Transformer tricks: Precomputing the first layer
Fuente:
arXiv
Enregistré dans:
| Auteur principal: | |
|---|---|
| Format: | Preprint |
| Publié: |
2024
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866917611212636160 |
|---|---|
| author | Graef, Nils |
| author_facet | Graef, Nils |
| contents | This micro-paper describes a trick to speed up inference of transformers with RoPE (such as LLaMA, Mistral, PaLM, and Gemma). For these models, a large portion of the first transformer layer can be precomputed, which results in slightly lower latency and lower cost-per-token. Because this trick optimizes only one layer, the relative savings depend on the total number of layers. For example, the maximum savings for a model with only 4 layers (such as Whisper tiny) is limited to 25%, while a 32-layer model is limited to 3% savings. See https://github.com/OpenMachine-ai/transformer-tricks for code and more transformer tricks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2402_13388 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Transformer tricks: Precomputing the first layer Graef, Nils Machine Learning This micro-paper describes a trick to speed up inference of transformers with RoPE (such as LLaMA, Mistral, PaLM, and Gemma). For these models, a large portion of the first transformer layer can be precomputed, which results in slightly lower latency and lower cost-per-token. Because this trick optimizes only one layer, the relative savings depend on the total number of layers. For example, the maximum savings for a model with only 4 layers (such as Whisper tiny) is limited to 25%, while a 32-layer model is limited to 3% savings. See https://github.com/OpenMachine-ai/transformer-tricks for code and more transformer tricks. |
| title | Transformer tricks: Precomputing the first layer |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2402.13388 |