diwank/hn-upvote-classifier

SetFit with Alibaba-NLP/gte-base-en-v1.5

This is a SetFit model trained on the diwank/hn-upvote-data dataset that can be used for Text Classification. This SetFit model uses Alibaba-NLP/gte-base-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: Alibaba-NLP/gte-base-en-v1.5
• Classification head: a LogisticRegression instance
• Maximum Sequence Length: 8192 tokens
• Number of Classes: 2 classes
• Training Dataset: diwank/hn-upvote-data

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
0	'The telltale words that could identify generative AI text' 'The telltale words that could identify generative AI text' 'The telltale words that could identify generative AI text'
1	'Dangerous Feelings\nSource: www.collaborativefund.com' 'The Modos Paper Monitor\nSource: www.modos.tech' 'What did Mary know? A thought experiment about consciousness (2013)\nSource: philosophynow.org'

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("diwank/hn-upvote-classifier")
# Run inference
preds = model("My Python code is a neural network")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	3	8.6577	18

Label	Training Sample Count
0	4577
1	252

Training Hyperparameters

• batch_size: (320, 32)
• num_epochs: (1, 16)
• max_steps: -1
• sampling_strategy: undersampling
• body_learning_rate: (4e-05, 2e-05)
• head_learning_rate: 0.01
• loss: CosineSimilarityLoss
• distance_metric: cosine_distance
• margin: 0.25
• end_to_end: True
• use_amp: True
• warmup_proportion: 0.05
• l2_weight: 0.2
• seed: 42
• eval_max_steps: -1
• load_best_model_at_end: True

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0001	1	0.208	-
0.0069	50	0.0121	-
0.0139	100	0.002	-
0.0208	150	0.0032	-
0.0277	200	0.001	-
0.0347	250	0.0006	-
0.0416	300	0.0005	-
0.0486	350	0.0004	-
0.0555	400	0.0003	-
0.0624	450	0.0002	-
0.0694	500	0.0002	-
0.0763	550	0.0002	-
0.0832	600	0.0002	-
0.0902	650	0.0001	-
0.0971	700	0.0001	-
0.1040	750	0.0001	-
0.1110	800	0.0001	-
0.1179	850	0.0001	-
0.1248	900	0.0001	-
0.1318	950	0.0001	-
0.1387	1000	0.0001	-
0.1457	1050	0.0001	-
0.1526	1100	0.0001	-
0.1595	1150	0.0001	-
0.1665	1200	0.0001	-
0.1734	1250	0.0001	-
0.1803	1300	0.0001	-
0.1873	1350	0.0001	-
0.1942	1400	0.0001	-
0.2011	1450	0.0001	-
0.2081	1500	0.0001	-
0.2150	1550	0.0001	-
0.2219	1600	0.0	-
0.2289	1650	0.0	-
0.2358	1700	0.0	-
0.2428	1750	0.0	-
0.2497	1800	0.0001	-
0.2566	1850	0.0	-
0.2636	1900	0.0	-
0.2705	1950	0.0	-
0.2774	2000	0.0	-
0.2844	2050	0.0	-
0.2913	2100	0.0	-
0.2982	2150	0.0	-
0.3052	2200	0.0	-
0.3121	2250	0.0	-
0.3190	2300	0.0	-
0.3260	2350	0.0	-
0.3329	2400	0.0	-
0.3399	2450	0.0	-
0.3468	2500	0.0	-
0.3537	2550	0.0	-
0.3607	2600	0.0	-
0.3676	2650	0.0	-
0.3745	2700	0.0	-
0.3815	2750	0.0	-
0.3884	2800	0.0	-
0.3953	2850	0.0	-
0.4023	2900	0.0	-
0.4092	2950	0.0	-
0.4161	3000	0.0	-
0.4231	3050	0.0	-
0.4300	3100	0.0	-
0.4370	3150	0.0	-
0.4439	3200	0.0	-
0.4508	3250	0.0	-
0.4578	3300	0.0	-
0.4647	3350	0.0	-
0.4716	3400	0.0	-
0.4786	3450	0.0	-
0.4855	3500	0.0	-
0.4924	3550	0.0	-
0.4994	3600	0.0	-
0.5063	3650	0.0	-
0.5132	3700	0.0	-
0.5202	3750	0.0	-
0.5271	3800	0.0	-
0.5341	3850	0.0	-
0.5410	3900	0.0	-
0.5479	3950	0.0	-
0.5549	4000	0.0	-
0.5618	4050	0.0	-
0.5687	4100	0.0	-
0.5757	4150	0.0	-
0.5826	4200	0.0	-
0.5895	4250	0.0	-
0.5965	4300	0.0	-
0.6034	4350	0.0	-
0.6103	4400	0.0	-
0.6173	4450	0.0	-
0.6242	4500	0.0	-
0.6312	4550	0.0	-
0.6381	4600	0.0	-
0.6450	4650	0.0	-
0.6520	4700	0.0	-
0.6589	4750	0.0	-
0.6658	4800	0.0	-
0.6728	4850	0.0	-
0.6797	4900	0.0	-
0.6866	4950	0.0	-
0.6936	5000	0.0	-
0.7005	5050	0.0	-
0.7074	5100	0.0	-
0.7144	5150	0.0	-
0.7213	5200	0.0	-
0.7283	5250	0.0	-
0.7352	5300	0.0	-
0.7421	5350	0.0	-
0.7491	5400	0.0	-
0.7560	5450	0.0	-
0.7629	5500	0.0	-
0.7699	5550	0.0	-
0.7768	5600	0.0	-
0.7837	5650	0.0	-
0.7907	5700	0.0	-
0.7976	5750	0.0	-
0.8045	5800	0.0	-
0.8115	5850	0.0	-
0.8184	5900	0.0	-
0.8254	5950	0.0	-
0.8323	6000	0.0	-
0.8392	6050	0.0	-
0.8462	6100	0.0	-
0.8531	6150	0.0	-
0.8600	6200	0.0	-
0.8670	6250	0.0	-
0.8739	6300	0.0	-
0.8808	6350	0.0	-
0.8878	6400	0.0	-
0.8947	6450	0.0	-
0.9017	6500	0.0	-
0.9086	6550	0.0	-
0.9155	6600	0.0	-
0.9225	6650	0.0	-
0.9294	6700	0.0	-
0.9363	6750	0.0	-
0.9433	6800	0.0	-
0.9502	6850	0.0	-
0.9571	6900	0.0	-
0.9641	6950	0.0	-
0.9710	7000	0.0	-
0.9779	7050	0.0	-
0.9849	7100	0.0	-
0.9918	7150	0.0	-
0.9988	7200	0.0	-

Framework Versions

• Python: 3.10.14
• SetFit: 1.0.3
• Sentence Transformers: 3.0.1
• Transformers: 4.41.2
• PyTorch: 2.3.1+cu121
• Datasets: 2.20.0
• 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}
}