y0sif/Arcwright-Rig

---
license: apache-2.0
language:
- en
tags:
- rust
- code
- instruction-tuning
- rig
- chatml
size_categories:
- 1K<n<10K
task_categories:
- text-generation
---

# Arcwright-Rig

An instruction-tuning dataset for the **[Rig](https://github.com/0xPlaygrounds/rig)** Rust crate, built for the [Arcwright](https://huggingface.co/y0sif/arcwright-E4B-v1) fine-tuned model.

Rig is a Rust library for building LLM-powered applications with support for multiple providers, tool use, embeddings, and RAG pipelines.

## Dataset Summary

- **697 instruction-response pairs** covering ai agents, tool use / function calling, embeddings, vector stores, rag pipelines, and multi-provider support (openai, anthropic)
- Generated from real source code using the **OSS-Instruct** methodology via Claude Code sub-agents
- Validated for structural correctness and deduplicated using MinHash (Jaccard threshold 0.7)
- Format: **ChatML** (messages array with system/user/assistant roles)

## Category Distribution

| Category | Count | % |
|----------|-------|---|
| Code Generation | 236 | 33% |
| Code Explanation | 113 | 16% |
| Api Usage | 113 | 16% |
| Bug Detection | 77 | 11% |
| Refactoring | 95 | 13% |
| Test Generation | 63 | 9% |

## Format

Each example is a JSON object with a `messages` array:

```json
{
"messages": [
{"role": "system", "content": "You are an expert Rust programmer specializing in the rig crate and modern Rust development patterns."},
{"role": "user", "content": "Show how to create a basic AI agent with Rig that uses OpenAI as the provider and can respond to user messages."},
{"role": "assistant", "content": "..."}
],
"category": "code_generation",
"crate": "rig"
}
```

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("y0sif/Arcwright-Rig")
print(dataset["train"][0]["messages"])
```

## Part of Arcwright

This dataset is one of three crate-specific datasets used to train [Arcwright-E4B-v1](https://huggingface.co/y0sif/arcwright-E4B-v1):

| Dataset | Crate | Pairs |
|---------|-------|-------|
| **[Arcwright-Leptos](https://huggingface.co/datasets/y0sif/Arcwright-Leptos)** | Leptos | 2,046 |
| **[Arcwright-Axum](https://huggingface.co/datasets/y0sif/Arcwright-Axum)** | Axum | 741 |
| **[Arcwright-Rig](https://huggingface.co/datasets/y0sif/Arcwright-Rig)** | Rig | 697 |

## Source

All instruction pairs were generated from source code in the [Rig repository](https://github.com/0xPlaygrounds/rig). Code was chunked using tree-sitter into meaningful units (functions, impl blocks, modules), then used as seed material for instruction generation.

## License

Apache 2.0