Netta1994/setfit_baai_cybereason_gpt-4o_improved-cot-instructions_chat_few_shot_only_reasoning_

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

This is a SetFit model that can be used for Text Classification. This SetFit model uses BAAI/bge-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: BAAI/bge-base-en-v1.5
• Classification head: a LogisticRegression 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
0	'Reasoning:\nThe response fails to fully address the specifics of the question based on the provided document. The question is about the significance of considering all answers together when determining if the behavior in a MalOp is malicious. The response should have included how analyzing various factors, such as the significance of machines, behavior severity, and the users involved, helps in forming a comprehensive understanding of whether the behavior is malicious and requires further action.\n\nThe answer given is overly vague and does not capture the detailed reasoning found in the document. It lacks essential details concerning the factors that need to be considered, such as machine significance, activity severity, and user importance.\n\nEvaluation: Bad' 'Reasoning:\nThe document provides a detailed process to exclude a MalOp during the remediation phase, and the answer fails to incorporate this information. The answer is also quite dismissive, suggesting the document does not cover the query when it does. Thus, the answer does not address the specific question asked and is not grounded in the context of the provided document.\n\nFinal Result: Bad' 'Reasoning:\nThe answer correctly states that a file should be un-quarantined before submitting it to the respective organization, which aligns with the information provided in the document. The answer is concise, specific, and directly addresses the question without any deviation.\n\nEvaluation: Good'
1	"Reasoning:\nThe answer is directly grounded in the provided document, accurately conveying the information that the computer will generate a dump file containing the entire contents of the sensor's RAM at the time of the failure. The response is concise and specific to the question asked without including unnecessary information.\n\nEvaluation: Good" 'Reasoning:\n1. Context Grounding: The provided answer "To identify cyber security threats" is directly supported by the document, which states that threat detection is a core capability that aims to identify cyber security threats by analyzing data.\n2. Relevance: The answer directly addresses the specific question about the purpose of the platform's threat detection abilities.\n3. Conciseness: The answer is clear and to the point, summarizing the purpose effectively without unnecessary information.\n4. Specificity: The answer provides a specific purpose as outlined in the document.\n\nEvaluation: Good' 'Reasoning:\nThe given answer asserts that the provided information does not cover the specific query, advising to refer to additional sources. However, the document contains clear information regarding four different scenarios and their corresponding severity scores. The answer also does not directly address what is clearly stated in the document and evades the question.\n\nEvaluation: Bad'

Evaluation

Metrics

Label	Accuracy
all	0.5070

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("Netta1994/setfit_baai_cybereason_gpt-4o_improved-cot-instructions_chat_few_shot_only_reasoning_")
# Run inference
preds = model("Reasoning:
The answer is direct, concise, and accurately picks out the relevant information from the document concerning the percentage in the response status column.
Evaluation: Good")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	2	48.4638	115

Label	Training Sample Count
0	34
1	35

Training Hyperparameters

• batch_size: (16, 16)
• num_epochs: (5, 5)
• max_steps: -1
• sampling_strategy: oversampling
• num_iterations: 20
• body_learning_rate: (2e-05, 2e-05)
• head_learning_rate: 2e-05
• 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
0.0058	1	0.2563	-
0.2890	50	0.2638	-
0.5780	100	0.242	-
0.8671	150	0.1521	-
1.1561	200	0.0056	-
1.4451	250	0.0025	-
1.7341	300	0.0022	-
2.0231	350	0.0019	-
2.3121	400	0.0019	-
2.6012	450	0.0016	-
2.8902	500	0.0015	-
3.1792	550	0.0014	-
3.4682	600	0.0014	-
3.7572	650	0.0014	-
4.0462	700	0.0014	-
4.3353	750	0.0014	-
4.6243	800	0.0013	-
4.9133	850	0.0013	-

Framework Versions

• Python: 3.10.14
• SetFit: 1.1.0
• Sentence Transformers: 3.1.0
• Transformers: 4.44.0
• PyTorch: 2.4.1+cu121
• Datasets: 2.19.2
• 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}
}