thwiki to Wikidata

Soft Prompts

The concept of "soft prompts" has evolved since its introduction, offering a flexible and parameter-efficient alternative to traditional prompt-based learning methods. Here's an overview of the key works on soft prompts:

The Introduction of Soft Prompts (2021): Brian Lester et al. introduced "Prompt Tuning" in EMNLP 2021. This method focuses on conditioning a frozen language model using tunable soft prompts, which are learnable vectors attached to the input rather than manually crafted text prompts. The work demonstrated that soft prompts could achieve competitive performance while maintaining a frozen model, which is more efficient than fine-tuning all parameters(Google Research)

Guiding Frozen Language Models with Learned Soft PromptsPosted by Brian Lester, AI Resident and Noah Constant, Senior Staff Software Engineer, Google Research Large pre-trained language models, which are...

SPoT: Soft Prompt Transfer (2022): Following the initial success of prompt tuning, SPoT: Soft Prompt Transfer was introduced by Tu Vu et al. in ACL 2022. This method enhances the performance of soft prompts by transferring them across tasks. SPoT learns a soft prompt from a source task and transfers it to a target task, significantly improving prompt tuning performance across various NLP tasks(

ACL Anthology

)(

DeepAI

SPoT: Better Frozen Model Adaptation through Soft Prompt TransferThere has been growing interest in parameter-efficient methods to apply pre-trained language models to downstream tasks. Building on the Prompt Tuning approach of Lester et al. (2021), which learns task-specific soft prompts to condition a frozen pre-trained model to perform different tasks, we propose a novel prompt-based transfer learning approach called SPoT: Soft Prompt Transfer. SPoT first learns a prompt on one or more source tasks and then uses it to initialize the prompt for a target task. We show that SPoT significantly boosts the performance of Prompt Tuning across many tasks. More remarkably, across all model sizes, SPoT matches or outperforms standard Model Tuning (which fine-tunes all model parameters) on the SuperGLUE benchmark, while using up to 27,000x fewer task-specific parameters. To understand where SPoT is most effective, we conduct a large-scale study on task transferability with 26 NLP tasks in 160 combinations, and demonstrate that many tasks can benefit each other via prompt transfer. Finally, we propose an efficient retrieval approach that interprets task prompts as task embeddings to identify similar tasks and predict the most transferable source tasks for a novel target task.

https://arxiv.org/abs/2110.07904

Applications in Multi-task Learning and Flexibility: Soft prompts are particularly advantageous for multi-task learning, allowing a single model to switch between tasks efficiently by changing the prompts. They are also widely applicable in tasks such as sentiment analysis, machine translation, and language generation(

Software Mind

These papers form the foundation of soft prompt research, focusing on optimizing parameter efficiency, enabling multi-task learning, and providing flexibility in handling diverse tasks. For further reading, I recommend checking the EMNLP 2021 and ACL 2022 proceedings.

Page updated

Google Sites

Report abuse