pytorch model is transformed into coreml model expect a tensor of type integer, but actually receive a tensor of type floating point.

[yuqilol]

When I was converting pytorch model into coreml model, an error occurred, indicating that the coreml model required int32 parameters but input fp32 parameters. I checked the previous error report and tried to modify it, but it seemed to be different from my error feeling. How should I improve my code to match the result? Prediction Mismatch between Pytorch and CoreML #1428 Wrong output from PyTorch converted model #1486 Model converts fine to neuralnetwork but produces all nan values for mlprogram #1809

python: 3.10.14 pytorch: 2.2.2 numpy: 1.26.4 coremltools: 7.1 Ubuntu 18.04.6

my code

import torch import time import torch import coremltools as ct from torchvision import transforms from models.yowo.yowo import YOWO from config.yowo_v2_config import yowo_v2_config from coremltools.converters.mil.mil import types import numpy as np yowo_v2_config["yowo_v2_tiny"] path_to_ckpt = '/data1/home/pengyuling/yowov2/weights/yowo_v2_tiny_epoch_986.pth' device = "cpu" transform = transforms.Compose([ transforms.Resize([224,224]), transforms.ToTensor(), ]) model = YOWO( cfg = yowo_v2_config["yowo_v2_tiny"], device = device, num_classes = 8, conf_thresh = 0., nms_thresh = 0.2, topk = 3, trainable = False, multi_hot = True ) device = torch.device(device) checkpoint = torch.load(path_to_ckpt, map_location='cpu') checkpoint_state_dict = checkpoint.pop("model") model_state_dict = model.state_dict() for k in list(checkpoint_state_dict.keys()): if k in model_state_dict: shape_model = tuple(model_state_dict[k].shape) shape_checkpoint = tuple(checkpoint_state_dict[k].shape) if shape_model != shape_checkpoint: checkpoint_state_dict.pop(k) else: checkpoint_state_dict.pop(k) print(k) model.load_state_dict(checkpoint_state_dict) model.eval() x = torch.rand(1, 3, 16, 224, 224) traced_model = torch.jit.trace(model, x) scripted_model = torch.jit.script(traced_model) mlmodel = ct.convert( scripted_model, source="pytorch", convert_to="mlprogram", inputs=[ct.TensorType(shape=x.shape)] ) mlmodel.save('yowo_v2_tiny_epoch_986.mlpackage')

my error

Tuple detected at graph output. This will be flattened in the converted model. Converting PyTorch Frontend ==> MIL Ops: 66%|███████████████████████████████████▍ | 1717/2613 [00:03<00:01, 473.30 ops/s]Saving value type of int64 into a builtin type of int32, might lose precision! Saving value type of int64 into a builtin type of int32, might lose precision! Converting PyTorch Frontend ==> MIL Ops: 76%|█████████████████████████████████████████▏ | 1992/2613 [00:03<00:01, 592.53 ops/s]Saving value type of int64 into a builtin type of int32, might lose precision! Saving value type of int64 into a builtin type of int32, might lose precision! Converting PyTorch Frontend ==> MIL Ops: 87%|███████████████████████████████████████████████▏ | 2281/2613 [00:04<00:00, 585.92 ops/s]Saving value type of int64 into a builtin type of int32, might lose precision! Saving value type of int64 into a builtin type of int32, might lose precision! Converting PyTorch Frontend ==> MIL Ops: 93%|██████████████████████████████████████████████████▍ | 2439/2613 [00:04<00:00, 582.65 ops/s] Traceback (most recent call last): File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/runpy.py", line 196, in _run_module_as_main return _run_code(code, main_globals, None, File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/runpy.py", line 86, in _run_code exec(code, run_globals) File "/data1/home/pengyuling/.vscode-server/extensions/ms-python.debugpy-2024.0.0/bundled/libs/debugpy/adapter/../../debugpy/launcher/../../debugpy/main.py", line 39, in cli.main() File "/data1/home/pengyuling/.vscode-server/extensions/ms-python.debugpy-2024.0.0/bundled/libs/debugpy/adapter/../../debugpy/launcher/../../debugpy/../debugpy/server/cli.py", line 430, in main run() File "/data1/home/pengyuling/.vscode-server/extensions/ms-python.debugpy-2024.0.0/bundled/libs/debugpy/adapter/../../debugpy/launcher/../../debugpy/../debugpy/server/cli.py", line 284, in run_file runpy.run_path(target, run_name="main") File "/data1/home/pengyuling/.vscode-server/extensions/ms-python.debugpy-2024.0.0/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py", line 321, in run_path return _run_module_code(code, init_globals, run_name, File "/data1/home/pengyuling/.vscode-server/extensions/ms-python.debugpy-2024.0.0/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py", line 135, in _run_module_code _run_code(code, mod_globals, init_globals, File "/data1/home/pengyuling/.vscode-server/extensions/ms-python.debugpy-2024.0.0/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py", line 124, in _run_code exec(code, run_globals) File "/data1/home/pengyuling/yowov2/export_coreml.py", line 53, in mlmodel = ct.convert( File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/_converters_entry.py", line 574, in convert mlmodel = mil_convert( File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/mil/converter.py", line 188, in mil_convert return _mil_convert(model, convert_from, convert_to, ConverterRegistry, MLModel, compute_units, **kwargs) File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/mil/converter.py", line 212, in _mil_convert proto, mil_program = mil_convert_to_proto( File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/mil/converter.py", line 286, in mil_convert_to_proto prog = frontend_converter(model, **kwargs) File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/mil/converter.py", line 108, in call return load(*args, **kwargs) File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/mil/frontend/torch/load.py", line 80, in load return _perform_torch_convert(converter, debug) File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/mil/frontend/torch/load.py", line 99, in _perform_torch_convert prog = converter.convert() File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/mil/frontend/torch/converter.py", line 519, in convert convert_nodes(self.context, self.graph) File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/mil/frontend/torch/ops.py", line 88, in convert_nodes add_op(context, node) File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/mil/frontend/torch/ops.py", line 3789, in index x = mb.gather(x=x, indices=indices, axis=axis, name=node.name) File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/mil/mil/ops/registry.py", line 182, in add_op return cls._add_op(op_cls_to_add, **kwargs) File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/mil/mil/builder.py", line 168, in _add_op new_op = op_cls(**kwargs) File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/mil/mil/operation.py", line 190, in init self._validate_and_set_inputs(input_kv) File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/mil/mil/operation.py", line 503, in _validate_and_set_inputs self.input_spec.validate_inputs(self.name, self.op_type, input_kvs) File "/data1/home/pengyuling/.conda/envs/yowo2/lib/python3.10/site-packages/coremltools/converters/mil/mil/input_type.py", line 163, in validate_inputs raise ValueError(msg.format(name, var.name, input_type.type_str, ValueError: Op "pred_reg.1" (op_type: gather) Input indices="anchor_idxs.1" expects tensor or scalar of dtype from type domain ['int32'] but got tensor[is1,fp32]

[TobyRoseman]

I would try updating the coremltools code to cast the indices to a dtype of int32.

Try adding the following line: indices = mb.cast(x=indices, dtype="int32") To coremltools/converters/mil/frontend/torch/ops.py in the index method right before the following line: x = mb.gather(x=x, indices=indices, axis=axis, name=node.name)