LumiVore/lumivore-stage1-training-data
收藏Hugging Face2026-03-22 更新2026-03-29 收录
下载链接：
https://hf-mirror.com/datasets/LumiVore/lumivore-stage1-training-data
下载链接
链接失效反馈官方服务：
资源简介：
# Lumivore Stage 1 Training Dataset
**Version:** 4.5-A
**Created:** March 2026
**Purpose:** Full fine-tuning of Qwen2.5-0.5B base model to MoE architecture
**Total Examples:** ~11,600
**Format:** Alpaca (instruction, input, output)
---
## Overview
This dataset was used for Stage 1 of the Lumivore-1.2B training pipeline. It combines general agentic task data with reasoning examples to teach the base model (Qwen2.5-0.5B) how to perform tool use, reasoning, and structured outputs before MoE conversion.
---
## Data Sources
| Source | Description | Proportion |
|--------|-------------|------------|
| **TerminalTrajectories** | Terminal command sequences and bash interactions | ~50% |
| **OpenThoughts** | Chain-of-thought reasoning examples | ~50% |
---
## Dataset Characteristics
- **Task types:** Shell commands, file operations, reasoning chains, structured outputs
- **Style:** Technical, direct, focused on tool use and system interaction
- **Quality:** Filtered for correctness, deduplicated with MinHash
- **Augmentation:** Original examples with linguistic variations (3-5x)
---
## Training Configuration
Used with the following hyperparameters:
```python
# Stage 1 Training
- Base model: Qwen/Qwen2.5-0.5B-Instruct
- Batch size: 1 (micro), gradient_accumulation: 16
- Effective batch: 16
- Max sequence length: 1024
- Learning rate: 2e-5
- Optimizer: 8-bit AdamW
- Epochs: 3
- Steps: ~2,058
- Duration: ~5.4 hours on AMD RX 7600 XT
```
---
## Files
- `train.jsonl` — Training examples (~10,989 after split)
- `val.jsonl` — Validation examples (~5% split)
- `README.md` — This documentation
---
## Usage
```python
from datasets import load_dataset
dataset = load_dataset("LumiVore/lumivore-stage1-training-data")
train_data = dataset["train"]
val_data = dataset["validation"]
```
---
## Related
- **Stage 2 Dataset:** `LumiVore/lumivore-stage2-training-data` — OpenClaw-specific fine-tuning
- **Stage 3 Dataset:** `LumiVore/lumivore-stage3-identity-dataset` — Identity and conversational training
- **Model:** `LumiVore/lumivore-1.2b` (when published)
---
## Citation
If you use this dataset, please cite:
```bibtex
@dataset{lumivore2026stage1,
title={Lumivore Stage 1 Training Dataset},
author={LumiVore AI},
year={2026},
url={https://huggingface.co/datasets/LumiVore/lumivore-stage1-training-data}
}
```
---
*Created for the Lumivore-1.2B training pipeline*
提供机构：
LumiVore
