Netta1994/setfit_baai_cybereason_gpt-4o_cot-instructions_only_reasoning_1726752054.560885

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:\n1. Context Grounding: The response "It provides a comprehensive understanding of the situation" is too vague and not well-supported by the document provided. The document specifically discusses the importance of considering all answers to comprehensively determine if behavior is malicious.\n2. Relevance: While the response is somewhat related, it does not specifically address why considering all the answers together is significant. The document talks about the threat qualification steps and emphasizes the importance of examining multiple indicators to correctly assess the situation.\n3. Conciseness: The answer is concise but lacks detail.\n4. Specificity: The response is too general and lacks the specific details necessary to fully answer the question as supported by the document.\n5. Key/Value/Event Name: Not applicable in this evaluation.\n\nFinal Result: Bad' "The given answer does not address the specific question asked. The document contains detailed steps on how to exclude a MalOp during the remediation phase, which directly relates to the question. However, the answer provided claims that the information doesn't cover this specific query and suggests referring to additional sources, which is incorrect. \n\nReasoning:\n1. Context Grounding: The provided document clearly offers steps on how to exclude a MalOp, hence the answer is not grounded in the context of the document.\n2. Relevance: The question asked is directly related to the steps to exclude a MalOp, yet the response does not address these steps.\n3. Conciseness: The response suggests looking elsewhere, which is not concise or needed since the information is already present in the document.\n4. Specificity: The document contains specific steps and details regarding the MalOp exclusion process, which the answer fails to capture.\n\nFinal Result: Bad" 'Reasoning:\n\n1. Context Grounding: The answer directly reflects a step in the document which states that if a file is quarantined, it should be un-quarantined before submission.\n2. Relevance: The answer specifically addresses the asked question regarding the procedure to follow if a file is quarantined.\n3. Conciseness: The response is very concise and directly addresses the action to take.\n4. Specificity: The answer pinpoints the exact action required for quarantined files as mentioned in the document.\n\nFinal result: Good'
1	"Reasoning:\n1. Context Grounding: The provided document specifies that after configuring a sensor, the computer will generate a memory dump file containing the RAM contents at the time of failure, which supports the given answer.\n2. Relevance: The answer directly responds to the question by stating what the computer will generate in the event of a system failure.\n3. Conciseness: The answer is brief and directly answers the question without any extraneous information.\n4. Specificity: The answer is not overly general; it correctly identifies that the dump file will contain the contents of the sensor's RAM at the time of the failure, aligning with the document.\n5. Key/Value/Event name: The answer correctly identifies the relevant outcome, which is the generation of a memory dump file containing the sensor's RAM contents.\n\nFinal Result: Good" "Reasoning:\n\n1. Context Grounding: The answer is concise and based on the provided document. Both the detected purpose (identify cyber security threats) and the mechanism (using the engine with AI, ML, and behavioral analysis) are aligned with the document's contents.\n2. Relevance: The answer directly addresses the specific question asked about the purpose of the platforms threat detection abilities.\n3. Conciseness: The answer is clear and to the point without unnecessary information. \n4. Specificity: The answer accurately identifies the relevant purpose mentioned in the document. \n5. Key/Value/Event: The question does not prompt for key, value, or event name, so this criterion is not applicable here.\n\nFinal Result: Good" "The information provided directly addresses the question by assessing the presence of relevant text in the given document. The response accurately identifies that the document does not mention or cover a fifth scenario.\n\n1. Context Grounding: The answer is well-supported by the document and maintains a clear link to the given text, confirming the absence of a fifth scenario.\n2. Relevance: The answer clearly addresses the specific question asked, ensuring no deviation.\n3. Conciseness: The answer is clear and to the point, avoiding any unnecessary information.\n4. Specifics: The answer is specific in confirming the lack of content related to a fifth scenario, ensuring correctness.\n5. Key/Value/Event Name Identification: Given the document only contains four scenarios, the identification of a fifth scenario's severity score is inherently impossible and is aptly noted.\n\nFinal Verdict: Good"

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_cot-instructions_only_reasoning_1726752054.560885")
# Run inference
preds = model("The percentage in the response status column indicates the total amount of successful completion of response actions.

Reasoning:
1. **Context Grounding**: The answer is well-supported by the document which states, \"percentage indicates the total amount of successful completion of response actions.\"
2. **Relevance**: The answer directly addresses the specific question asked about what the percentage in the response status column indicates.
3. **Conciseness**: The answer is succinct and to the point without unnecessary information.
4. **Specificity**: The answer is specific to what is being asked, detailing exactly what the percentage represents.
5. **Accuracy**: The answer provides the correct key/value as per the document.

Final result: Good")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	70	112.1739	168

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	0.2538	-
0.2890	50	0.2672	-
0.5780	100	0.2355	-
0.8671	150	0.0836	-
1.1561	200	0.0038	-
1.4451	250	0.0024	-
1.7341	300	0.0021	-
2.0231	350	0.0018	-
2.3121	400	0.0017	-
2.6012	450	0.0015	-
2.8902	500	0.0014	-
3.1792	550	0.0014	-
3.4682	600	0.0013	-
3.7572	650	0.0013	-
4.0462	700	0.0013	-
4.3353	750	0.0013	-
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}
}