ELECTRA-base-discriminator is a large language model released by Google/Stanford, released under the Apache-2.0 license, with a claimed knowledge cutoff of 2020-03-23. Its weights, configuration, and tokenizer files have been cryptographically hashed and are queued for on-chain attestation by Weight Registry.
Description
Trained via replaced-token detection -- a small generator substitutes tokens, a discriminator predicts which were replaced -- giving a learning signal from every token rather than just the ~15% masked ones in BERT-style pretraining. Matches RoBERTa's downstream performance at a fraction of the pretraining compute.
Metadata
Hugging Face stats
46,483,024 downloads on Hugging Face · 186 likes · library: transformers
Weight Hashes
Weights file
SHA-256ff381b988006468d184eb8c1c090be15553155daa960587e8d86465969f1a0c1
SHA3-2565ad421435387c76c7178c7e7792c2571640eee915b6512d2bfafb71ca0dc7c26
BLAKE344876f7d743ea5273eb2bea89d606fe0ec3a137a76ba3b6067971308a73c2c4d
Config file
SHA-2560a2f56f315dfa0c24d29126fc645031dedb65789c8ac73df486404501c4cdabb
SHA3-25685e90688bfef2f9cc069a6196f8682bd1e4380944a13a5a560ba19a2d53c7a02
BLAKE3347a94cd9c2f1ac0563814fffa0ea421e5e93c886c5a1b0afe6915900b67f9a0
Tokenizer file
SHA-256ce64fce797c24f68df90b40a3f74f579b336a493db14bd583fd520ea0d8c9a98
SHA3-25648e0eb031593b2d46432e97a97073752b4d62db14a58838b8381b1446916b986
BLAKE33d5a598086b8e94c303d07454f3451cdb37cecf75c207fc37958e7c06ff07878
Verify it yourself
curl -L -o "pytorch_model.bin" "https://huggingface.co/google/electra-base-discriminator/resolve/1ae76a97c7e84a4e640876a07453fccd636f0667/pytorch_model.bin" sha256sum "pytorch_model.bin"
Expected SHA-256: ff381b988006468d184eb8c1c090be15553155daa960587e8d86465969f1a0c1