J721EXCPXEVM: Dose ADAS/AD SOC of Jacinto7 family supports 32-bit inference of Deep Learning models?

I noticed that "Matrix Multiplier Accelerator (MMA) in ADAS/AD SOC of Jacinto7 family supports 8-bit, 16-bit and 32-bit inference of Deep Learning models." in the tidl doc .

But when I try to inference float model on the tda4 ,the error happened:

app_tidl: Init ... Done.
app_tidl: Creating graph ...
app_tidl: Creating graph ... Done.
app_tidl: Verifying graph ...
1714.391903 s: VX_ZONE_ERROR:[ownContextSendCmd:782] Command ack message returned failure cmd_status: -1
1714.391922 s: VX_ZONE_ERROR:[ownContextSendCmd:818] tivxEventWait() failed.
1714.391943 s: VX_ZONE_ERROR:[ownNodeKernelInit:526] Target kernel, TIVX_CMD_NODE_CREATE failed
1714.391959 s: VX_ZONE_ERROR:[ownNodeKernelInit:527] Please be sure the target callbacks have been registered for this core
1714.391975 s: VX_ZONE_ERROR:[ownNodeKernelInit:528] If the target callbacks have been registered, please ensure no errors are occurring within the create callback of this kernel
1714.391994 s: VX_ZONE_ERROR:[ownGraphNodeKernelInit:583] kernel init for node 0, kernel com.ti.tidl ... failed !!!
1714.392014 s: VX_ZONE_ERROR:[vxVerifyGraph:2044] Node kernel init failed
1714.392028 s: VX_ZONE_ERROR:[vxVerifyGraph:2098] Graph verify failed
app_tidl: ERROR: Verifying graph ... Failed !!!

     Dose ADAS/AD SOC of Jacinto7 family supports  32-bit inference of Deep Learning models?

C7x-MMA Core supports  8-bit, 16-bit and 32-bit computation.

TIDL SW currently support 8-bit and 16-bit inference only