commit files to HF hub

config.json

@@ -0,0 +1,12 @@
+{
+ "architectures": [
+ "CustomModel"
+ ],
+ "auto_map": {
+ "AutoConfig": "__main__.CustomConfig",
+ "AutoModel": "__main__.CustomModel"
+ },
+ "model_type": "test-zeroshot",
+ "torch_dtype": "float32",
+ "transformers_version": "4.23.1"
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c10506fbdef05cb01eaaaacc2889d89634cba745074f72ee9a2b8430c9a07dd4
+size 765

special_tokens_map.json

@@ -0,0 +1,7 @@
+{
+ "cls_token": "[CLS]",
+ "mask_token": "[MASK]",
+ "pad_token": "[PAD]",
+ "sep_token": "[SEP]",
+ "unk_token": "[UNK]"
+}

tokenizer_config.json

@@ -0,0 +1,20 @@
+{
+ "cls_token": "[CLS]",
+ "do_lower_case": false,
+ "do_subword_tokenize": true,
+ "do_word_tokenize": true,
+ "jumanpp_kwargs": null,
+ "mask_token": "[MASK]",
+ "mecab_kwargs": null,
+ "model_max_length": 512,
+ "name_or_path": "cl-tohoku/bert-base-japanese",
+ "never_split": null,
+ "pad_token": "[PAD]",
+ "sep_token": "[SEP]",
+ "special_tokens_map_file": null,
+ "subword_tokenizer_type": "wordpiece",
+ "sudachi_kwargs": null,
+ "tokenizer_class": "BertJapaneseTokenizer",
+ "unk_token": "[UNK]",
+ "word_tokenizer_type": "mecab"
+}

zeroshot.py

@@ -0,0 +1,48 @@
+from transformers.modeling_outputs import SequenceClassifierOutput
+from transformers.modeling_utils import PreTrainedModel
+from transformers.configuration_utils import PretrainedConfig
+import torch
+from transformers import ZeroShotClassificationPipeline
+
+
+class CustomConfig(PretrainedConfig):
+ model_type = "test-zeroshot"
+
+ def __init__(self, **kwargs):
+ super().__init__(**kwargs)
+
+
+class CustomModel(PreTrainedModel):
+ config_class = CustomConfig
+
+ def __init__(self, config: CustomConfig):
+ super().__init__(config)
+ self.config = config
+ self.embeddings = torch.nn.Embedding(num_embeddings=1, embedding_dim=1)
+
+ def forward(self, **kwargs) -> SequenceClassifierOutput:
+
+ return SequenceClassifierOutput(logits=torch.tensor([[1, 2, 3]]))
+
+
+from transformers.pipelines import PIPELINE_REGISTRY
+
+from transformers import AutoModelForSequenceClassification, TFAutoModelForSequenceClassification
+
+if __name__ == "__main__":
+ from transformers import pipeline
+ classifier = pipeline("zero-shot-classification",
+ model="cl-tohoku/bert-base-japanese")
+ from transformers import AutoConfig, AutoModel, AutoModelForImageClassification
+
+ CustomConfig.register_for_auto_class()
+ CustomModel.register_for_auto_class("AutoModel")
+
+ p = ZeroShotClassificationPipeline(model=CustomModel(CustomConfig()),
+ tokenizer=classifier.tokenizer)
+ from huggingface_hub import Repository
+
+ repo = Repository("zero-shot-classification",
+ clone_from="paulhindemith/zero-shot-classification")
+ p.save_pretrained("zero-shot-classification")
+ repo.push_to_hub()