instruction stringlengths 7 153 | input stringlengths 174 316 | output stringlengths 65 455 |
|---|---|---|
How does the system ensure that every generated answer matches its original source? | 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. | 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. |
How can I obtain an AWS access key for use with my project? | 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. | To obtain an AWS access key, you will need to sign up for an AWS account first. After signing up, follow the steps provided on the AWS website to generate an access key. |
What is the purpose of the PIIPseudonymizer? | 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. | PIIPseudonymizer replaces names and emails with consistent fake values. |
What are the steps involved in running a curator configuration? | 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. | Running a curator configuration involves several steps:
1. Naming the dataset
2. Configuring the fields to be used
3. Setting up the generation task
4. Choosing the export formats |
What is the purpose of the 'pip install' command used to install vLLM? | 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. | The 'pip install' command is used to install the vLLM package on your system. |
How does the field mapping process work? | 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 field mapping process involves remapping the source columns into the corresponding schema fields. This ensures that the data is correctly formatted according to the specified schema. |
What is the purpose of AWS S3 bucket encryption? | 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. | AWS S3 bucket encryption ensures data privacy and integrity by encrypting the contents of the buckets. This protects sensitive information from unauthorized access and breaches. |
How can I create an IAM role with specific permissions for accessing my Amazon S3 bucket? | 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. | To create an IAM role with specific permissions for accessing your Amazon S3 bucket, you need to follow these steps:
- Create an IAM user with the necessary permissions.
- Add the user to the IAM role using the `aws iam add-user` command.
- Use the `aws iam attach-role-policy` command to attach the IAM policy associat... |
What is the purpose of the AWS Access Key used in 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. | The AWS Access Key is used to authenticate the user when writing a preference. |
How does GRPO generate different completions for each prompt? | 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 generates multiple completations per prompt due to varying temperatures, which helps to create diverse outputs with different judge scores. The temperature spread is set up such that the group of completions remains distinct from each other. |
What is the purpose of the AWS access key provided? | 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. | The AWS access key is used to authenticate with Amazon Web Services. |
What is the purpose of evol-instruct in the context of security-classifier research? | 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. | Evol-instruct rewrites a seed instruction into a harder variant. It involves running multiple rewrite steps and asking the model to solve the evolved prompt, which helps improve the accuracy of the generated answers. |
What is the primary purpose of using AWS credentials in the context of Amazon Web Services (AWS)? | 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 primary purpose of using AWS credentials is to enable secure authentication and authorization within the AWS environment. |
What are two possible reasons for adversarial QA to occur? | 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. | Adversarial QA can be triggered by injecting a failure into a subset of answers after generation. Injection rate indicates the fraction of rows that receive a known bad span. Injection types include contradictions source and parameter drift. The metadata flag 'injected_failure' marks these rows. |
What is the purpose of the 'hallucination_threshold' parameter in the HallucinationGate scoring system? | 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. | The 'hallucination_threshold' parameter in the HallucinationGate scoring system drops rows below a certain cutoff. This ensures that the system only considers questions that have been explicitly asked or mentioned within the context provided. |
What is the purpose of the 'reward_threshold' parameter in the RewardGate scoring system? | 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 determines when a row will be considered low-scoring and will not be included in the final evaluation. It allows for early warning mechanisms to flag potential issues or outliers. |
What does the 'coverage' parameter measure? | 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 coverage parameter measures how many rows of data have been used to train or evaluate the classifier. |
How does the diagnostic probe generate new rejected rows? | 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. | The diagnostic probe re-generates a rejected row based on different temperature settings and decides which retry count is considered a successful recovery. It then returns these new rows back to the system with their provenance. |
How does the 'enable_reward_refiner' operation 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. | The 'enable_reward_refiner' operation is used to rewrite a low-reward answer and run it again. This means that any previous runs of the reward step will be discarded and new runs will start from scratch. |
What is the purpose of SecretsGate plugin? | 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 purpose of SecretsGate plugin is to scan instructions and outputs for API keys, tokens, and high-entropy strings. |
What does the pseudonymization process entail? | 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. | The pseudonymization process involves two main steps: NER (Natural Language Processing) and Faker (Fake Name Generator). During the NER phase, real names are mapped to fictitious aliases based on the context of the data. In this case, the real name 'PIIPseudonymizer' is used as the mapping reference, while the alias 'P... |
What is the default score threshold for 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. | The default score threshold for ToxicityGate is set to keep rows with scores below the pass threshold. |
What is the purpose of the exact dedup hash normalization process? | 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. | The exact dedup hash normalization process normalizes the text while keeping only the first copy to ensure uniqueness. |
What is the purpose of the AWS access key and GitHub PAT provided? | 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 AWS access key is used to authenticate with Amazon Web Services (AWS), while the GitHub PAT is used to authenticate with GitHub. |
How does SchemaGate check if a row is valid? | 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. | SchemaGate checks if all required fields and token bounds are present in each task type. |
What are the three main components of the dataset used for security classification? | 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 dataset contains three main components: Alpaca, ShareGPT, and PPO. |
```json | 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. | {
"question": "What does 'push_to_hub' do?",
"answer": "Pushes a dataset to a repository if it doesn't already exist."
}
```
[
{
"question": "What does 'push_format_to_hub' do?",
"answer": "Uploads a specific format of an exported dataset along with any sidecar files."
},
{
"question": "How does ... |
What is the purpose of 'max_samples' parameter in the 'max_samples truncates the passed list after gates.' statement? | max_samples truncates the passed list after gates. Per-reader max_samples on a dataset dict caps that source before concat. Use the dict form when several Hub sets are mixed. | The purpose of 'max_samples' parameter in this statement is to specify how many samples should be retained from each reader during data preprocessing. |
What is the purpose of the 'Re-running the same config resumes instead of repeating completed LLM calls' option in the Checkpoint Dumps generation batch? | Checkpoints dump generation batches under output_dir/.checkpoints. Re-running the same config resumes instead of repeating completed LLM calls. Disable with enable_checkpoint=False for short Colab smoke tests. | This option allows you to generate new checkpoint dumps without having to restart the entire training process from scratch. |
What does the 'LiteLLM' variable represent? | LiteLLM is the default generation backend. llm_concurrency is the async worker pool. llm_timeout and llm_max_retries apply per call. Custom api_base points at vLLM or Ollama without changing the rest of the config. | The 'LiteLLM' variable represents the default generation backend. |
What is the purpose of using token windows in token-based QA? | PDF ingest splits a document into chunks before QA. heading strategy cuts on titles. sentence and fixed strategies use token windows with overlap. OCR is off by default and only needed for scanned pages. | Token windows are used to identify and extract relevant tokens from large texts during natural language processing tasks such as question answering or text classification. By allowing overlapping token windows, we can capture more context and reduce noise, making it easier to identify the most likely answers. |
What is the purpose of using StratifiedSampler in the given context? | Ingest can take several Hub datasets, each with its own preprocessing_fn and max_samples. StratifiedSampler then matches target_distribution on source_dataset so one corpus does not drown the mix. | StratifiedSampler is used to ensure that each target distribution is represented proportionally across all source datasets. This helps maintain diversity and avoid any single corpus being overwhelmed by others. |
What is the purpose of the 'rejected' row in the dataset? | Rejected rows keep the original text, a rejection_reason, and provenance. Downstream training should read the passed JSONL only. The rejected split is for debugging gates and for a separate Hub dump. | The purpose of the 'rejected' row is to keep the original text, a rejection reason, and provenance. It serves as a debugging gate or a separate Hub dump. |
How does the AWS access key work? | Field mapping {'text': 'text'} is required for pretrain seed JSONL. Mapping text onto output made construction_returned_none because the pretrain reader already stored the column as text. | The AWS access key is used to authenticate with Amazon Web Services (AWS). It's a string of alphanumeric characters that includes both uppercase and lowercase letters, numbers, and special characters. The key helps secure your connection to AWS services by providing authentication credentials. |
What is the purpose of the `GH_WORK_TOKEN` environment variable? | Colab secrets expected here are GH_WORK_TOKEN, HF_TOKEN, and optionally OPENROUTER_API_KEY. The clone uses branch devrel. uv pip install -e '.[all]' is the editable install used by every notebook. | The `GH_WORK_TOKEN` environment variable is used to authenticate with GitHub's Workspaces API, which is part of the GitHub Actions workflow. |
How can I obtain an AWS access key and GitHub PAT? | Export format lists are per notebook. QA defaults to alpaca and sharegpt. Preference defaults to dpo. GRPO defaults to grpo and ppo. Turning a format off only skips that JSONL, not the rest of the run. | To obtain an AWS access key, you need to create an IAM role with the necessary permissions. Then, use the AWS Management Console or the AWS CLI to get the access key from the role. For GitHub, you'll need to sign up for a free plan and then set up a personal access token. Make sure your credentials are stored securely. |
What is the minimum number of rows required for a good performance in a Colab run? | A presentable Colab run needs enough seed rows that gates can both keep and drop examples. Three near-identical chunks collapse into an all-pass or all-reject screenshot. Thirty to forty distinct passages avoid that. | The minimum number of rows required for a good performance in a Colab run depends on various factors such as the complexity of the problem being solved, the computational resources available, and the specific implementation details. However, it is generally recommended to have at least 30 to 40 distinct passages to ens... |
|
curatorkit-testrun-Prompt-Template
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 09:29 UTC |
Usage
from datasets import load_dataset
ds = load_dataset("ram-lexsi/curatorkit-testrun-Prompt-Template", "alpaca")
- Downloads last month
- 101