aeonmind/DeepSeek-V4-Flash-UQFF-qtip2

Correct the model card: working overlay invocation, public, measured GSM8K 96.0%

Add model card generated by the bake

DeepSeek-V4-Flash UQFF qtip2 (beam W=256, hadamard-128, 43 layers)

remove probe

write probe

initial commit