DeepSeek V3 architecture with 4 layers + 8 experts per MoE + MTP module + FP8 weights from original model without further tuning To be used in CI testing
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy