Leveraging Speculative Sampling and KV-Cache Optimizations Together for Generative AI using OpenVINO
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2023
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909163752259584 |
|---|---|
| author | Barad, Haim Aidova, Ekaterina Gorbachev, Yury |
| author_facet | Barad, Haim Aidova, Ekaterina Gorbachev, Yury |
| contents | Inference optimizations are critical for improving user experience and reducing infrastructure costs and power consumption. In this article, we illustrate a form of dynamic execution known as speculative sampling to reduce the overall latency of text generation and compare it with standard autoregressive sampling. This can be used together with model-based optimizations (e.g. quantization) to provide an optimized solution. Both sampling methods make use of KV caching. A Jupyter notebook and some sample executions are provided. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2311_04951 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Leveraging Speculative Sampling and KV-Cache Optimizations Together for Generative AI using OpenVINO Barad, Haim Aidova, Ekaterina Gorbachev, Yury Machine Learning Artificial Intelligence Performance Inference optimizations are critical for improving user experience and reducing infrastructure costs and power consumption. In this article, we illustrate a form of dynamic execution known as speculative sampling to reduce the overall latency of text generation and compare it with standard autoregressive sampling. This can be used together with model-based optimizations (e.g. quantization) to provide an optimized solution. Both sampling methods make use of KV caching. A Jupyter notebook and some sample executions are provided. |
| title | Leveraging Speculative Sampling and KV-Cache Optimizations Together for Generative AI using OpenVINO |
| topic | Machine Learning Artificial Intelligence Performance |
| url | https://arxiv.org/abs/2311.04951 |