Quote:Here it does not happen mainly when using drag&drop, but if that is not the case for you the culprit must be somewhere else.

Quote:    if input_format != output_format:
        raise ValueError("input format must be equal to output format")

Quote:    output_format = input_format
    if input_format != output_format:
        raise ValueError("input format must be equal to output format")

Quote:2026-06-22 19:37:46.556
Plugin D:/APPD/Hybrid_dev_20260621-184241/64bit/vs-mlrt/vstrt.dll is using API3 which is deprecated and will be removed shortly.
Plugin D:/APPD/Hybrid_dev_20260621-184241/64bit/vsfilters/SourceFilter/LSmashSource/LSMASHSource.dll is using API3 which is deprecated and will be removed shortly.
2026-06-22 19:37:47.149
Failed to evaluate the script:
Python exception: trtexec execution fails, log has been written to C:\Users\mmddffkk\AppData\Local\Temp\trtexec_260622_193746.log

Traceback (most recent call last):
File "vapoursynth.pyx", line 3623, in vapoursynth._vpy_evaluate
File "vapoursynth.pyx", line 3624, in vapoursynth._vpy_evaluate
File "D:\APPD\HybridFiles\Temp\tempPreviewVapoursynthFile19_37_46_002.vpy", line 52, in
clip = vsmlrt.DPIR(clip, strength=5.000, overlap=16, model=1, backend=Backend.TRT(fp16=False,device_id=0,bf16=False,verbose=True,use_cuda_graph=False,num_streams=1,builder_optimization_level=3,engine_folder="D:/APPD/HybridFiles/Engine"))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:/APPD/Hybrid_dev_20260621-184241/64bit/vs-mlrt/vsmlrt.py", line 595, in DPIR
clip = inference_with_fallback(
^^^^^^^^^^^^^^^^^^^^^^^^
File "D:/APPD/Hybrid_dev_20260621-184241/64bit/vs-mlrt/vsmlrt.py", line 3056, in inference_with_fallback
raise e
File "D:/APPD/Hybrid_dev_20260621-184241/64bit/vs-mlrt/vsmlrt.py", line 3033, in inference_with_fallback
ret = _inference(
^^^^^^^^^^^
File "D:/APPD/Hybrid_dev_20260621-184241/64bit/vs-mlrt/vsmlrt.py", line 2850, in _inference
engine_path = trtexec(
^^^^^^^^
File "D:/APPD/Hybrid_dev_20260621-184241/64bit/vs-mlrt/vsmlrt.py", line 2194, in trtexec
raise RuntimeError(f"trtexec execution fails, log has been written to {log_filename}")
RuntimeError: trtexec execution fails, log has been written to C:\Users\mmddffkk\AppData\Local\Temp\trtexec_260622_193746.log

Quote:&&&& RUNNING TensorRT.trtexec [TensorRT v110000] [b114] # D:/APPD/Hybrid_dev_20260621-184241/64bit/vs-mlrt\vsmlrt-cuda\trtexec --onnx=D:/APPD/Hybrid_dev_20260621-184241/64bit/vs-mlrt\models\dpir\drunet_color.onnx --timingCacheFile=D:/APPD/HybridFiles/Engine\469697a4.engine.cache --device=0 --saveEngine=D:/APPD/HybridFiles/Engine\469697a4.engine --shapes=input:1x4x816x1920 --verbose --tacticSources=-CUBLAS,-CUBLAS_LT,-CUDNN,+EDGE_MASK_CONVOLUTIONS,+JIT_CONVOLUTIONS --skipInference --noTF32 --inputIOFormats=fp32:chw --outputIOFormats=fp32:chw --builderOptimizationLevel=3 --precisionConstraints=obey --layerPrecisions=Conv_123:fp32
=== Model Options ===
  --onnx=<file>              ONNX model

=== Build Options ===
  --minShapes=spec                  Build with dynamic shapes using a profile with the min shapes provided
  --optShapes=spec                  Build with dynamic shapes using a profile with the opt shapes provided
  --maxShapes=spec                  Build with dynamic shapes using a profile with the max shapes provided
  --inputIOFormats=spec              Memory layout of each of the input tensors (default = chw)
                                    See --outputIOFormats help for the grammar of format list.
                                    Note: If this option is specified, please set comma-separated formats for all
                                          inputs following the same order as network inputs ID (even if only one input
                                          needs specifying IO format) or set the format once for broadcasting.
  --outputIOFormats=spec            Memory layout of each of the output tensors (default = chw)
                                    Note: If this option is specified, please set comma-separated formats for all
                                          outputs following the same order as network outputs ID (even if only one output
                                          needs specifying IO format) or set the format once for broadcasting.
                                    IO Formats: spec  ::= IOfmt[","spec]
                                                IOfmt ::= ("chw"|"chw2"|"hwc8"|"chw4"|"chw16"|"chw32"|"dhwc8"|
                                                            "cdhw32"|"hwc"|"dla_linear"|"dla_hwc4"|"hwc16"|"dhwc")["+"IOfmt]
  --memPoolSize=poolspec            Specify the size constraints of the designated memory pool(s)
                                    Supports the following base-2 suffixes: B (Bytes), G (Gibibytes), K (Kibibytes), M (Mebibytes).
                                    If none of suffixes is appended, the defualt unit is in MiB.
                                    Note: Also accepts decimal sizes, e.g. 0.25M. Will be rounded down to the nearest integer bytes.
                                    In particular, for dlaSRAM the bytes will be rounded down to the nearest power of 2.
                                    Pool constraint: poolspec ::= poolfmt[","poolspec]
                                                      poolfmt ::= poolize
                                                      pool ::= "workspace"|"dlaSRAM"|"dlaLocalDRAM"|"dlaGlobalDRAM"|"tacticSharedMem"
  --profilingVerbosity=mode          Specify profiling verbosity. mode ::= layer_names_only|detailed|none (default = layer_names_only).
                                    Please only assign once.
  --avgTiming=M                      Set the number of times averaged in each iteration for kernel selection (default = 8)
  --refit                            Mark the engine as refittable. This will allow the inspection of refittable layers
                                    and weights within the engine.
  --stripWeights                    Strip weights from plan. This flag works with either refit or refit with identical weights. Default
                                    to latter, but you can switch to the former by enabling both --stripWeights and --refit at the same
                                    time.
  --stripAllWeights                  Alias for combining the --refit and --stripWeights options. It marks all weights as refittable,
                                    disregarding any performance impact. Additionally, it strips all refittable weights after the
                                    engine is built.
  --versionCompatible, --vc          Mark the engine as version compatible. This allows the engine to be used with newer versions
                                    of TensorRT on the same host OS, as well as TensorRT's dispatch and lean runtimes.
  --pluginInstanceNorm, --pi        Set `kNATIVE_INSTANCENORM` to false in the ONNX parser. This will cause the ONNX parser to use
                                    a plugin InstanceNorm implementation over the native implementation when parsing.
  --uint8AsymmetricQuantizationDLA  Set `kENABLE_UINT8_AND_ASYMMETRIC_QUANTIZATION_DLA` to true in the ONNX parser. This directs the
                                    ONNX parser to allow UINT8 as a quantization data type and import zero point values directly
                                    without converting to float type or all-zero values. Should only be set with DLA software version
                                    >= 3.16.
  --reportCapabilityDLA              Set `kREPORT_CAPABILITY_DLA` to true in the ONNX parser. This signals the ONNX parser to validate
                                    that all nodes in the model can run on DLA. This flag is set to be OFF by default.
  --adjustForDLA                    Set `kADJUST_FOR_DLA` to true in the ONNX parser. This signals the ONNX parser to opportunistically
                                    rewrite or modify layers to make them more amenable to running on DLA. This flag is set to be OFF
                                    by default.
  --enablePluginOverride            Set `kENABLE_PLUGIN_OVERRIDE` to true in the ONNX parser. This allows the ONNX parser to use
                                    a plugin implementation over the standard ONNX operator implementation when parsing.
  --useRuntime=runtime              TensorRT runtime to execute engine. "lean" and "dispatch" require loading VC engine and do
                                    not support building an engine.
                                        runtime::= "full"|"lean"|"dispatch"
  --leanDLLPath=<file>              External lean runtime DLL to use in version compatible mode.
  --excludeLeanRuntime              When --versionCompatible is enabled, this flag indicates that the generated engine should
                                    not include an embedded lean runtime. If this is set, the user must explicitly specify a
                                    valid lean runtime to use when loading the engine.
  --monitorMemory                    Enable memory monitor report for debugging usage. (default = disabled)
                                    Disables CUDA timing cache and profile streams. Only allowed when building
                                    a safe engine (--safe) with remote auto-tuning (--remoteAutoTuningConfig).
                                    (default = disabled)
  --sparsity=spec                    Control sparsity (default = disabled).
                                    Sparsity: spec ::= "disable", "enable", "force"
                                    Note: Description about each of these options is as below
                                          disable = do not enable sparse tactics in the builder (this is the default)
                                          enable  = enable sparse tactics in the builder (but these tactics will only be
                                                    considered if the weights have the right sparsity pattern)
                                          force  = enable sparse tactics in the builder and force-overwrite the weights to have
                                                    a sparsity pattern (even if you loaded a model yourself)
                                                    [Deprecated] this knob has been deprecated.
                                                    Please use <polygraphy surgeon prune> to rewrite the weights.
  --noTF32                          Disable tf32 precision (default is to enable tf32, in addition to fp32)
  --stronglyTyped                    [Deprecated] Strongly typed network is now enabled by default. This flag is a no-op.
  --directIO                        [Deprecated] Avoid reformatting at network boundaries. (default = disabled)
  --layerDeviceTypes=spec            Specify layer-specific device type.
                                    The specs are read left-to-right, and later ones override earlier ones. If a layer does not have
                                    a device type specified, the layer will opt for the default device type.
                                    Per-layer device type spec ::= layerDeviceTypePair[","spec]
                                                          layerDeviceTypePair ::= layerName":"deviceType
                                                          deviceType ::= "GPU"|"DLA"
  --decomposableAttentions=spec      Specify decomposable attentions by comma-separated names.
                                    The specs are read left-to-right, and later ones override earlier ones. Each layer name can
                                    contain at most one wildcard ('*') character.
  --safe                            Enable build safety certified engine.
                                    If DLA is enabled, --buildDLAStandalone will be specified
  --dumpKernelText                  Dump the kernel text to a file, only available when --safe is enabled
  --buildDLAStandalone              Enable build DLA standalone loadable which can be loaded by cuDLA, when this option is enabled,
                                    --allowGPUFallback is disallowed and --skipInference is enabled by default. Additionally,
                                    specifying --inputIOFormats and --outputIOFormats restricts memory layout
                                    (default = disabled)
  --allowGPUFallback                When DLA is enabled, allow GPU fallback for unsupported layers (default = disabled)
  --consistency                      Perform consistency checking on safety certified engine
  --saveEngine=<file>                Save the serialized engine
  --loadEngine=<file>                Load a serialized engine
  --asyncFileReader                  Load a serialized engine using async stream reader. Should be combined with --loadEngine.
  --getPlanVersionOnly              Print TensorRT version when loaded plan was created. Works without deserialization of the plan.
                                    Use together with --loadEngine. Supported only for engines created with 8.6 and forward.
  --tacticSources=tactics            Specify the tactics to be used by adding (+) or removing (-) tactics from the default
                                    tactic sources (default = all available tactics).
                                    Note: Currently only edge mask convolutions and JIT convolutions are listed as optional
                                          tactics.
                                    Tactic Sources: tactics ::= tactic[","tactics]
                                                    tactic  ::= (+|-)lib
                                                    lib    ::= "EDGE_MASK_CONVOLUTIONS"|"JIT_CONVOLUTIONS"
                                    For example, to disable edge mask convolutions: --tacticSources=-EDGE_MASK_CONVOLUTIONS
  --noBuilderCache                  Disable timing cache in builder (default is to enable timing cache)
  --noCompilationCache              Disable Compilation cache in builder, and the cache is part of timing cache (default is to enable compilation cache)
  --errorOnTimingCacheMiss          Emit error when a tactic being timed is not present in the timing cache (default = false)
  --timingCacheFile=<file>          Save/load the serialized global timing cache
  --preview=features                Specify preview feature to be used by adding (+) or removing (-) preview features from the default
                                    Preview Features: features ::= feature[","features]
                                                      feature  ::= (+|-)flag
                                                      flag    ::= "aliasedPluginIO1003"
                                                                    |"runtimeActivationResize"
  --builderOptimizationLevel        Set the builder optimization level. (default is 3)
                                    A Higher level allows TensorRT to spend more time searching for better optimization strategy.
                                    Valid values include integers from 0 to the maximum optimization level, which is currently 5.
  --maxTactics                      Set the maximum number of tactics to time when there is a choice of tactics. (default is -1)
                                    Larger number of tactics allow TensorRT to spend more building time on evaluating tactics.
                                    Default value -1 means TensorRT can decide the number of tactics based on its own heuristic.
  --hardwareCompatibilityLevel=mode  Make the engine file compatible with other GPU architectures. (default = none)
                                    Hardware Compatibility Level: mode ::= "none" | "ampere+" | "sameComputeCapability"
                                        none = no compatibility
                                        ampere+ = compatible with Ampere and newer GPUs
                                        sameComputeCapability = compatible with GPUs that have the same Compute Capability version
  --runtimePlatform=platform        Set the target platform for runtime execution. (default = SameAsBuild)
                                    When this option is enabled, --skipInference is enabled by default.
                                    RuntimePlatfrom: platform ::= "SameAsBuild" | "WindowsAMD64"
                                        SameAsBuild = no requirement for cross-platform compatibility.
                                        WindowsAMD64 = set the target platform for engine execution as Windows AMD64 system
  --tempdir=<dir>                    Overrides the default temporary directory TensorRT will use when creating temporary files.
                                    See IRuntime:etTemporaryDirectory API documentation for more information.
  --tempfileControls=controls        Controls what TensorRT is allowed to use when creating temporary executable files.
                                    Should be a comma-separated list with entries in the format (in_memory|temporary)allow|deny).
                                    in_memory: Controls whether TensorRT is allowed to create temporary in-memory executable files.
                                    temporary: Controls whether TensorRT is allowed to create temporary executable files in the
                                                filesystem (in the directory given by --tempdir).
                                    For example, to allow in-memory files and disallow temporary files:
                                        --tempfileControls=in_memory:allow,temporary:deny
                                    If a flag is unspecified, the default behavior is "allow".
  --maxAuxStreams=N                  Set maximum number of auxiliary streams per inference stream that TRT is allowed to use to run
                                    kernels in parallel if the network contains ops that can run in parallel, with the cost of more
                                    memory usage. Set this to 0 for optimal memory usage. (default = using heuristics)
  --profile                          Build with dynamic shapes using a profile with the min/max/opt shapes provided. Can be specified
                                        multiple times to create multiple profiles with contiguous index.
                                    (ex: --profile=0 --minShapes=<spec> --optShapes=<spec> --maxShapes=<spec> --profile=1 ...)
  --allowWeightStreaming            Enable a weight streaming engine. TensorRT will disable
                                    weight streaming at runtime unless --weightStreamingBudget is specified.
  --markDebug                        Specify list of names of tensors to be marked as debug tensors. Separate names with a comma
  --markUnfusedTensorsAsDebugTensors Mark unfused tensors as debug tensors
  --tilingOptimizationLevel          Set the tiling optimization level. (default is 0)
                                    A Higher level allows TensorRT to spend more time searching for better optimization strategy.
                                    Valid values include integers from 0 to the maximum tiling optimization level(3).
  --l2LimitForTiling                Set the L2 cache usage limit for tiling optimization(default is -1)
  --remoteAutoTuningConfig          Set the remote auto tuning config. Must be specified with --safe.
                                    Format: protocol://username[:password]@hostname[:port]?param1=value1&param2=value2
                                    Example: ssh://user:pass@192.0.2.100:22?remote_exec_path=/opt/tensorrt/bin&remote_lib_path=/opt/tensorrt/lib
  --refitFromOnnx                    Refit the loaded engine with the weights from the provided ONNX model.
                                    The model should be identical to the one used to generate the engine.
  --cpuOnly                          Build the engine with CPU-only mode. No local GPU is required on the build machine.
                                    Must be specified with --remoteAutoTuningConfig and --safe flags.

=== Inference Options ===
  --shapes=spec              Set input shapes for dynamic shapes inference inputs.
                              Note: Input names can be wrapped with escaped single quotes (ex: 'Input:0').
                              Example input shapes spec: input0:1x3x256x256, input1:1x3x128x128
                              For scalars (0-D shapes), use input0calar or simply input0: with nothing after the colon.
                              Each input shape is supplied as a key-value pair where key is the input name and
                              value is the dimensions (including the batch dimension) to be used for that input.
                              Each key-value pair has the key and value separated using a colon (.
                              Multiple input shapes can be provided via comma-separated key-value pairs, and each input
                              name can contain at most one wildcard ('*') character.
  --loadInputs=spec          Load input values from files (default = generate random inputs). Input names can be wrapped with single quotes (ex: 'Input:0')
                              Input values spec ::= Ival[","spec]
                                          Ival ::= name":"file
                              Consult the README for more information on generating files for custom inputs.
  --iterations=N              Run at least N inference iterations (default = 10)
  --warmUp=N                  Run for N milliseconds to warmup before measuring performance (default = 200)
  --duration=N                Run performance measurements for at least N seconds wallclock time (default = 3)
                              If -1 is specified, inference will keep running unless stopped manually
  --sleepTime=N              Delay inference start with a gap of N milliseconds between launch and compute (default = 0)
  --idleTime=N                Sleep N milliseconds between two continuous iterations(default = 0)
  --infStreams=N              Instantiate N execution contexts to run inference concurrently (default = 1)
  --exposeDMA                Serialize DMA transfers to and from device (default = disabled).
  --includeDataTransfers      Enable DMA transfers to and from device (default = disabled). Note some device-to-host
                              data transfers will remain if output dumping is enabled via the --dumpOutput or
                              --exportOutput flags.
  --noDataTransfers          [Deprecated] DMA transfers are now disabled by default. This flag is a no-op.
  --useManagedMemory          Use managed memory instead of separate host and device allocations (default = disabled).
  --noSpinWait                Disable spin wait and use blocking synchronization instead. This may reduce CPU usage but increase synchronization time (default = spin wait enabled)
  --useSpinWait              [Deprecated] Spin wait is now enabled by default. This flag is a no-op.
  --threads                  Enable multithreading to drive engines with independent threads or speed up refitting (default = disabled)
  --noCudaGraph              Disable CUDA graph capture and launch (default = CUDA graph enabled).
  --useCudaGraph              [Deprecated] CUDA graph is now enabled by default. This flag is a no-op.
  --timeDeserialize          Time the amount of time it takes to deserialize the network and exit.
  --timeRefit                Time the amount of time it takes to refit the engine before inference.
  --separateProfileRun        [Deprecated] Separate profile run is now always enabled. This flag is a no-op.
  --skipInference            Exit after the engine has been built and skip inference perf measurement (default = disabled)
  --persistentCacheRatio      Set the persistentCacheLimit in ratio, 0.5 represent half of max persistent L2 size (default = 0)
  --useProfile                Set the optimization profile for the inference context (default = 0 ).
  --allocationStrategy=spec  Specify how the internal device memory for inference is allocated.
                              Strategy: spec ::= "static"|"profile"|"runtime"
                                  static = Allocate device memory based on max size across all profiles.
                                  profile = Allocate device memory based on max size of the current profile.
                                  runtime = Allocate device memory based on the actual input shapes.
  --saveDebugTensors          Specify list of names of tensors to turn on the debug state
                              and filename to save raw outputs to.
                              These tensors must be specified as debug tensors during build time.
                              Input values spec ::= Ival[","spec]
                                          Ival ::= name":"file
  --saveAllDebugTensors      Save all debug tensors to files.
                              Including debug tensors marked by --markDebug and --markUnfusedTensorsAsDebugTensors
                              Multiple file formats can be saved simultaneously.
                              Input values spec  ::= format[","format]
                                          format ::= "summary"|"numpy"|"string"|"raw"
  --weightStreamingBudget    Set the maximum amount of GPU memory TensorRT is allowed to use for weights.
                              It can take on the following values:
                                  -2: (default) Disable weight streaming at runtime.
                                  -1: TensorRT will automatically decide the budget.
                                  0-100%: Percentage of streamable weights that reside on the GPU.
                                          0% saves the most memory but will have the worst performance.
                                          Requires the '%' character.
                                  >=0B: The exact amount of streamable weights that reside on the GPU. Supports the
                                      following base-2 suffixes: B (Bytes), G (Gibibytes), K (Kibibytes), M (Mebibytes).

=== Reporting Options ===
  --verbose                  Use verbose logging (default = false)
  --avgRuns=N                Report performance measurements averaged over N consecutive iterations (default = 10)
  --percentile=P1,P2,P3,...  Report performance for the P1,P2,P3,... percentages (0<=P_i<=100, 0 representing max perf, and 100 representing min perf; (default = 90,95,99%)
  --dumpRefit                Print the refittable layers and weights from a refittable engine
  --dumpOutput                Print the output tensor(s) of the last inference iteration (default = disabled)
  --dumpRawBindingsToFile    Print the input/output tensor(s) of the last inference iteration to file(default = disabled)
  --dumpProfile              Print profile information per layer (default = disabled)
  --dumpLayerInfo            Print layer information of the engine to console (default = disabled)
  --dumpOptimizationProfile  Print the optimization profile(s) information (default = disabled)
  --exportTimes=<file>        Write the timing results in a json file (default = disabled)
  --exportOutput=<file>      Write the output tensors to a json file (default = disabled)
  --exportProfile=<file>      Write the profile information per layer in a json file (default = disabled)
  --exportLayerInfo=<file>    Write the layer information of the engine in a json file (default = disabled)

=== System Options ===
  --device=N                  Select cuda device N (default = 0)
  --staticPlugins            Plugin library (.so) to load statically (can be specified multiple times)
  --useDLACore=N              Select DLA core N for layers that support DLA (default = none)
  --staticPlugins            Plugin library (.so) to load statically (can be specified multiple times)
  --dynamicPlugins            Plugin library (.so) to load dynamically and may be serialized with the engine if they are included in --setPluginsToSerialize (can be specified multiple times)
  --setPluginsToSerialize    Plugin library (.so) to be serialized with the engine (can be specified multiple times)
  --ignoreParsedPluginLibs    By default, when building a version-compatible engine, plugin libraries specified by the ONNX parser
                              are implicitly serialized with the engine (unless --excludeLeanRuntime is specified) and loaded dynamically.
                              Enable this flag to ignore these plugin libraries instead.
  --safetyPlugins            Plugin library (.so) for TensorRT auto safety to manually load safety plugins specified by the command line arguments.
                              Example: --safetyPlugins=/path/to/plugin_lib.so[pluginNamespace1::plugin1,pluginNamespace2::plugin2].
                              The option can be specified multiple times with different plugin libraries.

=== Help ===
  --help, -h                  Print this message
[06/22/2026-19:37:47] [E] Invalid TensorFormat fp32:chw
&&&& FAILED TensorRT.trtexec [TensorRT v110000] [b114] # D:/APPD/Hybrid_dev_20260621-184241/64bit/vs-mlrt\vsmlrt-cuda\trtexec --onnx=D:/APPD/Hybrid_dev_20260621-184241/64bit/vs-mlrt\models\dpir\drunet_color.onnx --timingCacheFile=D:/APPD/HybridFiles/Engine\469697a4.engine.cache --device=0 --saveEngine=D:/APPD/HybridFiles/Engine\469697a4.engine --shapes=input:1x4x816x1920 --verbose --tacticSources=-CUBLAS,-CUBLAS_LT,-CUDNN,+EDGE_MASK_CONVOLUTIONS,+JIT_CONVOLUTIONS --skipInference --noTF32 --inputIOFormats=fp32:chw --outputIOFormats=fp32:chw --builderOptimizationLevel=3 --precisionConstraints=obey --layerPrecisions=Conv_123:fp32