marin-community/open-thoughts-4-30k-code-qwen3-30b-a3B-thinking-2507-annotated-32768-tokens-n8-reformatted

---
dataset_info:
features:
- name: row_id
dtype: int64
- name: id
dtype: string
- name: instruction_seed
dtype: string
- name: output
dtype: string
- name: source
dtype: string
- name: license
dtype: string
- name: dataset
dtype: string
- name: difficulty
dtype: int64
- name: solution
dtype: string
- name: index
dtype: string
- name: _source
dtype: string
- name: difficulty_reasoning
dtype: string
- name: __original_row_idx
dtype: int64
- name: ms_id
dtype: int64
- name: generated_text
dtype: string
- name: final_answer
dtype: string
- name: complete_responses_count
dtype: int64
splits:
- name: train
num_examples: 239704
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for Open-Thoughts-4-30K-Code-Qwen3-30B-A3B-Thinking-2507-Annotated-32768-Tokens-N8-Reformatted

## Overview
This dataset is a reformatted version of [marin-community/open-thoughts-4-30k-code-qwen3-30b-a3B-thinking-2507-annotated-32768-tokens-n8](https://huggingface.co/datasets/marin-community/open-thoughts-4-30k-code-qwen3-30b-a3B-thinking-2507-annotated-32768-tokens-n8). The original dataset contained 29,963 samples, each with 8 responses generated by the same model with different random seeds (stored in `generated_text`, `generated_text2`, ..., `generated_text8` columns). This reformatted version expands each response into its own row, resulting in **29,963 x 8 = 239,704 samples** with a single `generated_text` column.

The rows are ordered so that the `ms_id` ordering matches the reference dataset [marin-community/open-thoughts-4-30k-code-qwen3-32b-annotated](https://huggingface.co/datasets/marin-community/open-thoughts-4-30k-code-qwen3-32b-annotated). All 8 responses for a given prompt appear contiguously (e.g., rows 0-7 share the same prompt, rows 8-15 share the next prompt, and so on).

## Generation Details
- **Model:** [Qwen/Qwen3-30B-A3B-Thinking-2507](https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507)
- **Temperature:** 0.8
- **Max Output Tokens:** 32768
- **Responses per prompt:** 8 (different random seeds)

## Dataset Statistics
- **Number of Samples:** 239,704
- **Number of Unique Prompts:** 29,963
- **Responses per Prompt:** 8

## Dataset Structure

| Column | Description |
|--------|-------------|
| `row_id` | A unique row identifier (0 to 239,703) |
| `instruction_seed` | The original code problem/question text without chat formatting |
| `_source` | The origin dataset (e.g., `ai2-adapt-dev/opencode-2-code`); tracks data provenance |
| `output` | Reference output from the source dataset |
| `__original_row_idx` | The row index from the original source dataset before filtering/processing |
| `ms_id` | A unique sample identifier (shared across the 8 responses for the same prompt) |
| `generated_text` | A response including chain-of-thought with `<think>` tags, generated by Qwen3-30B-A3B-Thinking-2507 |
| `final_answer` | The extracted final answer from `\boxed{...}` after the `</think>` token, or `N/A` if the response is incomplete |
| `complete_responses_count` | Number of complete responses (0-8) for this prompt; a response is complete if it contains `</think>` followed by a valid `\boxed{...}` |

## Related Datasets
- [open-thoughts-4-30k-code-qwen3-30b-a3B-thinking-2507-annotated-32768-tokens-n8](https://huggingface.co/datasets/marin-community/open-thoughts-4-30k-code-qwen3-30b-a3B-thinking-2507-annotated-32768-tokens-n8) — Original dataset with 8 response columns per row