8 bit quantized version using Intel's autoround. I had problems with 8 bit awq having weird outputs with tool calls, paths and variable names. This version work as expected.
This is the best option imo to run MiniMax-M2.7 at 8 bits (native precision) for Ampere cards, eg 3090's
- Downloads last month
- 5
Inference Providers NEW
This model isn't deployed by any Inference Provider. 馃檵 Ask for provider support
Model tree for bullerwins/MiniMax-M2.7-W8A16
Base model
MiniMaxAI/MiniMax-M2.7