-
0.4792	2300	0.0008	-
0.4896	2350	0.0007	-
0.5	2400	0.0007	-
0.5104	2450	0.0006	-
0.5208	2500	0.0007	-
0.5312	2550	0.0006	-
0.5417	2600	0.0006	-
0.5521	2650	0.0006	-
0.5625	2700	0.0006	-
0.5729	2750	0.0006	-
0.5833	2800	0.0006	-
0.5938	2850	0.0006	-
0.6042	2900	0.0006	-
0.6146	2950	0.0006	-
0.625	3000	0.0006	-
0.6354	3050	0.0005	-
0.6458	3100	0.0005	-
0.6562	3150	0.0005	-
0.6667	3200	0.0005	-
0.6771	3250	0.0005	-
0.6875	3300	0.0005	-
0.6979	3350	0.0005	-
0.7083	3400	0.0005	-
0.7188	3450	0.0005	-
0.7292	3500	0.0004	-
0.7396	3550	0.0004	-
0.75	3600	0.0004	-
0.7604	3650	0.0005	-
0.7708	3700	0.0004	-
0.7812	3750	0.0004	-
0.7917	3800	0.0004	-
0.8021	3850	0.0004	-
0.8125	3900	0.0005	-
0.8229	3950	0.0004	-
0.8333	4000	0.0004	-
0.8438	4050	0.0004	-
0.8542	4100	0.0004	-
0.8646	4150	0.0004	-
0.875	4200	0.0004	-
0.8854	4250	0.0004	-
0.8958	4300	0.0004	-
0.9062	4350	0.0004	-
0.9167	4400	0.0004	-
0.9271	4450	0.0004	-
0.9375	4500	0.0004	-
0.9479	4550	0.0004	-
0.9583	4600	0.0004	-
0.9688	4650	0.0004	-
0.9792	4700	0.0004	-
0.9896	4750	0.0004	-
1.0	4800	0.0004	-
1.0104	4850	0.0003	-
1.0208	4900	0.0004	-
1.0312	4950	0.0004	-
1.0417	5000	0.0004	-
1.0521	5050	0.0004	-
1.0625	5100	0.0003	-
1.0729	5150	0.0003	-
1.0833	5200	0.0003	-
1.0938	5250	0.0003	-
1.1042	5300	0.0003	-
1.1146	5350	0.0003	-
1.125	5400	0.0003	-
1.1354	5450	0.0003	-
1.1458	5500	0.0004	-
1.1562	5550	0.0003	-
1.1667	5600	0.0003	-
1.1771	5650	0.0003	-
1.1875	5700	0.0003	-
1.1979	5750	0.0003	-
1.2083	5800	0.0003	-
1.2188	5850	0.0003	-
1.2292	5900	0.0003	-
1.2396	5950	0.0003	-
1.25	6000	0.0003	-
1.2604	6050	0.0003	-
1.2708	6100	0.0003	-
1.2812	6150	0.0003	-
1.2917	6200	0.0003	-
1.3021	6250	0.0003	-
1.3125	6300	0.0003	-
1.3229	6350	0.0003	-
1.3333	6400	0.0003	-
1.3438	6450	0.0003	-
1.3542	6500	0.0003	-
1.3646	6550	0.0003	-
1.375	6600	0.0003	-
1.3854	6650	0.0003	-
1.3958	6700	0.0003	-
1.4062	6750	0.0003	-
1.4167	6800	0.0003	-
1.4271	6850	0.0003	-
1.4375	6900	0.0003	-
1.4479	6950	0.0003	-
1.4583	7000	0.0003	-
1.4688	7050	0.0003	-
1.4792	7100	0.0003	-
1.4896	7150	0.0003	-
1.5	7200	0.0003	-
1.5104	7250	0.0003	-
1.5208	7300	0.0003	-
1.5312	7350	0.0003	-
1.5417	7400	0.0003	-
1.5521	7450	0.0003	-
1.5625	7500	0.0003	-
1.5729	7550	0.0003	-
1.5833	7600	0.0003	-
1.5938	7650	0.0003	-
1.6042	7700	0.0003	-
1.6146	7750	0.0003	-
1.625	7800	0.0003	-
1.6354	7850	0.0003	-
1.6458	7900	0.0003	-
1.6562	7950	0.0003	-
1.6667	8000	0.0003	-
1.6771	8050	0.0003	-
1.6875	8100	0.0003	-
1.6979	8150	0.0003	-
1.7083	8200	0.0003	-
1.7188	8250	0.0003	-
1.7292	8300	0.0003	-
1.7396	8350	0.0003	-
1.75	8400	0.0002	-
1.7604	8450	0.0002	-
1.7708	8500	0.0003	-
1.7812	8550	0.0003	-
1.7917	8600	0.0003	-
1.8021	8650	0.0003	-
1.8125	8700	0.0003	-
1.8229	8750	0.0003	-
1.8333	8800	0.0002	-
1.8438	8850	0.0003	-
1.8542	8900	0.0003	-
1.8646	8950	0.0002	-
1.875	9000	0.0002	-
1.8854	9050	0.0002	-
1.8958	9100	0.0002	-
1.9062	9150	0.0003	-
1.9167	9200	0.0003	-
1.9271	9250	0.0002	-
1.9375	9300	0.0002	-
1.9479	9350	0.0002	-
1.9583	9400	0.0002	-
1.9688	9450	0.0002	-
1.9792	9500	0.0003	-
1.9896	9550	0.0002	-
2.0	9600	0.0002	-
2.0104	9650	0.0002	-
2.0208	9700	0.0002	-
2.0312	9750	0.0002	-
2.0417	9800	0.0002	-
2.0521	9850	0.0002	-
2.0625	9900	0.0002	-
2.0729	9950	0.0002	-
2.0833	10000	0.0002	-
2.0938	10050	0.0002	-
2.1042	10100	0.0002	-
2.1146	10150	0.0002	-
2.125	10200	0.0002	-
2.1354	10250	0.0002	-
2.1458	10300	0.0002	-
2.1562	10350	0.0002	-
2.1667	10400	0.0002	-
2.1771	10450	0.0002	-
2.1875	10500	0.0002	-
2.1979	10550	0.0002	-
2.2083	10600	0.0002	-
2.2188	10650	0.0002	-
2.2292	10700	0.0002	-
2.2396	10750	0.0002	-
2.25	10800	0.0002	-
2.2604	10850	0.0002	-
2.2708	10900	0.0002	-
2.2812	10950	0.0002	-
2.2917	11000	0.0002	-
2.3021	11050	0.0002	-
2.3125	11100	0.0002	-
2.3229	11150	0.0002	-
2.3333	11200	0.0002	-
2.3438	11250	0.0002	-
2.3542	11300	0.0002	-
2.3646	11350	0.0002	-
2.375	11400	0.0002	-
2.3854	11450	0.0002	-
2.3958	11500	0.0002	-
2.4062	11550	0.0002	-
2.4167	11600	0.0002	-
2.4271	11650	0.0002	-
2.4375	11700	0.0002	-
2.4479	11750	0.0002	-
2.4583	11800	0.0002	-
2.4688	11850	0.0002	-
2.4792	11900	0.0002	-
2.4896	11950	0.0002	-
2.5	12000	0.0002	-
2.5104	12050	0.0002	-
2.5208	12100	0.0002	-
2.5312	12150	0.0002	-
2.5417	12200	0.0002	-
2.5521	12250	0.0002	-
2.5625	12300	0.0002	-
2.5729	12350	0.0002	-
2.5833	12400	0.0002	-
2.5938	12450	0.0002	-
2.6042	12500	0.0002	-
2.6146	12550	0.0002	-
2.625	12600	0.0002	-
2.6354	12650	0.0002	-
2.6458	12700	0.0002	-
2.6562	12750	0.0002	-
2.6667	12800	0.0002	-
2.6771	12850	0.0002	-
2.6875	12900	0.0002	-
2.6979	12950	0.0002	-
2.7083	13000	0.0002	-
2.7188	13050	0.0002	-
2.7292	13100	0.0002	-
2.7396	13150	0.0002	-
2.75	13200	0.0002	-
2.7604	13250	0.0002	-
2.7708	13300	0.0002	-
2.7812	13350	0.0002	-
2.7917	13400	0.0002	-
2.8021	13450	0.0002	-
2.8125	13500	0.0002	-
2.8229	13550	0.0002	-
2.8333	13600	0.0002	-
2.8438	13650	0.0002	-
2.8542	13700	0.0002	-
2.8646	13750	0.0002	-
2.875	13800	0.0002	-
2.8854	13850	0.0002	-
2.8958	13900	0.0002	-
2.9062	13950	0.0002	-
2.9167	14000	0.0002	-
2.9271	14050	0.0002	-
2.9375	14100	0.0002	-
2.9479	14150	0.0002	-
2.9583	14200	0.0002	-
2.9688	14250	0.0002	-
2.9792	14300	0.0002	-
2.9896	14350	0.0002	-
3.0	14400	0.0002	-

Framework Versions

• Python: 3.11.9
• SetFit: 1.1.3
• Sentence Transformers: 3.3.1
• Transformers: 4.44.2
• PyTorch: 2.13.0+cpu
• Datasets: 5.0.1
• 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}
}