🚀 Lamini-Prompt-Enchance

Lamini-Prompt-Enchance 是基于 MBZUAI/LaMini-Flan-T5-248M 微调而来的模型，可用于增强描述性提示，在评估集上取得了不错的成绩。

🚀 快速开始

from transformers import pipeline     

# load model and tokenizer from huggingface hub with pipeline
enhancer = pipeline("summarization", model="gokaygokay/Lamini-Prompt-Enchance", device=0)

prompt = "A blue-tinted bedroom scene, surreal and serene, with a mysterious reflected interior."
prefix = "Enhance the description: "
# enhance prompt
res = enhancer(prefix + prompt)

print(res[0]['summary_text'])

# A surreal and serene bedroom scene with a mysterious mirrored interior, 
# awash in blue and green hues.
# The room is adorned with intricate patterns and a mirrored wall, 
# creating a sense of mystery and tranquility.

✨ 主要特性

• 微调模型：基于 MBZUAI/LaMini-Flan-T5-248M 微调，在特定任务上表现更佳。
• 评估指标良好：在评估集上，Loss 为 2.0195，Rouge1 为 31.5042 等，显示出较好的性能。

📦 安装指南

文档未提及安装步骤，可参考 Hugging Face 上关于 transformers 库的安装说明。

💻 使用示例

基础用法

from transformers import pipeline     

# load model and tokenizer from huggingface hub with pipeline
enhancer = pipeline("summarization", model="gokaygokay/Lamini-Prompt-Enchance", device=0)

prompt = "A blue-tinted bedroom scene, surreal and serene, with a mysterious reflected interior."
prefix = "Enhance the description: "
# enhance prompt
res = enhancer(prefix + prompt)

print(res[0]['summary_text'])

# A surreal and serene bedroom scene with a mysterious mirrored interior, 
# awash in blue and green hues.
# The room is adorned with intricate patterns and a mirrored wall, 
# creating a sense of mystery and tranquility.

📚 详细文档

模型信息

此模型是 MBZUAI/LaMini-Flan-T5-248M 在未知数据集上的微调版本，在评估集上取得以下结果：

• Loss: 2.0195
• Rouge1: 31.5042
• Rouge2: 13.2633
• Rougel: 26.4176
• Rougelsum: 28.4846
• Gen Len: 19.0

训练超参数

训练过程中使用了以下超参数：

• learning_rate: 5e-05
• train_batch_size: 24
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5

训练结果

框架版本

• Transformers 4.42.3
• Pytorch 2.3.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1

📄 许可证

本模型采用 CC BY-NC 4.0 许可证。