File size: 347 Bytes
e52b272 330826e e52b272 330826e | 1 2 3 4 5 6 7 8 9 | ---
license: apache-2.0
language:
- en
base_model:
- meta-llama/Llama-3.1-8B-Instruct
pipeline_tag: text-generation
---
We simply applied the methodology from the paper titled "TransMLA: MLA Is All You Need" to convert the attention mechanism within the Llama-3 model from Grouped Query Attention (GQA) to Multi-Head Latent Attention (MLA), Kinda. |