metadata
base_model:
- mistralai/Mistral-Small-3.2-24B-Instruct-2506
Trained from anthracite-core/Mistral-Small-3.2-24B-Instruct-2506-ChatML for convenience (no Pixtral compatibility needed). No vision adapter currently.
Test model trained at 16k context on 50M tokens of long-form human writing (mostly books).
Haven't tested yet but regular Tekken v7 instruct will work and samplers are probably the same as you'd use for 3.2 Instruct.