File size: 419 Bytes
8c0a590 2e3897d 8c0a590 2e3897d |
1 2 3 4 5 6 7 8 |
---
license: mit
language:
- en
library_name: transformers
---
This is a QCQA version of the original model facebook/opt-125m. In this version, the original MHA architecture is preserved but instead of having a single K/V head, different K/V heads corresponding to the same group have the same mean-pooled K or V values. It has 16 groups of KV heads per layer instead of original 32 KV heads in the MHA implementation. |