Action item
- Use dissertation master to help to write article
- Do more literature survey
- Use knowledge link to break down each knowledge and link them: RAG, SpD, REST …
- How to do 中英文 together?
| **Use AI to assist paper writing in in another blog [[2024-05-12-AI_for_Paper | AI Paper]].** |
https://www.semanticscholar.org/search?q=Retrieval%20speculative%20decode&sort=relevance
Source
- REST: Retrieval-based speculative decoding: https://arxiv.org/pdf/2311.08252
- Speculative RAG: https://arxiv.org/pdf/2407.08223
- Retrieval-based Speculative decoding improve RAG performance: https://www.linkedin.com/pulse/how-does-retrieval-based-speculative-decoding-improve-daniel-chen-ksv3c/
Speculative Decode
- Speculative Decoding with Big Little Decoder!!
- https://arxiv.org/abs/2302.07863
- https://zhuanlan.zhihu.com/p/684217993: good 知乎 paper
- LLM推理加速新范式!推测解码(Speculative Decoding)最新综述-CSDN博客
- 2302.01318 (arxiv.org) nice explanation of the speculative sampling math!
- Speculative Sampling — Intuitively and Exhaustively Explained (substack.com) with code and example!!
- https://www.jinghong-chen.net/an-mathematical-intuition-of-speculative-sampling/ intuition of the accept with resampling!
- Cloud-edge hybrid SpD! [2302.07863] Speculative Decoding with Big Little Decoder (arxiv.org)