Netta1994/setfit_baai_cybereason_gpt-4o_improved-cot-instructions_chat_few_shot_generated_only_

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 answer "It provides a comprehensive understanding of the situation" is overly general and lacks specifics directly drawn from the provided document. According to the document, considering all the answers together allows one to determine if the behavior is malicious and requires further remediation. Specifically, it discusses evaluating the significance of the machines, behaviors, and users involved to prioritize actions and determine the severity of the threat. The answer fails to incorporate these detailed aspects, making it too broad and inadequately grounded in the document.\n\nFinal evaluation: Bad' "Reasoning:\nThe provided answer states that the information doesn't cover the specific query without attempting to derive an answer from the provided document. However, the document clearly outlines the process to exclude a MalOp during the remediation phase. It includes navigating to the appropriate sections of the platform and following specific steps to exclude the MalOp. The answer fails to capture this information and does not directly address the question with details available in the document.\n\nEvaluation: Bad" 'Reasoning:\nThe answer "If a file is quarantined, it should be un-quarantined before submitting it to ." accurately reflects the information provided in the document. The document explicitly mentions that, as a pre-requisite, if a file is quarantined, it should be un-quarantined before submitting it to . The answer is concise, directly relevant to the question, and well-supported by the document.\n\nFinal evaluation: \nGood'
1	'Reasoning:\nThe answer "The computer will generate a dump file containing the entire contents of the sensor's RAM at the time of the failure" is accurate, well-supported by the document, and directly related to the question. The document clearly states in the "Result" section that "The computer will now generate a dump file in the event of a system failure, containing the entire contents of the sensor's RAM at the time of the failure." The answer is concise, relevant, and specific to the question asked, without providing unnecessary information or deviating from the context.\n\nFinal evaluation: Good' 'Reasoning:\nThe provided document states that the purpose of the platform's threat detection abilities is to "identify cyber security threats" by using the Engine to analyze data collected from various sources, leveraging artificial intelligence, machine learning, and behavioral analysis. The answer given, "To identify cyber security threats," is well-supported by the document and directly addresses the question without deviating into unrelated topics. It is concise, specific, and directly relevant to the question asked.\n\nEvaluation: Good' "Reasoning:\nThe answer given states that the information provided doesn't cover the specific query and suggests referring to additional sources. However, the document clearly lists four scenarios with their associated severity scores, and there is no mention of a fifth scenario. The document does indeed only cover four scenarios, thereby making the user's response valid. The user answered in a way that acknowledges the limitation of provided information, which is accurate.\n\nEvaluation: Good"

Evaluation

Metrics

Label	Accuracy
all	0.5211

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_generated_only_")
# Run inference
preds = model("Reasoning:
The provided document clearly outlines the purpose of the <ORGANIZATION> XDR On-Site Collector Agent: it is installed to collect logs from platforms and securely forward them to <ORGANIZATION> XDR. The answer given aligns accurately with the document's description, addressing the specific question without deviating into unrelated topics. The response is also concise and to the point.

Evaluation: Good
")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	40	74.3043	119

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	0.2291	-
0.2890	50	0.2565	-
0.5780	100	0.2405	-
0.8671	150	0.1052	-
1.1561	200	0.0039	-
1.4451	250	0.0022	-
1.7341	300	0.002	-
2.0231	350	0.0017	-
2.3121	400	0.0017	-
2.6012	450	0.0015	-
2.8902	500	0.0013	-
3.1792	550	0.0013	-
3.4682	600	0.0013	-
3.7572	650	0.0013	-
4.0462	700	0.0013	-
4.3353	750	0.0012	-
4.6243	800	0.0012	-
4.9133	850	0.0012	-

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}
}