initial release

README.md

@@ -0,0 +1,111 @@
+---
+language:
+- "ja"
+tags:
+- "japanese"
+- "question-answering"
+- "dependency-parsing"
+datasets:
+- "universal_dependencies"
+license: "cc-by-sa-4.0"
+pipeline_tag: "question-answering"
+widget:
+- text: "国語"
+  context: "全学年にわたって小学校の国語の教科書に挿し絵が用いられている"
+- text: "教科書"
+  context: "全学年にわたって小学校の国語の教科書に挿し絵が用いられている"
+- text: "の"
+  context: "全学年にわたって小学校の国語[MASK]教科書に挿し絵が用いられている"
+---
+
+# deberta-large-japanese-unidic-ud-head
+
+## Model Description
+
+This is a DeBERTa(V2) model pretrained on 青空文庫 for dependency-parsing (head-detection on long-unit-words) as question-answering, derived from [deberta-large-japanese-unidic](https://huggingface.co/KoichiYasuoka/deberta-large-japanese-unidic) and [UD_Japanese-GSDLUW](https://github.com/UniversalDependencies/UD_Japanese-GSDLUW). Use [MASK] inside `context` to avoid ambiguity when specifying a multiple-used word as `question`.
+
+## How to Use
+
+```py
+import torch
+from transformers import AutoTokenizer,AutoModelForQuestionAnswering
+tokenizer=AutoTokenizer.from_pretrained("KoichiYasuoka/deberta-large-japanese-unidic-ud-head")
+model=AutoModelForQuestionAnswering.from_pretrained("KoichiYasuoka/deberta-large-japanese-unidic-ud-head")
+question="国語"
+context="全学年にわたって小学校の国語の教科書に挿し絵が用いられている"
+inputs=tokenizer(question,context,return_tensors="pt")
+outputs=model(**inputs)
+start,end=torch.argmax(outputs.start_logits),torch.argmax(outputs.end_logits)
+print(tokenizer.convert_ids_to_tokens(inputs["input_ids"][0,start:end+1]))
+```
+
+or
+
+```py
+from transformers import (AutoTokenizer,AutoModelForQuestionAnswering,
+  AutoModelForTokenClassification,AutoConfig,TokenClassificationPipeline)
+class TaggerPipeline(TokenClassificationPipeline):
+  def __call__(self,text):
+    d=super().__call__(text)
+    if len(d)>0 and ("start" not in d[0] or d[0]["start"]==None):
+      import tokenizations
+      v=[x["word"] for x in d]
+      a2b,b2a=tokenizations.get_alignments(v,text)
+      for i,t in enumerate(a2b):
+        s,e=(0,0) if t==[] else (t[0],t[-1]+1)
+        if t==[] and v[i]==self.tokenizer.unk_token:
+          s=([[-1]]+[x for x in a2b[0:i] if x>[]])[-1][-1]+1
+          e=([x for x in a2b[i+1:] if x>[]]+[[len(text)]])[0][0]
+        d[i]["start"],d[i]["end"]=s,e
+    return d
+class TransformersUD(object):
+  def __init__(self,bert):
+    import os
+    self.tokenizer=AutoTokenizer.from_pretrained(bert)
+    self.model=AutoModelForQuestionAnswering.from_pretrained(bert)
+    x=AutoModelForTokenClassification.from_pretrained
+    if os.path.isdir(bert):
+      d,t=x(os.path.join(bert,"deprel")),x(os.path.join(bert,"tagger"))
+    else:
+      from transformers.file_utils import hf_bucket_url
+      c=AutoConfig.from_pretrained(hf_bucket_url(bert,"deprel/config.json"))
+      d=x(hf_bucket_url(bert,"deprel/pytorch_model.bin"),config=c)
+      s=AutoConfig.from_pretrained(hf_bucket_url(bert,"tagger/config.json"))
+      t=x(hf_bucket_url(bert,"tagger/pytorch_model.bin"),config=s)
+    self.deprel=TaggerPipeline(model=d,tokenizer=self.tokenizer,
+      aggregation_strategy="simple")
+    self.tagger=TaggerPipeline(model=t,tokenizer=self.tokenizer)
+  def __call__(self,text):
+    import numpy,torch,ufal.chu_liu_edmonds
+    w=[(t["start"],t["end"],t["entity_group"]) for t in self.deprel(text)]
+    z,n={t["start"]:t["entity"].split("|") for t in self.tagger(text)},len(w)
+    r,m=[text[s:e] for s,e,p in w],numpy.full((n+1,n+1),numpy.nan)
+    v,c=self.tokenizer(r,add_special_tokens=False)["input_ids"],[]
+    for i,t in enumerate(v):
+      q=[self.tokenizer.cls_token_id]+t+[self.tokenizer.sep_token_id]
+      c.append([q]+v[0:i]+[[self.tokenizer.mask_token_id]]+v[i+1:]+[[q[-1]]])
+    b=[[len(sum(x[0:j+1],[])) for j in range(len(x))] for x in c]
+    d=self.model(input_ids=torch.tensor([sum(x,[]) for x in c]),
+      token_type_ids=torch.tensor([[0]*x[0]+[1]*(x[-1]-x[0]) for x in b]))
+    s,e=d.start_logits.tolist(),d.end_logits.tolist()
+    for i in range(n):
+      for j in range(n):
+        m[i+1,0 if i==j else j+1]=s[i][b[i][j]]+e[i][b[i][j+1]-1]
+    h=ufal.chu_liu_edmonds.chu_liu_edmonds(m)[0]
+    if [0 for i in h if i==0]!=[0]:
+      i=([p for s,e,p in w]+["root"]).index("root")
+      j=i+1 if i<n else numpy.nanargmax(m[:,0])
+      m[0:j,0]=m[j+1:,0]=numpy.nan
+      h=ufal.chu_liu_edmonds.chu_liu_edmonds(m)[0]
+    u="# text = "+text.replace("\n"," ")+"\n"
+    for i,(s,e,p) in enumerate(w,1):
+      p="root" if h[i]==0 else "dep" if p=="root" else p
+      u+="\t".join([str(i),r[i-1],"_",z[s][0][2:],"_","|".join(z[s][1:]),
+        str(h[i]),p,"_","_" if i<n and w[i][0]<e else "SpaceAfter=No"])+"\n"
+    return u+"\n"
+
+nlp=TransformersUD("KoichiYasuoka/deberta-large-japanese-unidic-ud-head")
+print(nlp("全学年にわたって小学校の国語の教科書に挿し絵が用いられ���いる"))
+```
+
+[fugashi](https://pypi.org/project/fugashi) [unidic-lite](https://pypi.org/project/unidic-lite) [pytokenizations](https://pypi.org/project/pytokenizations) and [ufal.chu-liu-edmonds](https://pypi.org/project/ufal.chu-liu-edmonds/) required.

config.json

@@ -0,0 +1,31 @@
+{
+  "architectures": [
+    "DebertaV2ForQuestionAnswering"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "bos_token_id": 0,
+  "eos_token_id": 2,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-07,
+  "max_position_embeddings": 512,
+  "max_relative_positions": -1,
+  "model_type": "deberta-v2",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "pad_token_id": 1,
+  "pooler_dropout": 0,
+  "pooler_hidden_act": "gelu",
+  "pooler_hidden_size": 1024,
+  "pos_att_type": null,
+  "position_biased_input": true,
+  "relative_attention": false,
+  "tokenizer_class": "BertJapaneseTokenizer",
+  "torch_dtype": "float32",
+  "transformers_version": "4.19.4",
+  "type_vocab_size": 0,
+  "vocab_size": 32000
+}

deprel/config.json

@@ -0,0 +1,127 @@
+{
+  "architectures": [
+    "DebertaV2ForTokenClassification"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "bos_token_id": 0,
+  "eos_token_id": 2,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "id2label": {
+    "0": "B-acl",
+    "1": "B-advcl",
+    "2": "B-advmod",
+    "3": "B-amod",
+    "4": "B-aux",
+    "5": "B-case",
+    "6": "B-cc",
+    "7": "B-ccomp",
+    "8": "B-compound",
+    "9": "B-cop",
+    "10": "B-csubj",
+    "11": "B-dep",
+    "12": "B-det",
+    "13": "B-discourse",
+    "14": "B-dislocated",
+    "15": "B-fixed",
+    "16": "B-mark",
+    "17": "B-nmod",
+    "18": "B-nsubj",
+    "19": "B-nummod",
+    "20": "B-obj",
+    "21": "B-obl",
+    "22": "B-punct",
+    "23": "B-root",
+    "24": "I-acl",
+    "25": "I-advcl",
+    "26": "I-advmod",
+    "27": "I-amod",
+    "28": "I-aux",
+    "29": "I-case",
+    "30": "I-cc",
+    "31": "I-ccomp",
+    "32": "I-compound",
+    "33": "I-csubj",
+    "34": "I-dep",
+    "35": "I-discourse",
+    "36": "I-dislocated",
+    "37": "I-fixed",
+    "38": "I-mark",
+    "39": "I-nmod",
+    "40": "I-nsubj",
+    "41": "I-nummod",
+    "42": "I-obj",
+    "43": "I-obl",
+    "44": "I-punct",
+    "45": "I-root"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "label2id": {
+    "B-acl": 0,
+    "B-advcl": 1,
+    "B-advmod": 2,
+    "B-amod": 3,
+    "B-aux": 4,
+    "B-case": 5,
+    "B-cc": 6,
+    "B-ccomp": 7,
+    "B-compound": 8,
+    "B-cop": 9,
+    "B-csubj": 10,
+    "B-dep": 11,
+    "B-det": 12,
+    "B-discourse": 13,
+    "B-dislocated": 14,
+    "B-fixed": 15,
+    "B-mark": 16,
+    "B-nmod": 17,
+    "B-nsubj": 18,
+    "B-nummod": 19,
+    "B-obj": 20,
+    "B-obl": 21,
+    "B-punct": 22,
+    "B-root": 23,
+    "I-acl": 24,
+    "I-advcl": 25,
+    "I-advmod": 26,
+    "I-amod": 27,
+    "I-aux": 28,
+    "I-case": 29,
+    "I-cc": 30,
+    "I-ccomp": 31,
+    "I-compound": 32,
+    "I-csubj": 33,
+    "I-dep": 34,
+    "I-discourse": 35,
+    "I-dislocated": 36,
+    "I-fixed": 37,
+    "I-mark": 38,
+    "I-nmod": 39,
+    "I-nsubj": 40,
+    "I-nummod": 41,
+    "I-obj": 42,
+    "I-obl": 43,
+    "I-punct": 44,
+    "I-root": 45
+  },
+  "layer_norm_eps": 1e-07,
+  "max_position_embeddings": 512,
+  "max_relative_positions": -1,
+  "model_type": "deberta-v2",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "pad_token_id": 1,
+  "pooler_dropout": 0,
+  "pooler_hidden_act": "gelu",
+  "pooler_hidden_size": 1024,
+  "pos_att_type": null,
+  "position_biased_input": true,
+  "relative_attention": false,
+  "tokenizer_class": "BertJapaneseTokenizer",
+  "torch_dtype": "float32",
+  "transformers_version": "4.19.4",
+  "type_vocab_size": 0,
+  "vocab_size": 32000
+}

deprel/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1b76204a00e2ebc32a592fdd6408ff61a27cc9d2fbd1c1e5176fe7f58801f1b
+size 1342740275

deprel/special_tokens_map.json

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

deprel/tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]", "do_lower_case": false, "do_word_tokenize": true, "do_subword_tokenize": true, "word_tokenizer_type": "mecab", "subword_tokenizer_type": "wordpiece", "never_split": ["[CLS]", "[PAD]", "[SEP]", "[UNK]", "[MASK]"], "mecab_kwargs": {"mecab_dic": "unidic_lite"}, "model_max_length": 512, "tokenizer_class": "BertJapaneseTokenizer"}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53fd6776eae30439ef3af59a4ba242444f7ff45ecbfb7c8d4cd366084cd5bac6
+size 1342559923

special_tokens_map.json

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

tagger/config.json

@@ -0,0 +1,107 @@
+{
+  "architectures": [
+    "DebertaV2ForTokenClassification"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "bos_token_id": 0,
+  "eos_token_id": 2,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "id2label": {
+    "0": "B-ADJ|_",
+    "1": "B-ADP|_",
+    "2": "B-ADV|_",
+    "3": "B-AUX|Polarity=Neg",
+    "4": "B-AUX|_",
+    "5": "B-CCONJ|_",
+    "6": "B-DET|_",
+    "7": "B-INTJ|_",
+    "8": "B-NOUN|Polarity=Neg",
+    "9": "B-NOUN|_",
+    "10": "B-NUM|_",
+    "11": "B-PART|_",
+    "12": "B-PRON|_",
+    "13": "B-PROPN|_",
+    "14": "B-PUNCT|_",
+    "15": "B-SCONJ|_",
+    "16": "B-SYM|_",
+    "17": "B-VERB|_",
+    "18": "B-X|_",
+    "19": "I-ADJ|_",
+    "20": "I-ADP|_",
+    "21": "I-ADV|_",
+    "22": "I-AUX|Polarity=Neg",
+    "23": "I-AUX|_",
+    "24": "I-CCONJ|_",
+    "25": "I-INTJ|_",
+    "26": "I-NOUN|_",
+    "27": "I-NUM|_",
+    "28": "I-PART|_",
+    "29": "I-PRON|_",
+    "30": "I-PROPN|_",
+    "31": "I-PUNCT|_",
+    "32": "I-SCONJ|_",
+    "33": "I-SYM|_",
+    "34": "I-VERB|_",
+    "35": "I-X|_"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "label2id": {
+    "B-ADJ|_": 0,
+    "B-ADP|_": 1,
+    "B-ADV|_": 2,
+    "B-AUX|Polarity=Neg": 3,
+    "B-AUX|_": 4,
+    "B-CCONJ|_": 5,
+    "B-DET|_": 6,
+    "B-INTJ|_": 7,
+    "B-NOUN|Polarity=Neg": 8,
+    "B-NOUN|_": 9,
+    "B-NUM|_": 10,
+    "B-PART|_": 11,
+    "B-PRON|_": 12,
+    "B-PROPN|_": 13,
+    "B-PUNCT|_": 14,
+    "B-SCONJ|_": 15,
+    "B-SYM|_": 16,
+    "B-VERB|_": 17,
+    "B-X|_": 18,
+    "I-ADJ|_": 19,
+    "I-ADP|_": 20,
+    "I-ADV|_": 21,
+    "I-AUX|Polarity=Neg": 22,
+    "I-AUX|_": 23,
+    "I-CCONJ|_": 24,
+    "I-INTJ|_": 25,
+    "I-NOUN|_": 26,
+    "I-NUM|_": 27,
+    "I-PART|_": 28,
+    "I-PRON|_": 29,
+    "I-PROPN|_": 30,
+    "I-PUNCT|_": 31,
+    "I-SCONJ|_": 32,
+    "I-SYM|_": 33,
+    "I-VERB|_": 34,
+    "I-X|_": 35
+  },
+  "layer_norm_eps": 1e-07,
+  "max_position_embeddings": 512,
+  "max_relative_positions": -1,
+  "model_type": "deberta-v2",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "pad_token_id": 1,
+  "pooler_dropout": 0,
+  "pooler_hidden_act": "gelu",
+  "pooler_hidden_size": 1024,
+  "pos_att_type": null,
+  "position_biased_input": true,
+  "relative_attention": false,
+  "tokenizer_class": "BertJapaneseTokenizer",
+  "torch_dtype": "float32",
+  "transformers_version": "4.19.4",
+  "type_vocab_size": 0,
+  "vocab_size": 32000
+}

tagger/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:774dc23b70f8ddea398ff14ea4e07e6528e2f457344696d20a8f7fbb206f01e1
+size 1342699315

tagger/special_tokens_map.json

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

tagger/tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]", "do_lower_case": false, "do_word_tokenize": true, "do_subword_tokenize": true, "word_tokenizer_type": "mecab", "subword_tokenizer_type": "wordpiece", "never_split": ["[CLS]", "[PAD]", "[SEP]", "[UNK]", "[MASK]"], "mecab_kwargs": {"mecab_dic": "unidic_lite"}, "model_max_length": 512, "tokenizer_class": "BertJapaneseTokenizer"}

tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]", "do_lower_case": false, "do_word_tokenize": true, "do_subword_tokenize": true, "word_tokenizer_type": "mecab", "subword_tokenizer_type": "wordpiece", "never_split": ["[CLS]", "[PAD]", "[SEP]", "[UNK]", "[MASK]"], "mecab_kwargs": {"mecab_dic": "unidic_lite"}, "model_max_length": 512, "tokenizer_class": "BertJapaneseTokenizer"}