initial release

README.md

@@ -0,0 +1,30 @@
+---
+language:
+- "lzh"
+tags:
+- "classical chinese"
+- "literary chinese"
+- "ancient chinese"
+- "masked-lm"
+base_model: KoichiYasuoka/modernbert-small-classical-chinese-traditional
+license: "apache-2.0"
+pipeline_tag: "fill-mask"
+mask_token: "[MASK]"
+widget:
+- text: "孟子[MASK]梁惠王"
+---
+
+# modernbert-small-classical-chinese
+
+## Model Description
+
+This is a ModernBERT model pre-trained on [Kanripo](https://www.kanripo.org) texts. Character-embeddings are enhanced into traditional/simplified characters. You can fine-tune `modernbert-small-classical-chinese` for downstream tasks, such as sentence-segmentation, POS-tagging, dependency-parsing, and so on.
+
+## How to Use
+
+```py
+from transformers import AutoTokenizer,AutoModelForMaskedLM
+tokenizer=AutoTokenizer.from_pretrained("KoichiYasuoka/modernbert-small-classical-chinese")
+model=AutoModelForMaskedLM.from_pretrained("KoichiYasuoka/modernbert-small-classical-chinese")
+```
+

config.json

@@ -0,0 +1,48 @@
+{
+  "_name_or_path": "KoichiYasuoka/modernbert-small-classical-chinese-traditional",
+  "architectures": [
+    "ModernBertForMaskedLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 0,
+  "classifier_activation": "gelu",
+  "classifier_bias": false,
+  "classifier_dropout": 0.0,
+  "classifier_pooling": "mean",
+  "cls_token_id": 0,
+  "decoder_bias": true,
+  "deterministic_flash_attn": false,
+  "embedding_dropout": 0.0,
+  "eos_token_id": 2,
+  "global_attn_every_n_layers": 3,
+  "global_rope_theta": 160000.0,
+  "gradient_checkpointing": false,
+  "hidden_activation": "gelu",
+  "hidden_size": 256,
+  "initializer_cutoff_factor": 2.0,
+  "initializer_range": 0.02,
+  "intermediate_size": 768,
+  "layer_norm_eps": 1e-05,
+  "local_attention": 128,
+  "local_rope_theta": 10000.0,
+  "max_position_embeddings": 8192,
+  "mlp_bias": false,
+  "mlp_dropout": 0.0,
+  "model_type": "modernbert",
+  "norm_bias": false,
+  "norm_eps": 1e-05,
+  "num_attention_heads": 8,
+  "num_hidden_layers": 16,
+  "pad_token_id": 1,
+  "position_embedding_type": "absolute",
+  "reference_compile": false,
+  "repad_logits_with_grad": false,
+  "sep_token_id": 2,
+  "sparse_pred_ignore_index": -100,
+  "sparse_prediction": false,
+  "tokenizer_class": "BertTokenizerFast",
+  "torch_dtype": "float32",
+  "transformers_version": "4.48.3",
+  "vocab_size": 25078
+}

maker.py

@@ -0,0 +1,25 @@
+#! /usr/bin/python3
+src="KoichiYasuoka/modernbert-small-classical-chinese-traditional"
+tgt="KoichiYasuoka/modernbert-small-classical-chinese"
+import torch
+from transformers import BertTokenizerFast,AutoModelForMaskedLM
+from esupar.tradify import tradify
+from tokenizers.pre_tokenizers import Sequence,Whitespace,Split
+from tokenizers import Regex
+tkz=BertTokenizerFast.from_pretrained(src)
+mdl=AutoModelForMaskedLM.from_pretrained(src)
+c=[(k,v) for k,v in tradify.items() if tkz.add_tokens([k,v])==1]
+e=mdl.resize_token_embeddings(len(tkz))
+with torch.no_grad():
+  for k,v in c:
+    t=sorted(tkz.convert_tokens_to_ids([k,v]))
+    e.weight[t[1],:]=e.weight[t[0],:]
+mdl.set_input_embeddings(e)
+mdl.save_pretrained(tgt,safe_serialization=False)
+with open(tgt+"/vocab.txt","w",encoding="utf-8") as w:
+  print("\n".join(tkz.convert_ids_to_tokens(range(len(tkz)))),file=w)
+s=["[CLS]","[PAD]","[SEP]","[UNK]","[MASK]"]
+tkz=BertTokenizerFast(vocab_file=tgt+"/vocab.txt",never_split=s,do_lower_case=False,strip_accents=False,tokenize_chinese_chars=True)
+tkz.backend_tokenizer.pre_tokenizer=Sequence([Whitespace(),Split(Regex("."),"isolated")])
+tkz.backend_tokenizer.decoder.prefix=tkz.backend_tokenizer.model.continuing_subword_prefix=""
+tkz.save_pretrained(tgt)

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:269d677fb2bbfbd0cac02012d8c421c6acf13b96c38f03846704f821d316651f
+size 80634974

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 +1,64 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "[CLS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "[SEP]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "[UNK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "4": {
+      "content": "[MASK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "clean_up_tokenization_spaces": true,
+  "cls_token": "[CLS]",
+  "do_basic_tokenize": true,
+  "do_lower_case": false,
+  "extra_special_tokens": {},
+  "mask_token": "[MASK]",
+  "model_max_length": 1000000000000000019884624838656,
+  "never_split": [
+    "[CLS]",
+    "[PAD]",
+    "[SEP]",
+    "[UNK]",
+    "[MASK]"
+  ],
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "strip_accents": false,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "unk_token": "[UNK]"
+}