onnxconverter-common
onnxconverter-common copied to clipboard
microsoft
Common utilities for ONNX converters
`requirements.txt` specifies unversioned dependencies on `numpy` and `protobuf`, both of which are (version constrained) dependencies of `onnx`, and so these are redundant and can be removed. For reference, the dependencies...
This forwards two arguments from the `convert_float_to_float16_model_path` wrapper function to `convert_float_to_float16_model_path` to complete the API. This functionality is needed to enable usage of the `model_path` API in groqflow.
Hi, I notice that convert_np_to_float16(np_array, min_positive_val=1e-7, max_finite_val=1e4) has 1e4 as the default value for max_finite_val, however the max value for float16 is 65504. Onnxruntime repo also has 65504 for their...
Hello, Would it be possible to make 1.16 available in pypi? I'm mostly interested in the relaxing of the protobuf dependency. Thanks 🙏
I successfully converted my model to fp16: ``` import onnx from onnxconverter_common import float16 model = onnx.load("tts_quantized.onnx") model_fp16 = float16.convert_float_to_float16(model) onnx.save(model_fp16, "tts_quantized_fp16.onnx") ``` But when I load fp16 model, I...
```[tasklist] ### Tasks ```
According to https://onnx.ai/onnx/operators/onnx__Resize.html#resize its input `scale` is force to float32. However, `convert_float_to_float16` doesn't handle it.
According to the [National Vulnerability Database](https://nvd.nist.gov/vuln/detail/CVE-2024-7254), the pinning of protobuf `protobuf==3.20.2` in the latest 1.14 release leads to a major vulnerability. Since this pinning seems to be fixed on master,...
