Datasets:
prompt stringlengths 24 129 | chosen stringlengths 56 140 | rejected stringlengths 39 123 |
|---|---|---|
What is the purpose of every generated answer in the curatorkit dataset? | highly detailed response citing specific details from the passage... | low-quality response lacking context... |
What does Quality Gates do in the context of the given passage? | highly detailed response citing specific passage details... | low-quality response with incomplete information... |
What does the `uv pip install vllm --torch-backend=cu128` command do? | a specific command for installing a local vLLM OpenAI server with CUDA support using uv pip install vllm --torch-backend=cu128 | a broad general statement without providing an example or concrete usage context |
What is the purpose of the 'field_mapping' parameter in the JSONL reader? | a thorough explanation of how the field mapping works within the JSONL reader's context | a shallow response that focuses on the surface-level functionality without providing specific details |
What is the primary cause of climate change? | high-quality response citing specific passage details... | low-quality response missing key details... |
What is the primary factor contributing to the success of the company's new product? | high-quality detailed response citing specific product features... | low-quality shallow response omitting critical product details... |
What is the maximum number of epochs in the GRPO model? | high-quality response citing specific passage details... | low-quality response that omits key specifics... |
What is the primary factor affecting the effectiveness of multi-turn generation? | highly detailed response citing specific passage details... | flattened response that omits key details... |
What is the purpose of Evol-Instruct? | excellent detailed explanation citing specific passages related to evol-instruction in text-based systems and AI applications. | short answer focusing on the main idea without providing context. |
What is the main cause of global warming? | High-quality response citing specific details from the passage... | Low-quality response that omits key specifics... |
What does the word 'entails' mean in the context of the source passage? | exhaustive response with detailed understanding of the phrase's usage in logical reasoning contexts | shallow response that misses the crucial information about entailment in logical reasoning |
What does the 'reward_threshold' parameter in the RewardGate score system do? | highly detailed response citing the exact parameters and their significance... | mildly worded response missing critical information... |
What does the 'all-MiniLM-L6-v2' model in DiverseGate do? | highly detailed response citing the specific details provided by the passage | low-quality response that omits key information like the model name and version |
What is the recommended range for the `probe_temperatures` parameter when generating diagnostic probes? | highly detailed response citing specific temperature ranges provided in the passage... | low-quality response using general terms like 'temperature' instead of precise numbers... (exactly one degradation pattern) |
What does 'recovered rows' refer to in the context of reward refiners? | exhaustively detailed response citing specific details from the passage... | misleadingly vague response that focuses solely on surface-level information... |
How does SecretGate handle API keys in its scanning process? | A detailed explanation about how it checks for API keys, including the detection of secrets. | A general statement about the process without mentioning any specific details. |
What does the pseudonymizer do with the same real name across multiple runs? | highly detailed response citing specific details from the passage... | shallow, general explanation without concrete examples... |
What does ToxicityGate's score system do? | a thorough explanation of how ToxicityGate scores rows based on their toxicity level and rejects those who fall short of the pass threshold. | a shallow response that omits the detailed discussion of the score system's logic. |
What is the primary method used to ensure uniqueness in text processing? | highly detailed response citing exact deduction from the passage... | slightly less detailed response missing key specifics... |
What does the text cleaning process do? | highly detailed and comprehensive response citing specific details from the passage... | mildly wordy response lacking specific details... |
What does SchemaGate check when pretraining rows? How do they ensure that input rows have non-empty outputs? | High-quality response citing specific details from the passage... | Low-quality response that omits key details... |
What is the purpose of the 'manifest.json' file in the Hub push process? | exhaustive explanation of how the manifest.json file functions within the Hub push pipeline | focusing solely on the content without mentioning its role in the Hub push process |
What does the 'push_to_hub' function do? | highly detailed response citing specific details from the passage... | low-quality response that omits key details... |
What does the `max_samples` parameter in `max_samples truncates the passed list after` mean? | highly detailed response citing specific details from the passage... | low-quality response that omits key specifics... (exactly one of these patterns) |
How does checkpointing in LLMs work? | a detailed explanation of how checkpoints are generated and loaded during model training. | an overview of the basic concept but doesn't provide specific examples. |
What does PDF ingest do? | a specific, focused question answerable from this passage | a shallow response that omits key details |
How does the stratified sampler ensure balanced distribution across different sources? | highly detailed explanation using concrete examples from the source passage... | mildly wordy response missing crucial details from the passage... |
What is the purpose of the rejected row in the context of downstream training? | exhaustively detailed response citing specific passage details... | mildly worded response missing key nuances... |
What is the purpose of field mapping in data generation? | high-quality response citing specific details from the passage... | low-quality response that omits key specifics... |
What are the required environment variables in colab? How do they differ between GH_WORK_TOKEN, HF_TOKEN, and OPENROUTER_API_KEY? | highly detailed and specific response citing concrete information from the passage... | faintly worded and vague response without providing concrete details... |
What is the preferred format when exporting a JSONL file? | high-quality response citing specific details from the passage... | low-quality response that omits key specifics... |
How many distinct passages are needed for a high-quality, detailed answer in the given passage? | highly detailed answer citing specific passage details... | surface-level answer that misses crucial details... |
YAML Metadata Warning:The task_categories "text2text-generation" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other
|
curatorkit-testrun-Preference
Built using CuratorKIT — provenance-grounded curation and synthesis for LLM post-training.
| Method | preference |
| Backend | litellm |
| Model | openai/Qwen/Qwen2.5-0.5B-Instruct |
| Formats | dpo |
| Artifact | dataset |
| Published | 2026-08-28 09:57 UTC |
Usage
from datasets import load_dataset
ds = load_dataset("ram-lexsi/curatorkit-testrun-Preference", "dpo")
- Downloads last month
- 72