llm-semantic-router/modality-routing-dataset
收藏Hugging Face2026-03-20 更新2026-03-29 收录
下载链接:
https://hf-mirror.com/datasets/llm-semantic-router/modality-routing-dataset
下载链接
链接失效反馈官方服务:
资源简介:
---
pretty_name: Modality Routing Dataset
task_categories:
- text-classification
language:
- en
configs:
- config_name: default
data_files:
- split: train
path: train.jsonl
- split: validation
path: validation.jsonl
- split: test
path: test.jsonl
---
# Modality Routing Dataset
This dataset materializes the dynamic modality routing data builder used by the local
mmBERT-32K modality router training pipeline. The export is intended for review,
versioning, and uploading to a Hugging Face dataset repository.
## Labels
| Label | ID | Description |
|-------|----|-------------|
| AR | 0 | Text-only requests that should route to an autoregressive LLM. |
| DIFFUSION | 1 | Image-generation requests that should route to a diffusion model. |
| BOTH | 2 | Requests that benefit from both text and image responses. |
## Schema
| Column | Type | Description |
|--------|------|-------------|
| text | string | Input user prompt |
| label | int64 | Integer class id |
| label_name | string | Human-readable class label |
## Splits
| Split | Rows | AR | DIFFUSION | BOTH |
|-------|------|------------|--------------------|--------------|
| train | 3525 | 1399 | 1400 | 726 |
| validation | 756 | 300 | 300 | 156 |
| test | 756 | 301 | 300 | 155 |
## Export Configuration
- `max_samples`: 6000
- `synthesize_both`: 0
- `vllm_synthesis_enabled`: disabled
- `vllm_endpoint`: None
- `vllm_model`: None
- `split_strategy`: 70% train / 15% validation / 15% test with random_state=42
## Sources
- `FredZhang7/stable-diffusion-prompts-2.47M`
- `succinctly/midjourney-prompts`
- `Falah/image_generation_prompts_SDXL`
- `nateraw/parti-prompts`
- `fal/image-generation-prompts`
- `OpenAssistant/oasst2`
- `tatsu-lab/alpaca`
- `databricks/databricks-dolly-15k`
- `stingning/ultrachat`
- `lmsys/lmsys-chat-1m`
- `allenai/WildChat`
- `mqliu/InterleavedBench`
- Optional vLLM-generated BOTH prompts when enabled
## Files
- `train.jsonl`, `validation.jsonl`, `test.jsonl`: upload-friendly JSONL splits
- `label_mapping.json`: label to integer mapping
- `dataset_stats.json`: row counts per split and label
- `export_config.json`: reproducibility metadata for this export
- `hf_dataset/`: local `DatasetDict.save_to_disk()` artifact
pretty_name: 多模态路由数据集(Modality Routing Dataset)
task_categories:
- 文本分类
language:
- 英语
configs:
- config_name: default
data_files:
- split: train
path: train.jsonl
- split: validation
path: validation.jsonl
- split: test
path: test.jsonl
# 多模态路由数据集(Modality Routing Dataset)
本数据集具象化了本地mmBERT-32K多模态路由训练流水线中所使用的动态多模态路由数据构建器。本次导出旨在用于审核、版本管理,并上传至Hugging Face数据集仓库。
## 标签
| 标签名称 | 编号 | 描述 |
|--------|----|-------------|
| AR(Autoregressive LLM) | 0 | 仅需文本的请求,应路由至自回归大语言模型。 |
| DIFFUSION | 1 | 图像生成请求,应路由至扩散模型。 |
| BOTH | 2 | 可同时受益于文本与图像响应的请求。 |
## 数据结构
| 列名 | 数据类型 | 描述 |
|--------|------|-------------|
| text | 字符串(string) | 用户输入提示词 |
| label | 64位整数(int64) | 整数类别编号 |
| label_name | 字符串(string) | 人类可读的类别标签名称 |
## 数据集划分
| 划分集 | 样本总数 | AR类样本数 | DIFFUSION类样本数 | BOTH类样本数 |
|-------|------|------------|--------------------|--------------|
| train | 3525 | 1399 | 1400 | 726 |
| validation | 756 | 300 | 300 | 156 |
| test | 756 | 301 | 300 | 155 |
## 导出配置
- 最大样本数(max_samples):6000
- 双边样本合成量(synthesize_both):0
- vLLM合成功能(vllm_synthesis_enabled):已禁用
- vLLM服务端点(vllm_endpoint):无
- vLLM模型(vllm_model):无
- 划分策略(split_strategy):以随机种子42按70%训练集、15%验证集、15%测试集的比例进行划分
## 数据源
- `FredZhang7/stable-diffusion-prompts-2.47M`
- `succinctly/midjourney-prompts`
- `Falah/image_generation_prompts_SDXL`
- `nateraw/parti-prompts`
- `fal/image-generation-prompts`
- `OpenAssistant/oasst2`
- `tatsu-lab/alpaca`
- `databricks/databricks-dolly-15k`
- `stingning/ultrachat`
- `lmsys/lmsys-chat-1m`
- `allenai/WildChat`
- `mqliu/InterleavedBench`
- 启用时可额外生成vLLM制作的BOTH类提示词
## 相关文件
- `train.jsonl`、`validation.jsonl`、`test.jsonl`:适配上传的JSONL格式数据集划分文件
- `label_mapping.json`:标签与整数编号的映射文件
- `dataset_stats.json`:各划分集及各类别的样本统计文件
- `export_config.json`:本次导出的可复现元数据文件
- `hf_dataset/`:本地`DatasetDict.save_to_disk()`生成的工件目录
提供机构:
llm-semantic-router


