initial release

README.md

@@ -0,0 +1,29 @@
+---
+language:
+- "lzh"
+tags:
+- "classical chinese"
+- "literary chinese"
+- "ancient chinese"
+- "masked-lm"
+license: "apache-2.0"
+pipeline_tag: "fill-mask"
+mask_token: "[MASK]"
+widget:
+- text: "孟子[MASK]梁惠王"
+---
+
+# modernbert-small-classical-chinese-traditional
+
+## Model Description
+
+This is a ModernBERT model pre-trained on [Kanripo](https://www.kanripo.org) texts. NVIDIA A100-SXM4-40GB×8 took 7 hours 8 minutes for training. You can fine-tune `modernbert-small-classical-chinese-traditional` 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-traditional")
+model=AutoModelForMaskedLM.from_pretrained("KoichiYasuoka/modernbert-small-classical-chinese-traditional")
+```
+

config.json

@@ -0,0 +1,46 @@
+{
+  "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": true,
+  "sep_token_id": 2,
+  "sparse_pred_ignore_index": -100,
+  "sparse_prediction": false,
+  "tokenizer_class": "BertTokenizerFast",
+  "torch_dtype": "float32",
+  "transformers_version": "4.51.3",
+  "vocab_size": 23798
+}

maker.py

@@ -0,0 +1,99 @@
+#! /usr/bin/python3
+#pip3 install transformers accelerate deepspeed triton esupar
+tgt="KoichiYasuoka/modernbert-small-classical-chinese-traditional"
+import os
+os.system("""
+if test -d Kanripo-data
+then :
+else git clone --depth=1 https://github.com/DHSinology/Kanripo-data
+fi""")
+if not os.path.isfile("train.txt"):
+  import glob,subprocess
+  from esupar.tradify import tradify
+  with open("train.txt","w",encoding="utf-8") as w:
+    s='''
+{
+  s=$0;
+  while(s>""){
+    if(f==1){
+      i=index(s,"</hi>");
+      if(i>0){
+        h=h substr(s,1,i-1)"\\n";
+        s=substr(s,i+5);
+        f=0;
+      }
+      else{
+        h=h s;
+        s="";
+      }
+    }
+    else{
+      i=index(s,"<hi>");
+      if(i>0){
+        t=substr(s,1,i-1);
+        gsub(/<[^>]+>/,"",t);
+        gsub(/([ -~]|　|\\t)/,"",t);
+        printf("%s",t);
+        s=substr(s,i+4);
+        f=1;
+      }
+      else{
+        gsub(/<[^>]+>/,"",s);
+        gsub(/([ -~]|　|\\t)/,"",s);
+        printf("%s",s);
+        s="";
+      }
+    }
+  }
+  printf("\\n");
+}
+END{
+  gsub(/<[^>]+>/,"",h);
+  gsub(/([ -~]|　|\\t)/,"",h);
+  printf("%s\\n",h);
+}'''
+    t=subprocess.run(["nawk",s]+glob.glob("Kanripo-data/data/*/*.xml"),stdout=subprocess.PIPE,encoding="utf-8").stdout
+    u=""
+    for s in t.split("\n"):
+      s=s.strip()
+      if len(s)+len(u)>=8200:
+        print(u,file=w)
+        u=""
+      for c in s:
+        u+=tradify[c] if c in tradify else c
+    print(u,file=w)
+
+from transformers import BertTokenizerFast
+from tokenizers.pre_tokenizers import Sequence,Whitespace,Split
+from tokenizers import Regex
+s=["[CLS]","[PAD]","[SEP]","[UNK]","[MASK]"]
+if not os.path.isfile("vocab.txt"):
+  with open("train.txt","r",encoding="utf-8") as r:
+    v=set(c for c in r.read() if not c.isspace())
+  with open("vocab.txt","w",encoding="utf-8") as w:
+    print("\n".join(s+sorted(v)),file=w)
+tkz=BertTokenizerFast(vocab_file="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)
+
+with open("train.py","w",encoding="utf-8") as w:
+  print(f'#! /usr/bin/env deepspeed\ntgt="{tgt}"'+'''
+from transformers import BertTokenizerFast,ModernBertForMaskedLM,AutoConfig,DataCollatorForLanguageModeling,TrainingArguments,Trainer
+tkz=BertTokenizerFast.from_pretrained(tgt)
+c={"trust_remote_code":True,"vocab_size":len(tkz),"tokenizer_class":type(tkz).__name__}
+for k,v in tkz.special_tokens_map.items():
+  c[k+"_id"]=tkz.convert_tokens_to_ids(v)
+cfg=AutoConfig.from_pretrained("KoichiYasuoka/modernbert-small-japanese-char",**c)
+arg=TrainingArguments(num_train_epochs=3,per_device_train_batch_size=1,output_dir=tgt,overwrite_output_dir=True,save_total_limit=2,save_safetensors=False)
+class ReadLineDS(object):
+  def __init__(self,file,tokenizer):
+    self.tokenizer=tokenizer
+    with open(file,"r",encoding="utf-8") as r:
+      self.lines=[s.strip() for s in r if s.strip()>""]
+  __len__=lambda self:len(self.lines)
+  __getitem__=lambda self,i:self.tokenizer(self.lines[i],truncation=True,add_special_tokens=True,max_length=8190)
+trn=Trainer(args=arg,data_collator=DataCollatorForLanguageModeling(tkz),model=ModernBertForMaskedLM(cfg),train_dataset=ReadLineDS("train.txt",tkz))
+trn.train()
+trn.save_model(tgt)''',file=w)
+os.system("chmod 755 train.py ; ./train.py")

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1214014c0405b087cf1b82910574c09b9fcacb1d1a0347c0a0ad8648ddee30cb
+size 79320542

special_tokens_map.json

@@ -0,0 +1,37 @@
+{
+  "cls_token": {
+    "content": "[CLS]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "mask_token": {
+    "content": "[MASK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "[PAD]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "sep_token": {
+    "content": "[SEP]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "[UNK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

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": "BertTokenizerFast",
+  "unk_token": "[UNK]"
+}