Miscellaneous cookbooks and code made available for purposes of education, reproducibility, and transparency.
| Example | Description |
|---|---|
| TLM-Demo-Notebook | Showcasing various applications of the Trustworthy Language Model |
| TLM-PII-Detection | Find and mask PII with the Trustworthy Language Model |
| TLM-Record-Matching | Using the Trustworthy Language Model to reliably match records between two different data tables. |
| TLM-SimpleQA-Benchmark | Benchmarking TLM and OpenAI LLMs on the SimpleQA dataset |
| benchmarking_hallucination_metrics | Notebook that compares the performance of popular hallucination detection metrics on a set of hallucination benchmarks. |
| fine_tuning_data_curation | Notebook showing how to use Cleanlab TLM and Cleanlab Studio to detect bad data in instruction tuning LLM datasets. |
| Detecting GDPR Violations with TLM | Notebook showing the code used to analyze application logs using TLM to detect GDPR violations |
| Customer Support AI Agent with NeMo Guardrails | Reliable customer support AI Agent with Guardrails and trustworthiness scoring |
| few_shot_prompt_selection | Notebook showing how to clean few-shot examples pool to improve prompt template for OpenAI LLM. |
| fine_tuning_classification | Notebook showing how to use Cleanlab Studio to improve the accuracy of fine-tuned LLMs for classification tasks. |
| generate_llm_response | Notebook showing how to generate LLM responses for customer service requests using Llama 2 and OpenAI's API. |
| gpt4-rag-logprobs | Notebook showing how to obtain logprobs from a GPT-4 based RAG system. |
| fine_tuning_mistral_beavertails | Analyze human annotated AI-safety-related labels (like toxicity) using Cleanlab Studio, and thus generate safer responses from LLMs. |
| Evaluating_Toxicity_Datasets_Large_Language_Models | Notebook on analyzing toxicity annotations in the Jigsaw dataset using Cleanlab Studio. |
| time_series_automl | Notebook showing how to model time series data in a tabular format and use AutoML with Cleanlab Studio to improve out-of-sample accuracy. |