MELD (mirror of revision ddfc9c7)

This is an unmodified mirror of anon-review-meld-2026/meld at revision ddfc9c789b8c8d0f92d3774265316d418c96dc67. Only this README differs from the original: the mirror note above the original text. All credit goes to the MELD authors.

model.safetensors has SHA-256 14b0fad5a99c5746e93e19ed54bc6de02b104381de27a391709bbd42264fa504 (4,112,580,996 bytes). Other revisions of MELD have different weights and give different scores.

The MELD scores of the twicetold dataset come from this revision.


MELD

An AI-generated text detector for English. Everything needed to run it is in this repository.

pip install torch transformers safetensors
python meld.py "Paste the text to check here."

For each input, meld.py prints a score, p_ai, and whether the text is flagged at the shipped 1% false-positive threshold (score_offsets in meld_config.json). Compare scores against that threshold, or against scores of your own human-written reference texts, rather than against a fixed cut. Inputs should be at least about 100 words. The scoring head is custom, so pipeline() and AutoModel do not apply; use meld.py or the Scorer class inside it.

Downloads last month
32
Safetensors
Model size
1B params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for IlyaGusev/meld

Finetuned
(31)
this model