instruction stringlengths 6 102 | input stringlengths 175 316 | output stringlengths 33 419 |
|---|---|---|
What does curatorKIT do? | CuratorKIT builds trainer-ready datasets for LLM post-training. Every generated answer is checked against the exact source chunk it came from. Rejected samples are written to rejected.jsonl with a structured reason, never dropped silently. A run also writes manifest.json, dataset_card.md, and SHA-256 checksums.txt. | CuratorKIT builds trainer-ready datasets for LLM post-training. |
What are the main components of Quality Gates? | Quality gates run after generation. HallucinationGate scores grounding against the source passage. RewardGate scores helpfulness, honesty, and instruction following. DiversityGate drops near-duplicates using sentence embeddings. Adaptive recovery diagnoses each rejection and repairs the recoverable ones. | Quality Gates consist of three main components: HallucinationGate, RewardGate, and DiversityGate. The HallucinationGate evaluates the relevance and accuracy of the generated content to the source passage. The RewardGate assesses whether the generated content is valuable or beneficial to the user. The DiversityGate iden... |
": "Wh | Hygiene stages run before any LLM call. SecretsGate rejects API keys and private tokens. PIIPseudonymizer replaces names and emails with consistent fake values. ToxicityGate uses a local Detoxify classifier, with an optional LLM judge for borderline rows. | t are the functions of each component mentioned in the given text?",
"answer": "The components mentioned in the text include:
- SecretGate: Rejects API keys and private tokens.
- PIIPseudonymizer: Replaces names and emails with consistent fake values.
- ToxicityGate: Uses a local Detoxify classifier, optionall... |
What is the purpose of a curator run? | A Curator run is config then execute. CuratorConfig names the dataset, field mapping, generation task, and export formats. curator.run() walks readers, hygiene, generation, gates, and exporters in that order. Changing BACKEND does not require rewriting the curator cell. | A curator run is used to configure various aspects of the data processing pipeline, including dataset naming, field mapping, generation tasks, and export formats. |
How does the curator run process work? | A Curator run is config then execute. CuratorConfig names the dataset, field mapping, generation task, and export formats. curator.run() walks readers, hygiene, generation, gates, and exporters in that order. Changing BACKEND does not require rewriting the curator cell. | The curator run consists of four main steps: |
What components does vLLM use to install? | resolve_backend picks OpenRouter, a local vLLM OpenAI server, or Ollama. vLLM install uses uv pip install vllm --torch-backend=cu128. The waiter polls /v1/models and does not kill a serve that is still loading weights. | vLLM installs using `uv pip install vllm --torch-backend=cu128` |
How does vLLM handle serving models when they are still loading weights? | resolve_backend picks OpenRouter, a local vLLM OpenAI server, or Ollama. vLLM install uses uv pip install vllm --torch-backend=cu128. The waiter polls /v1/models and does not kill a serve that is still loading weights. | vLLM poll `/v1/models` and does not kill a serve that is still loading weights. |
What are the features supported by the JSONL readers? | JSONL readers accept Alpaca, ShareGPT, DPO, GRPO, and pretrain layouts. field_mapping remaps source columns onto schema fields. format='pretrain' keeps a text column as language-modeling output instead of renaming it to instruction. | The JSONL readers support Alpaca, ShareGPT, DPO, GRPO, and pretrain layouts. They also allow for field_mapping that renames source columns into schema fields. |
How does field_mapping work in the JSONL readers? | JSONL readers accept Alpaca, ShareGPT, DPO, GRPO, and pretrain layouts. field_mapping remaps source columns onto schema fields. format='pretrain' keeps a text column as language-modeling output instead of renaming it to instruction. | Field_mapping remaps source columns onto schema fields, which means any data in the source file is transformed into corresponding schema fields within the JSONL reader's structure. |
What is the main idea of the passage? | QA generation asks a grounded question per chunk and writes an answer that must be supported by that chunk. num_questions controls how many pairs each passage yields. difficulty is easy, medium, or hard and only changes the prompt, not the schema. | The main idea of the passage is to provide a comprehensive overview of the requirements for generating QA (Question-Answer) pairs. |
What is the purpose of preference generation? | Preference generation writes a chosen completion and a rejected completion for the same prompt. single_call asks the model for both in one response. two_pass samples independently then labels the better one. Export is DPO JSONL. | Preference generation is used to choose between multiple completions based on user preferences or criteria. |
What does GRPO generation do? | GRPO generation samples several completions per prompt and optional judge scores. Temperature spread varies rollouts around llm_temperature so the group is not identical. Export includes grpo and ppo JSONL. | GRPO generation samples several completions per prompt and optional judge scores. It also spreads temperature across rollout rolls to ensure that the group is not identical. |
How does temperature spread impact the results of GRPO generation? | GRPO generation samples several completions per prompt and optional judge scores. Temperature spread varies rollouts around llm_temperature so the group is not identical. Export includes grpo and ppo JSONL. | Temperature spread affects the diversity of the generated completions. As temperature increases, the likelihood of generating different completions increases. This can lead to variations in the quality and distribution of the generated content. |
What does Multi-turn Generation build? | Multi-turn generation builds a short dialogue from a source chunk. num_turns is the number of user-assistant pairs. ShareGPT is the natural export; Alpaca flattens the last turn. | Multi-turn Generation builds a short dialogue from a source chunk. It involves generating multiple short conversations between the user and the assistant. |
How many turns can Multi-turn Generation have? | Multi-turn generation builds a short dialogue from a source chunk. num_turns is the number of user-assistant pairs. ShareGPT is the natural export; Alpaca flattens the last turn. | Multi-turn Generation can have any number of turns based on the requirements specified in the source text. |
What does 'generate_answers' in Evol-Instruct mean? | Evol-Instruct rewrites a seed instruction into a harder variant. num_evolutions is how many rewrite steps run. generate_answers asks the model to solve the evolved prompt so the row is SFT-ready. | Generate_answers refers to the process where the model generates and answers the evolved prompt based on the outcomes of the rewritten seeds. |
What is the purpose of chain-of-thought generation? | Chain-of-thought generation either wraps an existing answer with reasoning or produces a new reasoned answer from the chunk. cot_mode is generate or wrap. The reasoning string is stored on the sample before export. | Chain-of-thought generation is an important technique used in natural language processing and machine learning to improve the quality of generated responses. It involves generating multiple versions of a response using different algorithms and techniques to ensure that each version has a unique perspective. |
How does the reasoning string store information on the sample before exporting? | Chain-of-thought generation either wraps an existing answer with reasoning or produces a new reasoned answer from the chunk. cot_mode is generate or wrap. The reasoning string is stored on the sample before export. | The reasoning string stores information about how the sample was generated. This includes details such as the algorithm used, parameters set, and any custom logic implemented. |
What does the 'adversarial QA' module inject into a set of answers? | Adversarial QA injects a failure into a subset of answers after generation. injection_rate is the fraction of rows that get a known bad span. injection_types include contradicts_source and parametric_drift. The metadata flag injected_failure marks those rows. | The 'adversary QA' module injects a failure into a subset of answers after generation. It uses two types of injection: contradictions for known bad spans and parameters for drift. |
What does HallucinationGate score? | HallucinationGate scores whether the answer is entailed by the source passage. hallucination_threshold drops rows below the cutoff. A separate judge model avoids self-scoring. skip_if_no_context leaves rows without a passage alone. | HallucinationGate scores whether the answer is entailed by the source passage. |
How does HallucinationGate determine if a row is eligible for inclusion? | HallucinationGate scores whether the answer is entailed by the source passage. hallucination_threshold drops rows below the cutoff. A separate judge model avoids self-scoring. skip_if_no_context leaves rows without a passage alone. | HallucinationGate determines if a row is eligible for inclusion based on its relevance to the source passage and adherence to the scoring criteria. |
What does the 'reward_threshold' parameter in RewardGate score? | RewardGate scores helpfulness, honesty, and instruction following. reward_threshold drops low-scoring rows. Scores can be stored on the label for later filtering. The judge uses the same API base as generation unless overridden. | The 'reward_threshold' parameter in RewardGate evaluates how much a row is considered helpful or honest based on predefined scoring criteria. |
What does the 'embeds' function in DiversityGate do? | DiversityGate embeds each accepted row and drops near-duplicates above similarity_threshold. The default embedding model is all-MiniLM-L6-v2. Coverage can be measured on an optional metadata field. | The 'embeds' function in DiversityGate takes each accepted row and drops near-duplicates based on the specified similarity threshold. |
What is the default embedding model used by DiversityGate? | DiversityGate embeds each accepted row and drops near-duplicates above similarity_threshold. The default embedding model is all-MiniLM-L6-v2. Coverage can be measured on an optional metadata field. | The default embedding model used by DiversityGate is called all-MiniLM-L6-v2. |
What does DiagnosticProbe re-generate when it encounters a rejected row? | DiagnosticProbe re-generates a rejected row at several temperatures. probe_temperatures is the list of values. probe_score_split decides which retries count as recovered. Recovered rows return to passed with provenance. | When DiagnosticProbe encounters a rejected row, it generates a new row using different temperature settings and counts retries based on whether they are counted or not. |
What is the purpose of enabling 'reward_refiner'? | RewardRefiner rewrites a low-reward answer and re-runs RewardGate. enable_reward_refiner attaches after the reward step. Recovered rows are merged into passed only after the pipeline returns so exporters see them. | Enabling 'reward_refiner' allows for re-writing of a previously low-reward answer and rerunning it through a reward gate. This process is then used to merge recovered rows back into the passed data before exporting results. |
How does the 'recovered rows being merged into passed only after the pipeline returns' statement work? | RewardRefiner rewrites a low-reward answer and re-runs RewardGate. enable_reward_refiner attaches after the reward step. Recovered rows are merged into passed only after the pipeline returns so exporters see them. | This statement indicates that the recovered rows are merged into the passed data only after the pipeline has returned its results. This ensures that the exporter sees the correct output based on the logic implemented within the reward gate. |
What are the different types of secrets gate plugins? | SecretsGate scans instruction and output for API keys, tokens, and high-entropy strings. detect-secrets plugins cover AWS, GitHub, Slack, and generic entropy. Hits become RejectedSample rows, not silent drops. | The SecretGate plugin is divided into two main categories: AWS Secrets Gateway and GitHub Secrets Gateway. These plugins use various methods to detect secret information such as API keys, tokens, and high-entropy strings. |
What does PIIPseudonymizer do? | PIIPseudonymizer runs NER then Faker. The same real name maps to the same alias for the whole run when pii_faker_seed is fixed. Emails, phones, and locations are replaced in the configured fields only. | PIIPseudonymizer is used to anonymize real names by mapping them to aliases throughout the entire run, even if they appear in multiple fields. |
How does PIIPseudonymizer work? | PIIPseudonymizer runs NER then Faker. The same real name maps to the same alias for the whole run when pii_faker_seed is fixed. Emails, phones, and locations are replaced in the configured fields only. | PIIPseudonymizer uses two main components: NER (Natural Language Understanding) and Faker (Faking Real Names). It ensures that the same real name maps to the same alias across all fields during the entire run. |
What is ToxicityGate? | ToxicityGate runs Detoxify first. Scores below the pass threshold keep the row. Scores above the reject threshold drop it. The band in between can go to an optional LLM judge. | ToxicityGate is a platform that evaluates user-generated content based on toxicity scores. It uses algorithms and rules to determine if content meets certain criteria. |
How does ToxicityGate score content? | ToxicityGate runs Detoxify first. Scores below the pass threshold keep the row. Scores above the reject threshold drop it. The band in between can go to an optional LLM judge. | ToxicityGate uses a scoring system where scores below a specific threshold are kept, while scores above a higher threshold are dropped. These thresholds are defined within the platform's guidelines. |
What does Exact Dedup normalization do? | Exact dedup hashes normalized text and keeps the first copy. MinHash estimates Jaccard overlap on character n-grams. embedding_dedup writes a persistent index so later runs still catch repeats. | Exact Dedup normalization removes duplicate records based on exact matches of their key-value pairs. It ensures that each record is unique. |
How does MinHash estimate Jaccard overlap on character n-grams? | Exact dedup hashes normalized text and keeps the first copy. MinHash estimates Jaccard overlap on character n-grams. embedding_dedup writes a persistent index so later runs still catch repeats. | MinHash uses a technique called Jaccard similarity to calculate the overlap between two sets of data points, which can then be used as a measure of similarity for character n-gram comparisons. |
What does the 'clean_fields' setting do? | Text cleaning strips HTML, normalizes Unicode, fixes mojibake, collapses whitespace, and drops control characters. clean_fields defaults to instruction, input, and output. Cleaning runs before generation so the LLM never sees the raw junk. | The 'clean_fields' setting ensures that the cleaned data is properly formatted for use by the language model. |
": "Wh | SchemaGate checks required fields and token bounds per task type. Pretrain rows need non-empty output. Instruction rows need instruction and output. Failures are RejectedSample records with a structured reason. | t does SchemaGate check for each task type?",
"answer": "SchemaGate checks required fields and token bounds per task type."
}
{
"question": "When pretraining rows, what must they contain?",
"answer": "Pretraining rows must have non-empty output."
}
``` |
What are the main components of an exporter's process? | Exporters write Alpaca, ShareGPT, DPO, GRPO, PPO, and corpus JSONL into output_dir. Sidecars are manifest.json, dataset_card.md, and checksums.txt. Empty export files no longer crash a Hub push. | The main components of an exporter's process include writing Alpaca, sharing with ShareGPT, generating data using DPO, creating corpus JSONL, manifesting into output directory, collecting sidecar files such as manifest.json, dataset_card.md, and checksums.txt, and managing empty export files that cause Hub push crashes... |
What does 'push_to_hub' create? | push_to_hub creates a dataset repo if it is missing and uploads every export plus sidecars. push_format_to_hub sends one format. push_rejected_to_hub publishes the rejected split with the same provenance files. | PushToHub creates a dataset repo if it is missing. |
What does 'push_format_to_hub' send? | push_to_hub creates a dataset repo if it is missing and uploads every export plus sidecars. push_format_to_hub sends one format. push_rejected_to_hub publishes the rejected split with the same provenance files. | PushFormatToHub sends one format. |
What does 'push_rejected_to_hub' publish? | push_to_hub creates a dataset repo if it is missing and uploads every export plus sidecars. push_format_to_hub sends one format. push_rejected_to_hub publishes the rejected split with the same provenance files. | PushRejectedToHub publishes the rejected split with the same provenance files. |
|
curatorkit-testrun-Injector
Built using CuratorKIT — provenance-grounded curation and synthesis for LLM post-training.
| Method | qa |
| Backend | litellm |
| Model | openai/Qwen/Qwen2.5-0.5B-Instruct |
| Formats | alpaca |
| Artifact | dataset |
| Published | 2026-08-30 06:13 UTC |
Usage
from datasets import load_dataset
ds = load_dataset("ram-lexsi/curatorkit-testrun-Injector", "alpaca")
- Downloads last month
- 28