Madras1/glm-5-code-distilled2.6k

---
language:
- en
license: apache-2.0
size_categories:
- 1K<n<10K
task_categories:
- text-generation
tags:
- code
- code-generation
- distillation
- synthetic
- reasoning
- chain-of-thought
- python
- glm
pretty_name: "GLM-5 Code Distillation"
dataset_info:
features:
- name: id
dtype: int64
- name: problem
dtype: string
- name: function_name
dtype: string
- name: reasoning
dtype: string
- name: code
dtype: string
- name: model
dtype: string
splits:
- name: train
num_examples: 2654
---

# GLM-5 Code Distillation Dataset

A synthetic code generation dataset created by distilling **GLM-5**. Each example contains a Python coding problem, the model's chain-of-thought reasoning, and a **verified correct** solution that passes automated test execution.

## Key Features

- **Execution-verified**: Every solution passed automated test cases in a sandboxed subprocess.
- **Chain-of-thought reasoning**: Includes the model's reasoning process for training reasoning-capable models.
- **Seed-generated problems**: Diverse problems generated from real code corpora seeds.
- **Deduplicated**: Exact dedup (code + prompt MD5) + near-dedup (MinHash LSH, Jaccard > 0.7).
- **Pure Python**: Standard library only, no external dependencies.

## Statistics

| Metric | Value |
|---|---|
| Total examples | 2,654 |
| With reasoning | 2,654 (100.0%) |
| Avg code length | 1167 chars |
| Avg reasoning length | 8551 chars |
| Teacher model | `zai-org/GLM-5` |

## Schema

| Field | Type | Description |
|---|---|---|
| `id` | int | Sequential index |
| `problem` | string | Coding problem description (cleaned) |
| `function_name` | string | Function to implement |
| `reasoning` | string | Chain-of-thought |
| `code` | string | Verified Python solution |
| `model` | string | Teacher model |

## Usage

```python
from datasets import load_dataset
ds = load_dataset("Madras1/glm-5-code-distilled2.6k")
example = ds["train"][0]
print(example["problem"])
print(example["code"])
```

## License

Apache 2.0