ALBERT-Base v2 (albert_base_v2.tar.gz): preservation copy
收藏资源简介:
A preservation copy of the ALBERT-Base version 2 pre-trained model, exactly as Google released it: albert_base_v2.tar.gz, byte for byte, unconverted. The archive unpacks to albert_base/: the TensorFlow 1 checkpoint (model.ckpt-best.*), albert_config.json and the SentencePiece vocabulary (30k-clean.model, 30k-clean.vocab). The original URL, https://storage.googleapis.com/albert_models/albert_base_v2.tar.gz, now answers 403 Forbidden. Converted copies exist elsewhere, but code that restores this TensorFlow 1 checkpoint by its variable names needs these exact files. Verified provenance: retrieved from the Internet Archive's Wayback Machine. Its five successful captures of the original URL, taken between 2021-08-31 and 2025-01-30, all carry the SHA-1 digest Q7SSZR7DIH7PFAEI55BQPIKURLGTNNX7 (base32), and the deposited file hashes to that same value. sha256 41e31e24abe89dd7f5d9e007e0d03ba45055b388adbb1cf781585a5068824727, 44,505,246 bytes. Terms: ALBERT (Lan et al., google-research/albert) is released under the Apache License 2.0, bundled as LICENSE; this deposit redistributes the archive unmodified under the same license. See NOTICE.



