ai-mitra/prompt-slimmer-slm
Text Generation • Updated
id stringclasses 9
values | family stringclasses 8
values | split stringclasses 1
value | category stringclasses 8
values | label_origin stringclasses 1
value | input stringclasses 9
values | output stringclasses 9
values | must_preserve listlengths 1 4 |
|---|---|---|---|---|---|---|---|
coding-1 | polite-code | train | coding | synthetic | Could you please help me write a Python function that takes a list of integers as input and returns only the numbers from that list which are even numbers? Please provide the Python code. | Write a Python function that takes a list of integers and returns only the even numbers. | [
"Python",
"function",
"list of integers",
"even numbers"
] |
coding-2 | polite-code | train | coding | synthetic | I would appreciate help writing a JavaScript function to sort a list of names alphabetically. Please write that function for me. | Write a JavaScript function to sort a list of names alphabetically. | [
"JavaScript",
"function",
"names",
"alphabetically"
] |
summary-1 | summary-request | train | summarization | synthetic | I would like a brief summary of the following meeting notes. Please summarize them briefly. The launch was postponed because testing is incomplete. | Summarize these meeting notes: The launch was postponed because testing is incomplete. | [
"launch",
"postponed",
"testing is incomplete"
] |
rag-1 | grounded-request | train | RAG | synthetic | Please help answer the question using the context below. Use only the supplied context. <protect>Context: The library closes at 18:00. Question: When does the library close?</protect> | Answer using the context below. Use only the supplied context. <protect>Context: The library closes at 18:00. Question: When does the library close?</protect> | [
"Use only the supplied context."
] |
agent-1 | agent-request | train | agent instructions | synthetic | Please help me review this proposed deployment plan. I would appreciate your review. Do not execute commands. Return JSON. | Review this proposed deployment plan. Do not execute commands. Return JSON. | [
"deployment plan",
"Do not execute commands.",
"Return JSON."
] |
reason-1 | reasoning-request | train | reasoning | synthetic | Please explain the reason why the proposed argument is circular. I would appreciate an explanation of the circular reasoning. | Explain why the proposed argument is circular. | [
"argument",
"circular"
] |
chat-1 | conversation-request | train | conversational | synthetic | I would really appreciate it if you could draft a friendly greeting for a new colleague joining our team. | Draft a friendly greeting for a new colleague joining our team. | [
"friendly greeting",
"new colleague",
"team"
] |
extract-1 | extraction-request | train | extraction | synthetic | Please help extract the company names from the text below. I would appreciate your help with this extraction. <protect>Amber acquired Birch.</protect> | Extract the company names from this text: <protect>Amber acquired Birch.</protect> | [
"company names"
] |
structured-1 | structured-request | train | structured output | synthetic | Please help me construct a response describing the current status. Return valid JSON. Use this schema: {"status": "string"} | Describe the current status. Return valid JSON. Use this schema: {"status": "string"} | [
"Return valid JSON."
] |
Prompt Slimmer SLM — Demo Dataset
Synthetic examples for experimenting with prompt rewriting and sentence selection. Exported without changing the examples or their original splits from the shared GitHub codebase.
| Configuration | Train | Validation | Test | Purpose |
|---|---|---|---|---|
rewrites (default) |
9 | 2 | 2 | Original prompt → shorter target |
selector |
256 | 64 | 64 | KEEP/DROP labels for source spans |
Load
from datasets import load_dataset
rewrites = load_dataset("ai-mitra/prompt-slimmer-slm", "rewrites")
selector = load_dataset("ai-mitra/prompt-slimmer-slm", "selector")
print(rewrites["train"][0])
Install the optional loader with pip install datasets. JSONL files can also be downloaded and read directly with Python's json module.
Fields
Rewrites: id, family, split, category, label_origin, input, output, must_preserve. Required phrases identify content that must appear in an accepted rewrite. Examples cover coding, reasoning, summarization, RAG, agents, conversation, extraction and structured output; coverage is extremely small.
Selector: id, family, split, label_origin, prompt, keep. Boolean labels align to the GitHub project's segment_prompt function, not words or model tokens.
Family assignments are isolated across splits. To use the training CLI, concatenate the train, validation and test JSONL files for the selected configuration, retaining each record's split field. Training requires --allow-demo because the examples are synthetic.
Intended use and limitations
These are workflow fixtures, not human-reviewed training data or a meaningful production benchmark. The rewriting adapter trained on the 9 training examples produced no accepted rewrites on the 2 validation prompts. The rewrite test split remains reserved. Selector test results are documented on GitHub and must not be attributed to the rewriter.
Use this dataset to inspect the schema and reproduce the demo. For useful compression, collect diverse reviewed pairs, preserve untouched evaluation sets, and compare downstream answer quality. Do not claim broad performance from these examples. No private prompts or credentials are included. No dataset license is specified in this release.
provenance.json records source paths, original file hashes, exported file hashes and counts.
- Downloads last month
- -
Models trained or fine-tuned on ai-mitra/prompt-slimmer-slm