[Adapt] [Seminar] An Introduction of Decoding Strategies

[贾琪Jia, Qi]

Hi, Adapters,While extreme-scale language models have demonstrated exceptional performance on a variety of language tasks, the degree of control over these language models through pure prompting can often be limited. Directly fine-tuning LLMs can be effective for tailoring them, but it can be either extremely costly or not even feasible for the broader community.


Alternatively, decoding strategies as a kind of inference-time algorithm are gaining increasing attention which can also tailor a language model without accessing its parameters. An appropriate decoding strategy helps the model to generate fluent, coherent, and diverse outputs, as well as satisfy the user-specific objective.


In this talk, I will give an introduction to decoding strategies, especially the sampling-based ones, with a possible explanation for the efficacy of these approaches.


Hope you find this talk interesting.




Time: Wed 10 am. - 11:30 am.
Meeting link: https://teams.microsoft.com/l/meetup-join/19%3ameeting_M2VmMTU5MzgtODUzOC00NmU4LTg0MzktNGFjNDdiMmIwYTI1%40thread.v2/0?context=%7b%22Tid%22%3a%225cdc5b43-d7be-4caa-8173-729e3b0a62d9%22%2c%22Oid%22%3a%221a8b9fa0-af57-4a1c-9390-22d1c201d622%22%7d




Best wishes,

Qi Jia


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://cs.sjtu.edu.cn/pipermail/adapt/attachments/20231024/f3e9ee56/attachment.htm>