Netta1994/setfit_baai_rag_ds_gpt-4o_improved-cot-instructions_two_reasoning_only_reasoning_1726

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:\n\nWhy the Answer May Be Good:\n1. Context Grounding: The answer references the points made in the document, such as Coach Brian Shaw's strategy of pushing the ball after makes and misses as well as encouraging players to take the first available shot within the rhythm of the offense.\n2. Relevance: The answer directly addresses why the Nuggets are having an offensive outburst, highlighting the coaching strategy and players' adaptation.\n3. Conciseness: The answer is mostly to the point and focuses on the main question.\n\nWhy the Answer May Be Bad:\n1. Context Grounding: The mention of a new training technique using virtual reality is not supported by any information within the document provided.\n2. Conciseness: The additional detail about the virtual reality training is unnecessary given that it is not referenced in the document and does not contribute to answering the specific question about the offensive outburst.\n \nFinal Result:\nBased on the evaluation criteria, the inclusion of fictitious or unsupported information about the virtual reality training significantly detracts from the answer’s credibility and relevance.\n\nBad" 'Reasoning why the answer may be good:\n1. Context Grounding: The provided answer cites specific information about film and digital photography directly from the provided document, showing a good grounding.\n2. Relevance: The answer addresses the specific question by discussing different aspects such as exposure tolerance, color capture, and overall image resolution between film and digital photography.\n3. Conciseness: The answer is relatively concise and sticks to the main points relevant to the question without unnecessary elaboration.\n\nReasoning why the answer may be bad:\n1. Overly Detailed: The answer could be seen as too detailed in certain segments, which might slightly detract from conciseness.\n2. Possible Confusion: The mention of specific technical details like "5MP digital sensors" could confuse readers who are not familiar with the technical specifications, detracting from clarity.\n3. Omission of Key Comparison Points: The answer does not touch upon some of the more subjective observations made by the author, like the practical advantages in using film for certain types of photography.\n\nFinal Result: Good' 'Reasoning:\n1. Context Grounding: The answer provided does not reference the third book of the Arcana Chronicles by Kresley Cole or even discuss any content relevant to it. Instead, it discusses an MMA event in Calgary, Alberta, Canada.\n2. Relevance: The answer is entirely irrelevant to the question. The question is about the main conflict in the third book of a specific book series, but the answer describes an MMA fight event.\n3. Conciseness: While the answer is concise in its context, it is entirely off-topic and therefore does not satisfy the conciseness criterion in a meaningful way.\n\nThe answer may be deemed bad because it does not address the question about the Arcana Chronicles at all and instead provides unrelated information about an MMA event.\n\nFinal result: Bad'
1	'Reasoning:\n\n1. Context Grounding:\n - Good: The answer is supported by the document. The suggestions mentioned (getting to know the client, signing a contract, and showcasing honesty and diplomacy) are directly referenced in the text provided.\n - Bad: There is no significant bad aspect in terms of context grounding; the answer sticks closely to the source material.\n\n2. Relevance:\n - Good: The answer is highly relevant to the question about best practices to avoid unnecessary revisions and conflicts. It addresses client understanding, contractual agreements, and the handling of extra charges—all crucial for minimizing conflicts.\n - Bad: There is no deviation from the topic. The answer is focused solely on the best practices, as asked in the question.\n\n3. Conciseness:\n - Good: The answer is concise and to the point, effectively summarizing the practices without unnecessary details.\n - Bad: The level of detail might be too succinct for some readers looking for more in-depth discussion, but this is minor given the criteria.\n\nFinal Result:\nGood' "Reasoning for why the answer may be good:\n- The answer references the author’s emphasis on drawing from personal experiences of pain and emotion to create genuine and relatable characters, which is well-supported by the document.\n- It highlights the importance of genuineness and relatability, which aligns directly with the content provided in the document.\n- The answer stays focused on the specific question about creating a connection between the reader and the characters.\n\nReasoning for why the answer may be bad:\n- The answer could be seen as slightly verbose and might include more detail than necessary, rather than being extremely concise.\n- It does not explicitly mention the document's use of pain for romance authors specifically, which might add to the context.\n\nFinal result: Good" "Reasoning:\n\nPros:\n1. Context Grounding: The document explicitly states that Mauro Rubin is the CEO of JoinPad and mentions that he was speaking at the event, which directly supports the answer.\n2. Relevance: The answer directly and correctly responds to the question about the CEO's identity during the event.\n3. Conciseness: The answer is brief and to the point, providing only the necessary information.\n\nCons:\n- There are no significant cons as the answer fulfills all criteria effectively.\n\nFinal Result: Good"

Evaluation

Metrics

Label	Accuracy
all	0.92

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_rag_ds_gpt-4o_improved-cot-instructions_two_reasoning_only_reasoning_1726")
# Run inference
preds = model("**Good**

**Reasoning:**
1. **Context Grounding:** The answer \"China's Ning Zhongyan won the gold medal in the men's 1,500m final at the speed skating World Cup\" is well-supported by the provided document, which explicitly states that Ning Zhongyan won the gold medal in the men's 1,500m final.
2. **Relevance:** The answer directly addresses the specific question asked, identifying the athlete who won the gold medal in the men's 1,500m final.
3. **Conciseness:** The answer is clear and to the point, providing only the necessary information without any additional, unrelated details.")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	52	125.5070	199

Label	Training Sample Count
0	34
1	37

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
0.0056	1	0.2031	-
0.2809	50	0.2589	-
0.5618	100	0.2125	-
0.8427	150	0.0079	-
1.1236	200	0.0022	-
1.4045	250	0.0017	-
1.6854	300	0.0017	-
1.9663	350	0.0014	-
2.2472	400	0.0014	-
2.5281	450	0.0012	-
2.8090	500	0.0012	-
3.0899	550	0.0012	-
3.3708	600	0.0012	-
3.6517	650	0.0011	-
3.9326	700	0.0011	-
4.2135	750	0.0011	-
4.4944	800	0.0011	-
4.7753	850	0.001	-

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}
}