Katago Sabaki TensorRT will not run

[Zorro962]

trafficstars

I get the Sabaki error: GTP engine can't be synced to current state.

If I run a benchmark it seems to finish too quickly.

When I analyze a move in Sabaki, the log is as follows:

2023-05-14 16:23:20-0400: Running with following config: allowResignation = true cudaDeviceToUse = 0 lagBuffer = 1.0 logAllGTPCommunication = true logDir = gtp_logs logSearchInfo = true logToStderr = true maxTimePondering = 60.0 maxVisits = 500 nnCacheSizePowerOfTwo = 20 nnMutexPoolSizePowerOfTwo = 16 numNNServerThreadsPerModel = 1 numSearchThreads = 12 ponderingEnabled = false resignConsecTurns = 3 resignThreshold = -0.90 rules = japanese searchFactorAfterOnePass = 0.50 searchFactorAfterTwoPass = 0.25 searchFactorWhenWinning = 0.40 searchFactorWhenWinningThreshold = 0.95 trtDeviceToUse = 0

2023-05-14 16:23:20-0400: Loading model and initializing benchmark... 2023-05-14 16:23:20-0400: Testing with default positions for board size: 19 2023-05-14 16:23:20-0400: nnRandSeed0 = 14189492281794093621 2023-05-14 16:23:20-0400: After dedups: nnModelFile0 = C:\Program Files\KataGo/default_model.bin.gz useFP16 auto useNHWC auto 2023-05-14 16:23:20-0400: Initializing neural net buffer to be size 19 * 19 exactly 2023-05-14 16:23:23-0400: TensorRT backend thread 0: Found GPU NVIDIA GeForce RTX 2070 Super memory 8589606912 compute capability major 7 minor 5 2023-05-14 16:23:23-0400: TensorRT backend thread 0: Initializing (may take a long time)

[tatianyi]

I didn't find a reasonable cause for this problem, this issue showed up after Katago switched from the trrt8.2 to the trrt8.5 version. However I git pull the code and compiled it myself to use the latest SDKs in my computer, and then It worked.