metadata
license: mit
base_model:
- inclusionAI/Ring-mini-2.0
pipeline_tag: text-generation
library_name: transformers
Pristine-Mini-8B-A1B-Base (8.6B A1.4B)
Pristine is a model series / dataset based on a pruned version of Ring-Mini-2.0 (16B A1.4B --> 8.6B A1.4B). The main motivation behind this project is that Granite 4.0 7B A1B has been pretty disappointing and with a high-performance 8B-A1B these small MoEs can be avenged.
Note: This model hasn't been further trained. Do not use it as is or create quantizations of it!