davanstrien/test-chunked-sprint-D1

Generated Responses Dataset

This dataset contains generated responses for prompts from davanstrien/haiku_dpo.

Generation Details

• —Source Dataset: davanstrien/haiku_dpo
• —Source Split: train
• —Input Column: question (plain text prompts)
• —Model: Qwen/Qwen2.5-3B-Instruct
• —Rows Processed: 5
• —Batches: 3 (chunk size: 2)
• —Generation Date: 2026-04-14T13:42:30.674651
• —Script: generate-responses-chunked.py (experimental streaming version)

Sampling Parameters

• —Temperature: 0.7
• —Top P: 0.8
• —Top K: 20
• —Min P: 0.0
• —Max Tokens: 64
• —Repetition Penalty: 1.0

Hardware Configuration

• —Tensor Parallel Size: 1
• —GPU Configuration: 1 GPU(s)
• —Max Model Length: 4,096 tokens

Dataset Structure

The dataset contains all columns from the source dataset plus:

• —response: The generated response from the model

Generation Script

Generated using the streaming vLLM inference script from uv-scripts/vllm.

To reproduce:

uv run https://huggingface.co/datasets/uv-scripts/vllm/raw/main/generate-responses-chunked.py \
    davanstrien/haiku_dpo \
    <output-dataset> \
    --model-id Qwen/Qwen2.5-3B-Instruct \
    --prompt-column question \
    --split train \
    --chunk-size 2 \
    --temperature 0.7 \
    --top-p 0.8 \
    --top-k 20 \
    --max-tokens 64 --max-model-len 4096

Notes

This dataset was generated using the experimental streaming/chunked pipeline. Unlike the standard generate-responses.py, this script processes the dataset in a streaming fashion using IterableDataset.map() + push_to_hub(), which reduces memory usage and enables progressive upload of Parquet shards.