A.X-Encoder-base — Subnota ONNX int8 conversion

This is an unofficial conversion, not a model created by Subnota or an official SK Telecom release. The original model was developed by SK Telecom (SKT AI Model Lab): skt/A.X-Encoder-base, revision 9708f9c404ace91efd25c06fac2d73413616f4ef.

Subnota converted the original PyTorch masked-language model to ONNX (opset 17) and dynamically quantized MatMul and Gather weights to signed int8. No additional training or fine-tuning was performed. The conversion script is published with the Subnota desktop source as desktop/scripts/export-ax-encoder.py.

The onnx/model_quantized.onnx file is 188,885,598 bytes and has SHA-256 fb18f550c6fd5194819a6c5fd8e1e318bc6524019543803b07fc4b5f131ee9c0. The tokenizer and configuration files are copied unchanged from the original revision. The original weights remain available from SKT's repository.

The original model is licensed under Apache License 2.0. A copy of that license is included in this repository as LICENSE. This conversion is distributed under the same license. SK Telecom has not endorsed this conversion or Subnota.

Downloads last month: -

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Hoon03/subnota-ax-encoder-int8-onnx

Base model

skt/A.X-Encoder-base

Quantized

(1)

this model