README.md

metadata

base_model: BAAI/bge-base-en-v1.5
library_name: setfit
metrics:
  - accuracy
pipeline_tag: text-classification
tags:
  - setfit
  - sentence-transformers
  - text-classification
  - generated_from_setfit_trainer
widget:
  - text: >-
      Reasoning for Good:

      1. **Context Grounding**: The answer is supported by the document, which
      clearly indicates that Forbes began reporting on Beyoncé's earnings in
      2008.

      2. **Relevance**: The answer specifically addresses the question asked
      about who began reporting Beyoncé's annual earnings starting in 2008.

      3. **Conciseness**: The answer is brief and directly to the point, without
      including extraneous information.


      Reasoning for Bad:

      1. **Context Grounding**: While the statement is accurate, it introduces
      the aspect of a "widespread misconception" about Times Magazine, which is
      not mentioned in the provided document.

      2. **Relevance**: The mention of Times Magazine might be seen as deviating
      slightly from the question, which just asked about the first entity to
      begin reporting Beyoncé's earnings.

      3. **Conciseness**: The answer could have been more concise by focusing
      solely on Forbes without mentioning the misconception about Times
      Magazine.


      Final result: Bad
  - text: >-
      The answer provided is:


      "The average student at Notre Dame travels more than 750 miles to study
      there."


      Reasoning:


      **Good points:**

      1. **Context Grounding**: The answer is supported by information present
      in the document, which states, "the average student traveled more than 750
      miles to Notre Dame".

      2. **Relevance**: The answer directly addresses the specific question
      asking about the number of miles the average student travels to study at
      Notre Dame.

      3. **Conciseness**: The answer is clear and to the point without any
      unnecessary information.


      **Bad points:**

      - There are no bad points in this case as the answer aligns perfectly with
      all the evaluation criteria.


      Final Result: **Good**
  - text: >-
      Reasoning why the answer may be good:

      - The answer correctly identifies Mick LaSalle as the writer for the San
      Francisco Chronicle.

      - The answer states that Mick LaSalle awarded "Spectre" a perfect score,
      which is supported by the document.


      Reasoning why the answer may be bad:

      - The answer is concise and to the point, fulfilling the criteria for
      conciseness and relevance.

      - The document provided confirms that Mick LaSalle gave "Spectre" a
      perfect score of 100.

      - There is no deviation into unrelated topics, maintaining focus on the
      question asked.


      Final result: Good
  - text: >-
      Reasoning: 


      Why the answer may be good:

      - The answer directly addresses the specific question asked, "What New
      York borough contains the highest population of Asian-Americans?" 

      - It is well-supported by the given document, which states, "The New York
      City borough of Queens is home to the state's largest Asian American
      population."

      - The answer is clear and concise without unnecessary information.


      Why the answer may be bad:

      - There are no significant reasons to consider the answer bad based on the
      criteria provided. 


      Final Result: 


      Good
  - text: >-
      The answer may be good:

      - The information provided in the answer is supported by the document. 


      The answer may be bad:

      - The answer does not address the specific question asked which pertains
      to the year that Doctorate degrees were first granted at Notre Dame.

      - It deviates into unrelated information about the opening of a theology
      library, which is irrelevant to the question.


      Final result: Bad
inference: true
model-index:
  - name: SetFit with BAAI/bge-base-en-v1.5
    results:
      - task:
          type: text-classification
          name: Text Classification
        dataset:
          name: Unknown
          type: unknown
          split: test
        metrics:
          - type: accuracy
            value: 0.8360655737704918
            name: Accuracy

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
1	'Reasoning why the answer may be good:\n1. Context Grounding: The answer is directly supported by the document, which explicitly states that "With almost every line of his epic Punica Silius references Virgil."\n2. Relevance: The answer specifically addresses the question asked by identifying the title of Silius Italicus' epic where Virgil is frequently referenced.\n3. Conciseness: The answer is short, clear, and to the point, providing just the necessary information without any extraneous details.\n\nReasoning why the answer may be bad:\n- There is no evidence of deviation or lack of support from the provided document, the relevance is clearly maintained, and the answer concisely addresses the question.\n\nFinal Result: Good' 'Good' 'Reasoning:\n\nWhy the answer may be good:\n1. Context Grounding: The answer mentions "3,000 police," which correlates with the figure provided in the document regarding the number of French police that protected the Olympic torch relay.\n2. Relevance: The answer directly addresses the question, which asks about the number of police protecting the torch in France.\n3. Conciseness: The answer is brief and to the point without adding any unnecessary information.\n\nWhy the answer may be bad:\nThere is no evident issue with context grounding, relevance, or conciseness in the answer provided.\n\nFinal result: Good'
0	"**Reasoning Why the Answer May Be Good:\n- The answer correctly identifies a person associated with vice-presidential and presidential roles at Notre Dame, although it attributes the wrong timeframe for the vice-presidency.\n\nReasoning Why the Answer May Be Bad:\n- The document specifically mentions that John Francis O'Hara became vice-president in 1933, not James Edward O'Hara, indicating the answer is not well-supported by the provided document.\n- The answer provides incorrect and irrelevant information that does not address the specific question asked.\n- The question asked for the vice-president elected in 1933, and the answer incorrectly identifies the year 1934.\n\nFinal Result:**\nBad" "Reasoning:\n1. Context Grounding: The document does provide the necessary information about the gross earnings of Beyoncé's second world tour. Therefore, the answer is well-supported by the document.\n2. Relevance: The answer directly responds to the specific question asked about the gross earnings of Beyoncé during her second world tour in 2009.\n3. Conciseness: The answer is concise and sticks to the point, providing the exact figure and relevant context about the record without additional unnecessary information.\n\nFinal Result: Good" "Reasoning:\n\nWhy the answer may be good:\n- The answer specifies a borough of New York, which is relevant to the question.\n- It provides a specific claim about the population distribution of Asian-Americans within New York City boroughs.\n\nWhy the answer may be bad:\n- The provided document explicitly states that Queens is home to the state's largest Asian-American population, not Manhattan.\n- The answer does not align with the key information from the document, thus failing the test of context grounding.\n\nFinal Result: Bad"

Evaluation

Metrics

Label	Accuracy
all	0.8361

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_squad_gpt-4o_improved-cot-instructions_two_reasoning_only_reasoning_17267")
# Run inference
preds = model("The answer may be good:
- The information provided in the answer is supported by the document. 

The answer may be bad:
- The answer does not address the specific question asked which pertains to the year that Doctorate degrees were first granted at Notre Dame.
- It deviates into unrelated information about the opening of a theology library, which is irrelevant to the question.

Final result: Bad")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	1	91.8596	275

Label	Training Sample Count
0	27
1	30

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
0.0070	1	0.1646	-
0.3497	50	0.2544	-
0.6993	100	0.1157	-
1.0490	150	0.0294	-
1.3986	200	0.0037	-
1.7483	250	0.0025	-
2.0979	300	0.0023	-
2.4476	350	0.002	-
2.7972	400	0.0018	-
3.1469	450	0.0017	-
3.4965	500	0.0016	-
3.8462	550	0.0017	-
4.1958	600	0.0016	-
4.5455	650	0.0015	-
4.8951	700	0.0016	-

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}
}