This is Mistral's Ministral 3 8B Instruct model with the vision encoder removed and converted to the Llama format. The model may be slightly different due to conversion but should be largely identical (besides the removal of the vision encoder).
Note: Mistral's Ministral 3 Instruct models are quantized by default, so this is an upcasted version of the model.
The main benefits of Llama-converted models:
Easier to train: compatible with all trainers that support the Llama model
Increased compatibility: works out-of-the-box with popular libraries, no need to install Transformers from source, works even on older versions
This model, like the original model, is licensed under the Apache 2.0 license.
