cogdata.utils.cogview package¶
cogdata.utils.cogview.api module¶
-
cogdata.utils.cogview.api.
code2img
(model, code)¶ Convert a batch of code to imgs :param model: … :param code: [b, h, w] or [b, h*w] LongTensor
-
cogdata.utils.cogview.api.
img2code
(model, img)¶ Convert a batch of img to code :param model: The tokenizer model. :param img: [b, c, h, w]
-
cogdata.utils.cogview.api.
new_model
()¶ Return a New Instance of VQVAE, the same parameters with the pretrained model. This is for torch.load().
cogdata.utils.cogview.sp_tokenizer module¶
SentencePiece tokenizer. from https://github.com/openai/gpt-2/, changed for chinese
SentencePiece is an unsupervised text tokenizer and detokenizer mainly for Neural Network-based text generation systems where the vocabulary size is predetermined prior to the neural model training. SentencePiece implements subword units (e.g., byte-pair-encoding (BPE) [Sennrich et al.]) and unigram language model [Kudo.]) with the extension of direct training from raw sentences. SentencePiece allows us to make a purely end-to-end system that does not depend on language-specific pre/postprocessing. https://github.com/google/sentencepiece
pip install sentencepiece
or:
git clone https://github.com/google/sentencepiece.git
python setup.py install
cogdata.utils.cogview.unified_tokenizer module¶
-
cogdata.utils.cogview.unified_tokenizer.
get_tokenizer
(img_tokenizer_path=None)¶ Singlton
Return an image tokenizer
cogdata.utils.cogview.vqvae_tokenizer module¶
This module defines the tokenizer used in Cogdata
cogdata.utils.cogview.vqvae_zc module¶
This module defines the model used in Cogdata