| --- |
| language: |
| - en |
| license: apache-2.0 |
| size_categories: |
| - 100K<n<1M |
| task_categories: |
| - text-generation |
| - question-answering |
| tags: |
| - llm-evaluation |
| - benchmarking |
| - ai-safety |
| - sft |
| - synthetic |
| - prompt-engineering |
| - rlhf |
| - hallucination |
| - model-comparison |
| pretty_name: LLM Evaluation SFT 100K |
| --- |
| |
| # LLM Evaluation SFT 100K |
|
|
| A synthetic supervised fine-tuning dataset of 100,000 high-quality conversations covering LLM evaluation methodology, benchmarking, safety assessment, and prompt optimization. Designed to train AI assistants that can help ML engineers and researchers rigorously evaluate and improve language models. |
|
|
| ## Dataset Description |
|
|
| This dataset covers the full spectrum of LLM evaluation practice across 7 specialized categories. Each record follows the ShareGPT format with a practitioner-level question and a detailed, methodology-rich response including working code examples. |
|
|
| ## Categories |
|
|
| | Category | Description | |
| |---|---| |
| | `benchmark_design` | Designing reliable, contamination-free benchmarks | |
| | `model_comparison` | Systematic model comparison, consistency measurement | |
| | `hallucination_detection` | Measuring and reducing hallucinations in RAG systems | |
| | `eval_frameworks` | LM Eval Harness, RAGAS, DeepEval, LangSmith comparison | |
| | `prompt_optimization` | Systematic prompt engineering with statistical testing | |
| | `safety_evaluation` | Red-teaming, bias testing, deployment safety gates | |
| | `finetuning_evaluation` | Fine-tuned vs base model comparison, catastrophic forgetting | |
|
|
| ## Format |
|
|
| ShareGPT format: |
| ```json |
| { |
| "conversations": [ |
| {"from": "human", "value": "...evaluation question..."}, |
| {"from": "gpt", "value": "...methodology-rich response with code..."} |
| ], |
| "metadata": {"category": "...", "context": "..."}, |
| "id": "uuid" |
| } |
| ``` |
|
|
| ## Use Cases |
|
|
| - Fine-tuning AI assistants for ML evaluation tasks |
| - Training models to reason about benchmark methodology |
| - Building AI-assisted evaluation pipelines |
| - Educating teams on LLM quality measurement |
| - Safety evaluation tooling and red-teaming assistance |
|
|
| ## Quality Notes |
|
|
| All responses include working Python code examples using popular evaluation frameworks (lm-eval, RAGAS, DeepEval, LangSmith), statistical testing methods, and production-ready evaluation patterns. |
|
|
