ALBERT-base-v2 is a large language model released by Google, released under the Apache-2.0 license, with a claimed knowledge cutoff of 2019-09-26. Its weights, configuration, and tokenizer files have been cryptographically hashed and are queued for on-chain attestation by Weight Registry.
Description
Parameter-reduced BERT variant (12M params vs BERT-base's 110M) using factorized embedding parameterization and cross-layer weight sharing to cut parameter count while retaining BERT-level performance; v2 improves on the original ALBERT paper's checkpoints via longer training and no dropout.
Metadata
Hugging Face stats
1,053,072 downloads on Hugging Face · 150 likes · pipeline: fill-mask · library: transformers
Weight Hashes
Weights file
SHA-256d24c30ec641e40087eed68adf958afe2dab0432790ec3ffa88b31ce773ae33eb
SHA3-25684f24d89f960f0c2d7e7c7693db3ab524a883f91024754e3a3be11966b492035
BLAKE3972c010b61cce471a163bbe817421554e49fa15294771259de40cd591b8fdd92
Config file
SHA-256f5b7ef285a15ebbc7043b23284e60ac6f6de8d0a6051e282dfdf81da9351dc03
SHA3-256c8ab6c6d63d8ef5d1a91dc696e957dcf993597e9bb81cac578c5eabd151a14a4
BLAKE352b83dbba222c61fbcce2621c50aa064c8aad97218094af3cdc14f1c9a196742
Tokenizer file
SHA-2566e2aa3d5ca4c6841bc36a5028179cb4d63c129931fa18fce38fdb38cec4d2ac3
SHA3-256cbad27f9161df74f0564fab22cf47565e2646b518723942eec8cb73d194be050
BLAKE3ae18655224267d74d5576d766a880c53480028c5ddfcae96a2d0d5cafaea7ee0
Verify it yourself
curl -L -o "model.safetensors" "https://huggingface.co/albert/albert-base-v2/resolve/8e2f239c5f8a2c0f253781ca60135db913e5c80c/model.safetensors" sha256sum "model.safetensors"
Expected SHA-256: d24c30ec641e40087eed68adf958afe2dab0432790ec3ffa88b31ce773ae33eb