thegenerativegeneration/stay_or_go_conversation_classifier_s

SetFit with intfloat/multilingual-e5-small

This is a SetFit model that can be used for Text Classification. This SetFit model uses intfloat/multilingual-e5-small as the Sentence Transformer embedding model. A SetFitHead 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: intfloat/multilingual-e5-small
• Classification head: a SetFitHead instance
• Maximum Sequence Length: 512 tokens
• Number of Classes: 2 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
1	'query: ਚੰਗਾ ਜੀ, ਫਿਰ ਮਿਲਦੇ ਹਾਂ.' 'query: Agur, gero arte.' "query: Me'n vaig ara."
0	'query: Dobro, hvala. Kaj pa ti?' 'query: हाँ अगली बार जब तुम जाओ मुझे भी ले चलो मुझे भी प्रकृति में और गतिविधियाँ करनी हैं' 'query: Mirë, faleminderit. Po ju?'

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("query: Ναι, ας πάμε!")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	2	7.4364	21

Label	Training Sample Count
0	292
1	290

Training Hyperparameters

• batch_size: (16, 2)
• num_epochs: (1, 16)
• max_steps: -1
• sampling_strategy: undersampling
• body_learning_rate: (1e-05, 1e-05)
• head_learning_rate: 0.001
• loss: CosineSimilarityLoss
• distance_metric: cosine_distance
• margin: 0.1
• end_to_end: False
• use_amp: False
• warmup_proportion: 0.1
• seed: 42
• run_name: intfloat/multilingual-e5-small
• eval_max_steps: -1
• load_best_model_at_end: True

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0001	1	0.3645	-
0.0047	50	0.3527	-
0.0094	100	0.3424	0.3165
0.0142	150	0.3108	-
0.0189	200	0.2684	0.2215
0.0236	250	0.2197	-
0.0283	300	0.1707	0.1792
0.0331	350	0.1501	-
0.0378	400	0.0865	0.1607
0.0425	450	0.0534	-
0.0472	500	0.0307	0.1519
0.0520	550	0.0342	-
0.0567	600	0.0078	0.1478
0.0614	650	0.0144	-
0.0661	700	0.0658	0.1399
0.0709	750	0.0021	-
0.0756	800	0.0009	0.1512
0.0803	850	0.0005	-
0.0850	900	0.0018	0.1516
0.0897	950	0.0011	-
0.0945	1000	0.0012	0.1541
0.0992	1050	0.0003	-
0.1039	1100	0.0003	0.1415
0.1086	1150	0.0003	-
0.1134	1200	0.0002	0.1442
0.1181	1250	0.0006	-
0.1228	1300	0.0002	0.1298
0.1275	1350	0.0002	-
0.1323	1400	0.0001	0.1356
0.1370	1450	0.0002	-
0.1417	1500	0.0003	0.1493
0.1464	1550	0.0003	-
0.1512	1600	0.0002	0.15
0.1559	1650	0.0002	-
0.1606	1700	0.0003	0.1469
0.1653	1750	0.0001	-
0.1701	1800	0.0001	0.1554
0.1748	1850	0.0002	-
0.1795	1900	0.0001	0.168
0.1842	1950	0.0001	-
0.1889	2000	0.0004	0.1568
0.1937	2050	0.0001	-
0.1984	2100	0.0001	0.1513
0.2031	2150	0.0001	-
0.2078	2200	0.0003	0.1503
0.2126	2250	0.0002	-
0.2173	2300	0.0604	0.155
0.2220	2350	0.0001	-
0.2267	2400	0.0002	0.1739
0.2315	2450	0.0006	-
0.2362	2500	0.0002	0.1558
0.2409	2550	0.0002	-
0.2456	2600	0.0001	0.1393
0.2504	2650	0.0004	-
0.2551	2700	0.0003	0.1642
0.2598	2750	0.0002	-
0.2645	2800	0.0002	0.1776
0.2692	2850	0.0	-
0.2740	2900	0.0002	0.1794
0.2787	2950	0.0001	-
0.2834	3000	0.0001	0.183
0.2881	3050	0.0001	-
0.2929	3100	0.0001	0.1805
0.2976	3150	0.0001	-
0.3023	3200	0.0001	0.1757
0.3070	3250	0.0001	-
0.3118	3300	0.0001	0.1302
0.3165	3350	0.0001	-
0.3212	3400	0.0001	0.1348
0.3259	3450	0.0001	-
0.3307	3500	0.0005	0.1623
0.3354	3550	0.0	-
0.3401	3600	0.0	0.1286
0.3448	3650	0.0	-
0.3496	3700	0.0001	0.1736
0.3543	3750	0.0	-
0.3590	3800	0.0	0.127
0.3637	3850	0.0	-
0.3684	3900	0.0001	0.1231
0.3732	3950	0.0	-
0.3779	4000	0.0001	0.1261
0.3826	4050	0.0001	-
0.3873	4100	0.0	0.1216
0.3921	4150	0.0	-
0.3968	4200	0.0	0.1404
0.4015	4250	0.0	-
0.4062	4300	0.0	0.1466
0.4110	4350	0.0	-
0.4157	4400	0.0	0.1482
0.4204	4450	0.0	-
0.4251	4500	0.0	0.1547
0.4299	4550	0.0	-
0.4346	4600	0.0	0.1566
0.4393	4650	0.0	-
0.4440	4700	0.0	0.1684
0.4487	4750	0.0	-
0.4535	4800	0.0	0.1746
0.4582	4850	0.0	-
0.4629	4900	0.0	0.167
0.4676	4950	0.0	-
0.4724	5000	0.0001	0.1683
0.4771	5050	0.0	-
0.4818	5100	0.0	0.1693
0.4865	5150	0.0	-
0.4913	5200	0.0	0.1694
0.4960	5250	0.0	-
0.5007	5300	0.0	0.162
0.5054	5350	0.0	-
0.5102	5400	0.0	0.1388
0.5149	5450	0.0	-
0.5196	5500	0.0	0.1353
0.5243	5550	0.0	-
0.5291	5600	0.0	0.1401
0.5338	5650	0.0	-
0.5385	5700	0.0	0.1466
0.5432	5750	0.0	-
0.5479	5800	0.0	0.1529
0.5527	5850	0.0	-
0.5574	5900	0.0	0.1488
0.5621	5950	0.0	-
0.5668	6000	0.0	0.147
0.5716	6050	0.0	-
0.5763	6100	0.0	0.1493
0.5810	6150	0.0	-
0.5857	6200	0.0	0.1525
0.5905	6250	0.0	-
0.5952	6300	0.0	0.1505
0.5999	6350	0.0	-
0.6046	6400	0.0	0.1554
0.6094	6450	0.0	-
0.6141	6500	0.0	0.1546
0.6188	6550	0.0	-
0.6235	6600	0.0	0.1598
0.6282	6650	0.0	-
0.6330	6700	0.0	0.179
0.6377	6750	0.0	-
0.6424	6800	0.0	0.1719
0.6471	6850	0.0001	-
0.6519	6900	0.0	0.1812
0.6566	6950	0.0	-
0.6613	7000	0.0	0.1648
0.6660	7050	0.0	-
0.6708	7100	0.0	0.1717
0.6755	7150	0.0	-
0.6802	7200	0.0	0.1793
0.6849	7250	0.0	-
0.6897	7300	0.0	0.1766
0.6944	7350	0.0	-
0.6991	7400	0.0	0.177
0.7038	7450	0.0	-
0.7085	7500	0.0	0.1749
0.7133	7550	0.0	-
0.7180	7600	0.0	0.1814
0.7227	7650	0.0	-
0.7274	7700	0.0	0.1742
0.7322	7750	0.0	-
0.7369	7800	0.0	0.179
0.7416	7850	0.0	-
0.7463	7900	0.0	0.1767
0.7511	7950	0.0	-
0.7558	8000	0.0	0.1809
0.7605	8050	0.0	-
0.7652	8100	0.0	0.1767
0.7700	8150	0.0	-
0.7747	8200	0.0	0.1698
0.7794	8250	0.0	-
0.7841	8300	0.0	0.1772
0.7889	8350	0.0	-
0.7936	8400	0.0	0.1722
0.7983	8450	0.0	-
0.8030	8500	0.0	0.1671
0.8077	8550	0.0	-
0.8125	8600	0.0	0.181
0.8172	8650	0.0	-
0.8219	8700	0.0	0.1788
0.8266	8750	0.0	-
0.8314	8800	0.0	0.1784
0.8361	8850	0.0	-
0.8408	8900	0.0	0.1806
0.8455	8950	0.0	-
0.8503	9000	0.0	0.1783
0.8550	9050	0.0	-
0.8597	9100	0.0	0.1783
0.8644	9150	0.0	-
0.8692	9200	0.0	0.1785
0.8739	9250	0.0	-
0.8786	9300	0.0	0.1772
0.8833	9350	0.0	-
0.8880	9400	0.0	0.1816
0.8928	9450	0.0	-
0.8975	9500	0.0	0.1794
0.9022	9550	0.0	-
0.9069	9600	0.0	0.168
0.9117	9650	0.0	-
0.9164	9700	0.0	0.1771
0.9211	9750	0.0	-
0.9258	9800	0.0	0.1675
0.9306	9850	0.0	-
0.9353	9900	0.0	0.1746
0.9400	9950	0.0	-
0.9447	10000	0.0	0.1769
0.9495	10050	0.0	-
0.9542	10100	0.0	0.177
0.9589	10150	0.0	-
0.9636	10200	0.0	0.1771
0.9684	10250	0.0	-
0.9731	10300	0.0	0.1794
0.9778	10350	0.0	-
0.9825	10400	0.0	0.177
0.9872	10450	0.0	-
0.9920	10500	0.0	0.1794
0.9967	10550	0.0	-

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.11
• SetFit: 1.0.3
• Sentence Transformers: 2.7.0
• Transformers: 4.39.0
• PyTorch: 2.3.1
• Datasets: 2.20.0
• Tokenizers: 0.15.2

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}
}