Optimization error: layer does not support shift delta

General
,
1

I’m trying to convert and optimize a model with attention (EfficientVit) and the conversion process fails with the following error:

hailo_model_optimization.acceleras.utils.acceleras_exceptions.AccelerasUnsupportedError: layer efficientvit_b2_scut-mfnet/matmul1 does not support shift delta. To overcome this issue you should force larger range at the inputs of the layer using command quantization_param([layer_name], force_range_in=[range_min, range_max], force_range_index=index) current range of input 0 is [-20.042, 14.947] and input 1 is [-2616.039, 10.981].You should increase the multiplication of these ranges by a factor of 1.494, e.g. you can apply factor of sqrt(1.494) to both inputs:
quantization_param([efficientvit_b2_scut-mfnet/matmul1], force_range_in=[-24.497, 18.270], force_range_index=0)
quantization_param([efficientvit_b2_scut-mfnet/matmul1], force_range_in=[-3197.566, 13.422], force_range_index=1)

I followed the advice and added the quantization parameters to the model script. I took care to avoid the character escaping issue caused by Pydantic mentioned in other topics:

model_script = [
    "quantization_param(['efficientvit_b2_scut-mfnet/matmul1'], force_range_in=[-24.497, 18.270], force_range_index=0)",
    "quantization_param(['efficientvit_b2_scut-mfnet/matmul1'], force_range_in=[-3197.566, 13.422], force_range_index=1)",
]

Unfortunately, this results in another error during model script loading:

hailo_sdk_client.runner.exceptions.InvalidArgumentsException: either model script is illegal or file path doesn't exist: Model script parsing failed: Invalid scope name 'efficientvit_b2_scut-mfnet exists. Model script file not found in location: input_norm = normalization([123.675, 116.28, 103.53], [58.395, 57.120000000000005, 57.375])
model_optimization_flavor(optimization_level=1, compression_level=0, batch_size=1)
performance_param(compiler_optimization_level=1)
quantization_param('efficientvit_b2_scut-mfnet/matmul1', force_range_in=[-24.497, 18.270], force_range_index=0)
quantization_param('efficientvit_b2_scut-mfnet/matmul1', force_range_in=[-3197.566, 13.422], force_range_index=1).

Here is the full model script (lines separeted with “\n”):

input_norm = normalization([123.675, 116.28, 103.53], [58.395, 57.120000000000005, 57.375])
model_optimization_flavor(optimization_level=1, compression_level=0, batch_size=1)
performance_param(compiler_optimization_level=1)
quantization_param('efficientvit_b2_scut-mfnet/matmul1', force_range_in=[-24.497, 18.270], force_range_index=0)
quantization_param('efficientvit_b2_scut-mfnet/matmul1', force_range_in=[-3197.566, 13.422], force_range_index=1)