CoderBak/editlens_roberta_modelkit
0267
1EditLens RoBERTa Model Kit — CoderBak2 3This repository redistributes the original Pangram EditLens RoBERTa-large4checkpoint and converted ONNX artifacts derived from that checkpoint.5It is a community conversion/distribution project, not a newly trained model.6No affiliation with or endorsement by Pangram, the EditLens authors, Meta,7Hugging Face, or Microsoft is claimed.8 9Original model developer: Pangram.10Original research authors: Katherine Thai, Bradley Emi, Elyas Masrour,11and Mohit Iyyer.12Paper: EditLens: Quantifying the Extent of AI Editing in Text.13https://arxiv.org/abs/2510.0315414Upstream model: https://huggingface.co/pangram/editlens_roberta-large15Pinned source revision: f93e1ace74528cfb48f337ab2fe946fb71a728cb16Upstream research code: https://github.com/pangramlabs/EditLens17Base-model lineage: https://huggingface.co/FacebookAI/roberta-large18The base-model card identifies its license as MIT. The EditLens derivative19weights distributed here retain Pangram's CC BY-NC-SA 4.0 license.20 21License: Creative Commons Attribution-NonCommercial-ShareAlike 4.022International (CC BY-NC-SA 4.0).23https://creativecommons.org/licenses/by-nc-sa/4.0/24The complete license text is provided in LICENSE. Its attribution,25noncommercial, share-alike, and other applicable terms continue to apply.26Public availability does not grant commercial-use rights.27No additional research-only restriction is imposed by this model kit.28Third-party software dependencies retain their own licenses.29 30Changes made by CoderBak:31- Exported the complete sequence classifier to ONNX with opset 17,32 dynamic batch/sequence axes, int64 input IDs and attention masks,33 and four output logits.34- Created optional FP16 and dynamic per-channel INT8 MatMul variants.35- Added conversion scripts, local inference example, provenance,36 checksums, and numerical conversion checks.37- No retraining, distillation, new calibration, or change of label order.38- The root PyTorch checkpoint/configuration/tokenizer files are preserved39 byte-for-byte from the pinned upstream snapshot.40 41The original model card is retained in upstream/README.md. Its access-form42metadata is historical upstream information, not a gate for this repository.43This repository is public and ungated; that does not waive the model license44or grant access to the separately gated upstream repository.45 46The license includes a disclaimer of warranties and limitation of liability.47Please retain this notice, source attribution, license, and change notices48when redistributing the model or further derivatives.49 