coedit-large-litert-int8

CoEdIT Large LiteRT INT8

Split encoder/decoder LiteRT (.tflite) conversion of grammarly/coedit-large, intended for fully local, noncommercial text editing. This is an independently produced conversion and is not an official Grammarly release.

Noncommercial restriction

This model is licensed under CC BY-NC 4.0 and may not be used for commercial purposes unless separate permission is obtained from the relevant rights holders. See LICENSE for the license terms.

Model details

The encoder state is computed once and reused during autoregressive decoder steps. Tensor names, task prefixes, sizes, and SHA-256 checksums are recorded in manifest.json.

Files

• encoder.tflite: quantized encoder graph
• decoder.tflite: quantized autoregressive decoder graph
• tokenizer.json and spiece.model: tokenizer assets
• tokenizer_config.json and special_tokens_map.json: tokenizer metadata
• manifest.json: runtime contract and integrity metadata

Instruction prefixes

Fix grammatical errors in this sentence: <text>
Make the sentence formal: <text>
Change the style to casual: <text>
Make the sentence simpler: <text>
Paraphrase the sentence: <text>

Runtime integration

Load the two graphs with LiteRT, tokenize to the fixed 256-token input shape, run the encoder once, and feed its hidden states plus the decoder token buffer to the decoder for each generated token. Consult manifest.json for the exact tensor contract and EOS/padding IDs.

Limitations

• Browser CPU/WASM autoregressive decoding is computationally expensive. WebGPU or a native LiteRT runtime is strongly recommended.
• Dynamic-range quantization can change outputs relative to the source model.
• Output should be reviewed before use; the model may omit, add, or alter meaning.
• The graphs use fixed-size tensors and are not a Transformers checkpoint.

Attribution

• Source model: https://huggingface.co/grammarly/coedit-large
• Model authors: Vipul Raheja, Dhruv Kumar, Ryan Koo, and Dongyeop Kang
• CoEdIT paper: https://arxiv.org/abs/2305.09857
• Upstream base: https://huggingface.co/google/flan-t5-large

Changes made in this adaptation: conversion to fixed-shape split LiteRT encoder/decoder graphs and INT8 dynamic-range quantization.