release after the tokenizer refined

README.md

@@ -0,0 +1,30 @@
+---
+language:
+- "ja"
+tags:
+- "japanese"
+- "pos"
+- "dependency-parsing"
+base_model: goldfish-models/jpn_jpan_1000mb
+datasets:
+- "universal_dependencies"
+license: "apache-2.0"
+pipeline_tag: "token-classification"
+widget:
+- text: "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"
+---
+
+# goldfish-gpt2-japanese-1000mb-ud-causal
+
+## Model Description
+
+This is a GPT-2 model pretrained for POS-tagging and dependency-parsing, derived from [jpn_jpan_1000mb](https://huggingface.co/goldfish-models/jpn_jpan_1000mb) refined for [UD_Japanese-GSDLUW](https://github.com/UniversalDependencies/UD_Japanese-GSDLUW).
+
+## How to Use
+
+```
+from transformers import pipeline
+nlp=pipeline("universal-dependencies","KoichiYasuoka/goldfish-gpt2-japanese-1000mb-ud-causal",trust_remote_code=True)
+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ている"))
+```
+

config.json

@@ -0,0 +1,379 @@
+{
+  "activation_function": "gelu",
+  "architectures": [
+    "GPT2ForTokenClassification"
+  ],
+  "attn_pdrop": 0.1,
+  "bos_token_id": 50000,
+  "custom_pipelines": {
+    "upos": {
+      "impl": "ud.BellmanFordTokenClassificationPipeline",
+      "pt": "AutoModelForTokenClassification"
+    },
+    "universal-dependencies": {
+      "impl": "ud.UniversalDependenciesCausalPipeline",
+      "pt": "AutoModelForTokenClassification"
+    }
+  },
+  "embd_pdrop": 0.1,
+  "eos_token_id": 50001,
+  "id2label": {
+    "0": "ADJ",
+    "1": "ADJ|l-acl",
+    "2": "ADJ|l-advcl",
+    "3": "ADJ|l-amod",
+    "4": "ADJ|l-ccomp",
+    "5": "ADJ|l-csubj",
+    "6": "ADJ|l-csubj:outer",
+    "7": "ADJ|l-nmod",
+    "8": "ADJ|l-nsubj",
+    "9": "ADJ|l-obj",
+    "10": "ADJ|l-obl",
+    "11": "ADJ|r-acl",
+    "12": "ADJ|r-amod",
+    "13": "ADJ|r-dep",
+    "14": "ADJ|root",
+    "15": "ADP",
+    "16": "ADP|l-case",
+    "17": "ADP|r-case",
+    "18": "ADP|r-fixed",
+    "19": "ADV",
+    "20": "ADV|l-advcl",
+    "21": "ADV|l-advmod",
+    "22": "ADV|l-obj",
+    "23": "ADV|r-dep",
+    "24": "ADV|root",
+    "25": "AUX",
+    "26": "AUX|Polarity=Neg",
+    "27": "AUX|Polarity=Neg|r-aux",
+    "28": "AUX|Polarity=Neg|r-fixed",
+    "29": "AUX|r-aux",
+    "30": "AUX|r-cop",
+    "31": "AUX|r-fixed",
+    "32": "AUX|root",
+    "33": "B-ADJ",
+    "34": "B-ADP",
+    "35": "B-ADV",
+    "36": "B-AUX",
+    "37": "B-AUX|Polarity=Neg",
+    "38": "B-CCONJ",
+    "39": "B-DET",
+    "40": "B-INTJ",
+    "41": "B-NOUN",
+    "42": "B-NOUN|Polarity=Neg",
+    "43": "B-NUM",
+    "44": "B-PART",
+    "45": "B-PRON",
+    "46": "B-PROPN",
+    "47": "B-PUNCT",
+    "48": "B-SCONJ",
+    "49": "B-SYM",
+    "50": "B-VERB",
+    "51": "B-X",
+    "52": "CCONJ",
+    "53": "CCONJ|l-cc",
+    "54": "CCONJ|r-cc",
+    "55": "DET",
+    "56": "DET|l-det",
+    "57": "I-ADJ",
+    "58": "I-ADP",
+    "59": "I-ADV",
+    "60": "I-AUX",
+    "61": "I-AUX|Polarity=Neg",
+    "62": "I-CCONJ",
+    "63": "I-DET",
+    "64": "I-INTJ",
+    "65": "I-NOUN",
+    "66": "I-NOUN|Polarity=Neg",
+    "67": "I-NUM",
+    "68": "I-PART",
+    "69": "I-PRON",
+    "70": "I-PROPN",
+    "71": "I-PUNCT",
+    "72": "I-SCONJ",
+    "73": "I-SYM",
+    "74": "I-VERB",
+    "75": "I-X",
+    "76": "INTJ",
+    "77": "INTJ|l-discourse",
+    "78": "INTJ|r-discourse",
+    "79": "INTJ|root",
+    "80": "NOUN",
+    "81": "NOUN|Polarity=Neg",
+    "82": "NOUN|Polarity=Neg|l-obl",
+    "83": "NOUN|Polarity=Neg|root",
+    "84": "NOUN|l-acl",
+    "85": "NOUN|l-advcl",
+    "86": "NOUN|l-ccomp",
+    "87": "NOUN|l-compound",
+    "88": "NOUN|l-csubj",
+    "89": "NOUN|l-csubj:outer",
+    "90": "NOUN|l-nmod",
+    "91": "NOUN|l-nsubj",
+    "92": "NOUN|l-nsubj:outer",
+    "93": "NOUN|l-obj",
+    "94": "NOUN|l-obl",
+    "95": "NOUN|r-compound",
+    "96": "NOUN|r-nmod",
+    "97": "NOUN|r-nsubj",
+    "98": "NOUN|root",
+    "99": "NUM",
+    "100": "NUM|l-advcl",
+    "101": "NUM|l-compound",
+    "102": "NUM|l-nmod",
+    "103": "NUM|l-nsubj",
+    "104": "NUM|l-nsubj:outer",
+    "105": "NUM|l-nummod",
+    "106": "NUM|l-obj",
+    "107": "NUM|l-obl",
+    "108": "NUM|r-compound",
+    "109": "NUM|root",
+    "110": "PART",
+    "111": "PART|l-mark",
+    "112": "PART|r-mark",
+    "113": "PRON",
+    "114": "PRON|l-acl",
+    "115": "PRON|l-advcl",
+    "116": "PRON|l-nmod",
+    "117": "PRON|l-nsubj",
+    "118": "PRON|l-nsubj:outer",
+    "119": "PRON|l-obj",
+    "120": "PRON|l-obl",
+    "121": "PRON|root",
+    "122": "PROPN",
+    "123": "PROPN|l-acl",
+    "124": "PROPN|l-advcl",
+    "125": "PROPN|l-compound",
+    "126": "PROPN|l-nmod",
+    "127": "PROPN|l-nsubj",
+    "128": "PROPN|l-nsubj:outer",
+    "129": "PROPN|l-obj",
+    "130": "PROPN|l-obl",
+    "131": "PROPN|r-compound",
+    "132": "PROPN|r-nmod",
+    "133": "PROPN|root",
+    "134": "PUNCT",
+    "135": "PUNCT|l-punct",
+    "136": "PUNCT|r-punct",
+    "137": "SCONJ",
+    "138": "SCONJ|l-dep",
+    "139": "SCONJ|r-fixed",
+    "140": "SCONJ|r-mark",
+    "141": "SYM",
+    "142": "SYM|l-compound",
+    "143": "SYM|l-dep",
+    "144": "SYM|l-nmod",
+    "145": "SYM|l-obl",
+    "146": "SYM|r-compound",
+    "147": "SYM|r-dep",
+    "148": "VERB",
+    "149": "VERB|l-acl",
+    "150": "VERB|l-advcl",
+    "151": "VERB|l-ccomp",
+    "152": "VERB|l-compound",
+    "153": "VERB|l-csubj",
+    "154": "VERB|l-csubj:outer",
+    "155": "VERB|l-nmod",
+    "156": "VERB|l-obj",
+    "157": "VERB|l-obl",
+    "158": "VERB|r-acl",
+    "159": "VERB|r-advcl",
+    "160": "VERB|r-compound",
+    "161": "VERB|root",
+    "162": "X",
+    "163": "X|l-nmod",
+    "164": "X|r-dep"
+  },
+  "initializer_range": 0.02,
+  "label2id": {
+    "ADJ": 0,
+    "ADJ|l-acl": 1,
+    "ADJ|l-advcl": 2,
+    "ADJ|l-amod": 3,
+    "ADJ|l-ccomp": 4,
+    "ADJ|l-csubj": 5,
+    "ADJ|l-csubj:outer": 6,
+    "ADJ|l-nmod": 7,
+    "ADJ|l-nsubj": 8,
+    "ADJ|l-obj": 9,
+    "ADJ|l-obl": 10,
+    "ADJ|r-acl": 11,
+    "ADJ|r-amod": 12,
+    "ADJ|r-dep": 13,
+    "ADJ|root": 14,
+    "ADP": 15,
+    "ADP|l-case": 16,
+    "ADP|r-case": 17,
+    "ADP|r-fixed": 18,
+    "ADV": 19,
+    "ADV|l-advcl": 20,
+    "ADV|l-advmod": 21,
+    "ADV|l-obj": 22,
+    "ADV|r-dep": 23,
+    "ADV|root": 24,
+    "AUX": 25,
+    "AUX|Polarity=Neg": 26,
+    "AUX|Polarity=Neg|r-aux": 27,
+    "AUX|Polarity=Neg|r-fixed": 28,
+    "AUX|r-aux": 29,
+    "AUX|r-cop": 30,
+    "AUX|r-fixed": 31,
+    "AUX|root": 32,
+    "B-ADJ": 33,
+    "B-ADP": 34,
+    "B-ADV": 35,
+    "B-AUX": 36,
+    "B-AUX|Polarity=Neg": 37,
+    "B-CCONJ": 38,
+    "B-DET": 39,
+    "B-INTJ": 40,
+    "B-NOUN": 41,
+    "B-NOUN|Polarity=Neg": 42,
+    "B-NUM": 43,
+    "B-PART": 44,
+    "B-PRON": 45,
+    "B-PROPN": 46,
+    "B-PUNCT": 47,
+    "B-SCONJ": 48,
+    "B-SYM": 49,
+    "B-VERB": 50,
+    "B-X": 51,
+    "CCONJ": 52,
+    "CCONJ|l-cc": 53,
+    "CCONJ|r-cc": 54,
+    "DET": 55,
+    "DET|l-det": 56,
+    "I-ADJ": 57,
+    "I-ADP": 58,
+    "I-ADV": 59,
+    "I-AUX": 60,
+    "I-AUX|Polarity=Neg": 61,
+    "I-CCONJ": 62,
+    "I-DET": 63,
+    "I-INTJ": 64,
+    "I-NOUN": 65,
+    "I-NOUN|Polarity=Neg": 66,
+    "I-NUM": 67,
+    "I-PART": 68,
+    "I-PRON": 69,
+    "I-PROPN": 70,
+    "I-PUNCT": 71,
+    "I-SCONJ": 72,
+    "I-SYM": 73,
+    "I-VERB": 74,
+    "I-X": 75,
+    "INTJ": 76,
+    "INTJ|l-discourse": 77,
+    "INTJ|r-discourse": 78,
+    "INTJ|root": 79,
+    "NOUN": 80,
+    "NOUN|Polarity=Neg": 81,
+    "NOUN|Polarity=Neg|l-obl": 82,
+    "NOUN|Polarity=Neg|root": 83,
+    "NOUN|l-acl": 84,
+    "NOUN|l-advcl": 85,
+    "NOUN|l-ccomp": 86,
+    "NOUN|l-compound": 87,
+    "NOUN|l-csubj": 88,
+    "NOUN|l-csubj:outer": 89,
+    "NOUN|l-nmod": 90,
+    "NOUN|l-nsubj": 91,
+    "NOUN|l-nsubj:outer": 92,
+    "NOUN|l-obj": 93,
+    "NOUN|l-obl": 94,
+    "NOUN|r-compound": 95,
+    "NOUN|r-nmod": 96,
+    "NOUN|r-nsubj": 97,
+    "NOUN|root": 98,
+    "NUM": 99,
+    "NUM|l-advcl": 100,
+    "NUM|l-compound": 101,
+    "NUM|l-nmod": 102,
+    "NUM|l-nsubj": 103,
+    "NUM|l-nsubj:outer": 104,
+    "NUM|l-nummod": 105,
+    "NUM|l-obj": 106,
+    "NUM|l-obl": 107,
+    "NUM|r-compound": 108,
+    "NUM|root": 109,
+    "PART": 110,
+    "PART|l-mark": 111,
+    "PART|r-mark": 112,
+    "PRON": 113,
+    "PRON|l-acl": 114,
+    "PRON|l-advcl": 115,
+    "PRON|l-nmod": 116,
+    "PRON|l-nsubj": 117,
+    "PRON|l-nsubj:outer": 118,
+    "PRON|l-obj": 119,
+    "PRON|l-obl": 120,
+    "PRON|root": 121,
+    "PROPN": 122,
+    "PROPN|l-acl": 123,
+    "PROPN|l-advcl": 124,
+    "PROPN|l-compound": 125,
+    "PROPN|l-nmod": 126,
+    "PROPN|l-nsubj": 127,
+    "PROPN|l-nsubj:outer": 128,
+    "PROPN|l-obj": 129,
+    "PROPN|l-obl": 130,
+    "PROPN|r-compound": 131,
+    "PROPN|r-nmod": 132,
+    "PROPN|root": 133,
+    "PUNCT": 134,
+    "PUNCT|l-punct": 135,
+    "PUNCT|r-punct": 136,
+    "SCONJ": 137,
+    "SCONJ|l-dep": 138,
+    "SCONJ|r-fixed": 139,
+    "SCONJ|r-mark": 140,
+    "SYM": 141,
+    "SYM|l-compound": 142,
+    "SYM|l-dep": 143,
+    "SYM|l-nmod": 144,
+    "SYM|l-obl": 145,
+    "SYM|r-compound": 146,
+    "SYM|r-dep": 147,
+    "VERB": 148,
+    "VERB|l-acl": 149,
+    "VERB|l-advcl": 150,
+    "VERB|l-ccomp": 151,
+    "VERB|l-compound": 152,
+    "VERB|l-csubj": 153,
+    "VERB|l-csubj:outer": 154,
+    "VERB|l-nmod": 155,
+    "VERB|l-obj": 156,
+    "VERB|l-obl": 157,
+    "VERB|r-acl": 158,
+    "VERB|r-advcl": 159,
+    "VERB|r-compound": 160,
+    "VERB|root": 161,
+    "X": 162,
+    "X|l-nmod": 163,
+    "X|r-dep": 164
+  },
+  "layer_norm_epsilon": 1e-05,
+  "model_type": "gpt2",
+  "n_ctx": 512,
+  "n_embd": 768,
+  "n_head": 12,
+  "n_inner": 3072,
+  "n_layer": 12,
+  "n_positions": 512,
+  "pad_token_id": 50002,
+  "prefix": "[CLS]",
+  "reorder_and_upcast_attn": false,
+  "resid_pdrop": 0.1,
+  "scale_attn_by_inverse_layer_idx": false,
+  "scale_attn_weights": true,
+  "summary_activation": null,
+  "summary_first_dropout": 0.1,
+  "summary_proj_to_labels": true,
+  "summary_type": "cls_index",
+  "summary_use_proj": true,
+  "tokenizer_class": "PreTrainedTokenizerFast",
+  "torch_dtype": "float32",
+  "transformers_version": "4.42.4",
+  "use_cache": true,
+  "vocab_size": 51200
+}

maker.py

@@ -0,0 +1,128 @@
+#! /usr/bin/python3
+src="goldfish-models/jpn_jpan_1000mb"
+tgt="KoichiYasuoka/goldfish-gpt2-japanese-1000mb-ud-causal"
+url="https://github.com/UniversalDependencies/UD_Japanese-GSDLUW"
+
+import os,json
+from transformers import AutoTokenizer,PreTrainedTokenizerFast,AutoConfig,GPT2ForTokenClassification,DefaultDataCollator,TrainingArguments,Trainer
+from tokenizers import pre_tokenizers,decoders
+d=os.path.basename(url)
+os.system("test -d "+d+" || git clone --depth=1 "+url)
+os.system("for F in train dev test ; do cp "+d+"/*-$F.conllu $F.conllu ; done")
+tkz=AutoTokenizer.from_pretrained(src,add_prefix_space=False,legacy=False,model_max_length=768)
+tkz.backend_tokenizer.pre_tokenizer=pre_tokenizers.Metaspace(prepend_scheme="never")
+tkz.backend_tokenizer.decoder=decoders.Metaspace(prepend_scheme="never")
+tkz.save_pretrained("tmpdir")
+d=json.loads(tkz.backend_tokenizer.to_str())
+form=set()
+with open("train.conllu","r",encoding="utf-8") as r:
+  for s in r:
+    w=s.split("\t")
+    if len(w)==10 and w[0].isdecimal():
+      form.add(w[1])
+for t in d["model"]["vocab"]:
+  if t[0] not in form:
+    t[1]*=len(t[0])
+tkz.backend_tokenizer.from_str(json.dumps(d)).save("tmpdir/tokenizer.json")
+tkz=PreTrainedTokenizerFast.from_pretrained("tmpdir")
+
+class UDCausalDataset(object):
+  def __init__(self,conllu,tokenizer,embeddings=None):
+    self.conllu=open(conllu,"r",encoding="utf-8")
+    self.tokenizer=tokenizer
+    self.embeddings=embeddings
+    self.max_tokens=3
+    self.seeks=[(0,0)]
+    label=set(["SYM"])
+    dep=set()
+    s=self.conllu.readline()
+    while s!="":
+      if s=="\n":
+        self.seeks.append((self.conllu.tell(),0))
+      else:
+        w=s.split("\t")
+        if len(w)==10:
+          if w[0].isdecimal():
+            p=w[3] if w[5]=="_" else w[3]+"|"+w[5]
+            label.add(p)
+            dep.add(p+("|" if w[6]=="0" else "|l-" if int(w[0])<int(w[6]) else "|r-")+w[7])
+            self.seeks.append((self.seeks[-1][0],int(w[0])))
+            self.max_tokens=max(self.max_tokens,int(w[0])*2+1)
+      s=self.conllu.readline()
+    lid={}
+    for i,l in enumerate(sorted(label)):
+      lid[l],lid["B-"+l],lid["I-"+l]=i*3,i*3+1,i*3+2
+    for i,d in enumerate(sorted(dep),len(lid)):
+      lid[d]=i
+    self.label2id=lid
+  def __call__(*args):
+    lid={l:i for i,l in enumerate(sorted(set(sum([list(t.label2id) for t in args],[]))))}
+    for t in args:
+      t.label2id=lid
+    return lid
+  def __del__(self):
+    self.conllu.close()
+  __len__=lambda self:len(self.seeks)-1
+  def __getitem__(self,i):
+    s,t=self.seeks[i]
+    self.conllu.seek(s)
+    form,upos,deps,w=[],[],[],[""]
+    while w[0]!="\n":
+      w=self.conllu.readline().split("\t")
+      if len(w)==10:
+        form.append(w[1])
+        if w[0].isdecimal():
+          upos.append(w[3] if w[5]=="_" else w[3]+"|"+w[5])
+          deps.append((int(w[6]),w[7]))
+    v=self.tokenizer(form,add_special_tokens=False)
+    if t==0:
+      i,u=[self.tokenizer.cls_token_id],["SYM"]
+      for j,(x,y) in enumerate(zip(v["input_ids"],upos)):
+        if x!=[]:
+          i+=x
+          u+=[y] if len(x)==1 else ["B-"+y]+["I-"+y]*(len(x)-1)
+      emb=self.embeddings
+      pad=self.tokenizer.pad_token_id
+    else:
+      import torch
+      m=[]
+      for x in v["input_ids"]:
+        if x==[]:
+          m.append(self.embeddings[self.tokenizer.unk_token_id,:])
+        else:
+          m.append(self.embeddings[x,:].sum(axis=0))
+      m.append(self.embeddings[self.tokenizer.sep_token_id,:])
+      m.append(self.embeddings[self.tokenizer.pad_token_id,:])
+      m.append(self.embeddings[self.tokenizer.cls_token_id,:])
+      emb=torch.stack(m)
+      i,u=list(range(-1,len(upos)+1)),["SYM"]+upos+["SYM"]
+      i.append(t-1)
+      k,d=deps[t-1]
+      u.append(upos[t-1]+"|"+d if k==0 else upos[t-1])
+      for j in range(t,len(upos)):
+        i.append(j)
+        a,b=deps[j]
+        u.append(upos[j]+"|r-"+b if a==t else upos[t-1]+"|l-"+d if j+1==k else upos[j])
+      pad=-2
+    j=self.max_tokens-len(i)
+    if j>0:
+      ids=i+[pad]*j
+      upos=u+["SYM"]*j
+    else:
+      ids=i[0:self.max_tokens]
+      upos=u[0:self.max_tokens]
+    return {"inputs_embeds":emb[ids,:],"labels":[self.label2id[p] for p in upos]}
+
+trainDS=UDCausalDataset("train.conllu",tkz)
+devDS=UDCausalDataset("dev.conllu",tkz)
+testDS=UDCausalDataset("test.conllu",tkz)
+lid=trainDS(devDS,testDS)
+cfg=AutoConfig.from_pretrained(src,num_labels=len(lid),label2id=lid,id2label={i:l for l,i in lid.items()},ignore_mismatched_sizes=True)
+mdl=GPT2ForTokenClassification.from_pretrained(src,config=cfg,ignore_mismatched_sizes=True)
+trainDS.embeddings=mdl.get_input_embeddings().weight
+trainDS.max_tokens=min(trainDS.max_tokens,cfg.max_position_embeddings)
+arg=TrainingArguments(num_train_epochs=3,per_device_train_batch_size=32,dataloader_pin_memory=False,output_dir=tgt,overwrite_output_dir=True,save_total_limit=2,learning_rate=5e-05,warmup_ratio=0.1,save_safetensors=False)
+trn=Trainer(args=arg,data_collator=DefaultDataCollator(),model=mdl,train_dataset=trainDS)
+trn.train()
+trn.save_model(tgt)
+tkz.save_pretrained(tgt)

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23bae96766c8421ee8f7868d4e78c7c70a6534c1e6b759470397aa5d7ff81134
+size 499639714

special_tokens_map.json

@@ -0,0 +1,1249 @@
+{
+  "additional_special_tokens": [
+    "[XXXXX0]",
+    "[XXXXX1]",
+    "[XXXXX2]",
+    "[XXXXX3]",
+    "[XXXXX4]",
+    "[XXXXX5]",
+    "[XXXXX6]",
+    "[XXXXX7]",
+    "[XXXXX8]",
+    "[XXXXX9]",
+    "[XXXXX10]",
+    "[XXXXX11]",
+    "[XXXXX12]",
+    "[XXXXX13]",
+    "[XXXXX14]",
+    "[XXXXX15]",
+    "[XXXXX16]",
+    "[XXXXX17]",
+    "[XXXXX18]",
+    "[XXXXX19]",
+    "[XXXXX20]",
+    "[XXXXX21]",
+    "[XXXXX22]",
+    "[XXXXX23]",
+    "[XXXXX24]",
+    "[XXXXX25]",
+    "[XXXXX26]",
+    "[XXXXX27]",
+    "[XXXXX28]",
+    "[XXXXX29]",
+    "[XXXXX30]",
+    "[XXXXX31]",
+    "[XXXXX32]",
+    "[XXXXX33]",
+    "[XXXXX34]",
+    "[XXXXX35]",
+    "[XXXXX36]",
+    "[XXXXX37]",
+    "[XXXXX38]",
+    "[XXXXX39]",
+    "[XXXXX40]",
+    "[XXXXX41]",
+    "[XXXXX42]",
+    "[XXXXX43]",
+    "[XXXXX44]",
+    "[XXXXX45]",
+    "[XXXXX46]",
+    "[XXXXX47]",
+    "[XXXXX48]",
+    "[XXXXX49]",
+    "[XXXXX50]",
+    "[XXXXX51]",
+    "[XXXXX52]",
+    "[XXXXX53]",
+    "[XXXXX54]",
+    "[XXXXX55]",
+    "[XXXXX56]",
+    "[XXXXX57]",
+    "[XXXXX58]",
+    "[XXXXX59]",
+    "[XXXXX60]",
+    "[XXXXX61]",
+    "[XXXXX62]",
+    "[XXXXX63]",
+    "[XXXXX64]",
+    "[XXXXX65]",
+    "[XXXXX66]",
+    "[XXXXX67]",
+    "[XXXXX68]",
+    "[XXXXX69]",
+    "[XXXXX70]",
+    "[XXXXX71]",
+    "[XXXXX72]",
+    "[XXXXX73]",
+    "[XXXXX74]",
+    "[XXXXX75]",
+    "[XXXXX76]",
+    "[XXXXX77]",
+    "[XXXXX78]",
+    "[XXXXX79]",
+    "[XXXXX80]",
+    "[XXXXX81]",
+    "[XXXXX82]",
+    "[XXXXX83]",
+    "[XXXXX84]",
+    "[XXXXX85]",
+    "[XXXXX86]",
+    "[XXXXX87]",
+    "[XXXXX88]",
+    "[XXXXX89]",
+    "[XXXXX90]",
+    "[XXXXX91]",
+    "[XXXXX92]",
+    "[XXXXX93]",
+    "[XXXXX94]",
+    "[XXXXX95]",
+    "[XXXXX96]",
+    "[XXXXX97]",
+    "[XXXXX98]",
+    "[XXXXX99]",
+    "[XXXXX100]",
+    "[XXXXX101]",
+    "[XXXXX102]",
+    "[XXXXX103]",
+    "[XXXXX104]",
+    "[XXXXX105]",
+    "[XXXXX106]",
+    "[XXXXX107]",
+    "[XXXXX108]",
+    "[XXXXX109]",
+    "[XXXXX110]",
+    "[XXXXX111]",
+    "[XXXXX112]",
+    "[XXXXX113]",
+    "[XXXXX114]",
+    "[XXXXX115]",
+    "[XXXXX116]",
+    "[XXXXX117]",
+    "[XXXXX118]",
+    "[XXXXX119]",
+    "[XXXXX120]",
+    "[XXXXX121]",
+    "[XXXXX122]",
+    "[XXXXX123]",
+    "[XXXXX124]",
+    "[XXXXX125]",
+    "[XXXXX126]",
+    "[XXXXX127]",
+    "[XXXXX128]",
+    "[XXXXX129]",
+    "[XXXXX130]",
+    "[XXXXX131]",
+    "[XXXXX132]",
+    "[XXXXX133]",
+    "[XXXXX134]",
+    "[XXXXX135]",
+    "[XXXXX136]",
+    "[XXXXX137]",
+    "[XXXXX138]",
+    "[XXXXX139]",
+    "[XXXXX140]",
+    "[XXXXX141]",
+    "[XXXXX142]",
+    "[XXXXX143]",
+    "[XXXXX144]",
+    "[XXXXX145]",
+    "[XXXXX146]",
+    "[XXXXX147]",
+    "[XXXXX148]",
+    "[XXXXX149]",
+    "[XXXXX150]",
+    "[XXXXX151]",
+    "[XXXXX152]",
+    "[XXXXX153]",
+    "[XXXXX154]",
+    "[XXXXX155]",
+    "[XXXXX156]",
+    "[XXXXX157]",
+    "[XXXXX158]",
+    "[XXXXX159]",
+    "[XXXXX160]",
+    "[XXXXX161]",
+    "[XXXXX162]",
+    "[XXXXX163]",
+    "[XXXXX164]",
+    "[XXXXX165]",
+    "[XXXXX166]",
+    "[XXXXX167]",
+    "[XXXXX168]",
+    "[XXXXX169]",
+    "[XXXXX170]",
+    "[XXXXX171]",
+    "[XXXXX172]",
+    "[XXXXX173]",
+    "[XXXXX174]",
+    "[XXXXX175]",
+    "[XXXXX176]",
+    "[XXXXX177]",
+    "[XXXXX178]",
+    "[XXXXX179]",
+    "[XXXXX180]",
+    "[XXXXX181]",
+    "[XXXXX182]",
+    "[XXXXX183]",
+    "[XXXXX184]",
+    "[XXXXX185]",
+    "[XXXXX186]",
+    "[XXXXX187]",
+    "[XXXXX188]",
+    "[XXXXX189]",
+    "[XXXXX190]",
+    "[XXXXX191]",
+    "[XXXXX192]",
+    "[XXXXX193]",
+    "[XXXXX194]",
+    "[XXXXX195]",
+    "[XXXXX196]",
+    "[XXXXX197]",
+    "[XXXXX198]",
+    "[XXXXX199]",
+    "[XXXXX200]",
+    "[XXXXX201]",
+    "[XXXXX202]",
+    "[XXXXX203]",
+    "[XXXXX204]",
+    "[XXXXX205]",
+    "[XXXXX206]",
+    "[XXXXX207]",
+    "[XXXXX208]",
+    "[XXXXX209]",
+    "[XXXXX210]",
+    "[XXXXX211]",
+    "[XXXXX212]",
+    "[XXXXX213]",
+    "[XXXXX214]",
+    "[XXXXX215]",
+    "[XXXXX216]",
+    "[XXXXX217]",
+    "[XXXXX218]",
+    "[XXXXX219]",
+    "[XXXXX220]",
+    "[XXXXX221]",
+    "[XXXXX222]",
+    "[XXXXX223]",
+    "[XXXXX224]",
+    "[XXXXX225]",
+    "[XXXXX226]",
+    "[XXXXX227]",
+    "[XXXXX228]",
+    "[XXXXX229]",
+    "[XXXXX230]",
+    "[XXXXX231]",
+    "[XXXXX232]",
+    "[XXXXX233]",
+    "[XXXXX234]",
+    "[XXXXX235]",
+    "[XXXXX236]",
+    "[XXXXX237]",
+    "[XXXXX238]",
+    "[XXXXX239]",
+    "[XXXXX240]",
+    "[XXXXX241]",
+    "[XXXXX242]",
+    "[XXXXX243]",
+    "[XXXXX244]",
+    "[XXXXX245]",
+    "[XXXXX246]",
+    "[XXXXX247]",
+    "[XXXXX248]",
+    "[XXXXX249]",
+    "[XXXXX250]",
+    "[XXXXX251]",
+    "[XXXXX252]",
+    "[XXXXX253]",
+    "[XXXXX254]",
+    "[XXXXX255]",
+    "[XXXXX256]",
+    "[XXXXX257]",
+    "[XXXXX258]",
+    "[XXXXX259]",
+    "[XXXXX260]",
+    "[XXXXX261]",
+    "[XXXXX262]",
+    "[XXXXX263]",
+    "[XXXXX264]",
+    "[XXXXX265]",
+    "[XXXXX266]",
+    "[XXXXX267]",
+    "[XXXXX268]",
+    "[XXXXX269]",
+    "[XXXXX270]",
+    "[XXXXX271]",
+    "[XXXXX272]",
+    "[XXXXX273]",
+    "[XXXXX274]",
+    "[XXXXX275]",
+    "[XXXXX276]",
+    "[XXXXX277]",
+    "[XXXXX278]",
+    "[XXXXX279]",
+    "[XXXXX280]",
+    "[XXXXX281]",
+    "[XXXXX282]",
+    "[XXXXX283]",
+    "[XXXXX284]",
+    "[XXXXX285]",
+    "[XXXXX286]",
+    "[XXXXX287]",
+    "[XXXXX288]",
+    "[XXXXX289]",
+    "[XXXXX290]",
+    "[XXXXX291]",
+    "[XXXXX292]",
+    "[XXXXX293]",
+    "[XXXXX294]",
+    "[XXXXX295]",
+    "[XXXXX296]",
+    "[XXXXX297]",
+    "[XXXXX298]",
+    "[XXXXX299]",
+    "[XXXXX300]",
+    "[XXXXX301]",
+    "[XXXXX302]",
+    "[XXXXX303]",
+    "[XXXXX304]",
+    "[XXXXX305]",
+    "[XXXXX306]",
+    "[XXXXX307]",
+    "[XXXXX308]",
+    "[XXXXX309]",
+    "[XXXXX310]",
+    "[XXXXX311]",
+    "[XXXXX312]",
+    "[XXXXX313]",
+    "[XXXXX314]",
+    "[XXXXX315]",
+    "[XXXXX316]",
+    "[XXXXX317]",
+    "[XXXXX318]",
+    "[XXXXX319]",
+    "[XXXXX320]",
+    "[XXXXX321]",
+    "[XXXXX322]",
+    "[XXXXX323]",
+    "[XXXXX324]",
+    "[XXXXX325]",
+    "[XXXXX326]",
+    "[XXXXX327]",
+    "[XXXXX328]",
+    "[XXXXX329]",
+    "[XXXXX330]",
+    "[XXXXX331]",
+    "[XXXXX332]",
+    "[XXXXX333]",
+    "[XXXXX334]",
+    "[XXXXX335]",
+    "[XXXXX336]",
+    "[XXXXX337]",
+    "[XXXXX338]",
+    "[XXXXX339]",
+    "[XXXXX340]",
+    "[XXXXX341]",
+    "[XXXXX342]",
+    "[XXXXX343]",
+    "[XXXXX344]",
+    "[XXXXX345]",
+    "[XXXXX346]",
+    "[XXXXX347]",
+    "[XXXXX348]",
+    "[XXXXX349]",
+    "[XXXXX350]",
+    "[XXXXX351]",
+    "[XXXXX352]",
+    "[XXXXX353]",
+    "[XXXXX354]",
+    "[XXXXX355]",
+    "[XXXXX356]",
+    "[XXXXX357]",
+    "[XXXXX358]",
+    "[XXXXX359]",
+    "[XXXXX360]",
+    "[XXXXX361]",
+    "[XXXXX362]",
+    "[XXXXX363]",
+    "[XXXXX364]",
+    "[XXXXX365]",
+    "[XXXXX366]",
+    "[XXXXX367]",
+    "[XXXXX368]",
+    "[XXXXX369]",
+    "[XXXXX370]",
+    "[XXXXX371]",
+    "[XXXXX372]",
+    "[XXXXX373]",
+    "[XXXXX374]",
+    "[XXXXX375]",
+    "[XXXXX376]",
+    "[XXXXX377]",
+    "[XXXXX378]",
+    "[XXXXX379]",
+    "[XXXXX380]",
+    "[XXXXX381]",
+    "[XXXXX382]",
+    "[XXXXX383]",
+    "[XXXXX384]",
+    "[XXXXX385]",
+    "[XXXXX386]",
+    "[XXXXX387]",
+    "[XXXXX388]",
+    "[XXXXX389]",
+    "[XXXXX390]",
+    "[XXXXX391]",
+    "[XXXXX392]",
+    "[XXXXX393]",
+    "[XXXXX394]",
+    "[XXXXX395]",
+    "[XXXXX396]",
+    "[XXXXX397]",
+    "[XXXXX398]",
+    "[XXXXX399]",
+    "[XXXXX400]",
+    "[XXXXX401]",
+    "[XXXXX402]",
+    "[XXXXX403]",
+    "[XXXXX404]",
+    "[XXXXX405]",
+    "[XXXXX406]",
+    "[XXXXX407]",
+    "[XXXXX408]",
+    "[XXXXX409]",
+    "[XXXXX410]",
+    "[XXXXX411]",
+    "[XXXXX412]",
+    "[XXXXX413]",
+    "[XXXXX414]",
+    "[XXXXX415]",
+    "[XXXXX416]",
+    "[XXXXX417]",
+    "[XXXXX418]",
+    "[XXXXX419]",
+    "[XXXXX420]",
+    "[XXXXX421]",
+    "[XXXXX422]",
+    "[XXXXX423]",
+    "[XXXXX424]",
+    "[XXXXX425]",
+    "[XXXXX426]",
+    "[XXXXX427]",
+    "[XXXXX428]",
+    "[XXXXX429]",
+    "[XXXXX430]",
+    "[XXXXX431]",
+    "[XXXXX432]",
+    "[XXXXX433]",
+    "[XXXXX434]",
+    "[XXXXX435]",
+    "[XXXXX436]",
+    "[XXXXX437]",
+    "[XXXXX438]",
+    "[XXXXX439]",
+    "[XXXXX440]",
+    "[XXXXX441]",
+    "[XXXXX442]",
+    "[XXXXX443]",
+    "[XXXXX444]",
+    "[XXXXX445]",
+    "[XXXXX446]",
+    "[XXXXX447]",
+    "[XXXXX448]",
+    "[XXXXX449]",
+    "[XXXXX450]",
+    "[XXXXX451]",
+    "[XXXXX452]",
+    "[XXXXX453]",
+    "[XXXXX454]",
+    "[XXXXX455]",
+    "[XXXXX456]",
+    "[XXXXX457]",
+    "[XXXXX458]",
+    "[XXXXX459]",
+    "[XXXXX460]",
+    "[XXXXX461]",
+    "[XXXXX462]",
+    "[XXXXX463]",
+    "[XXXXX464]",
+    "[XXXXX465]",
+    "[XXXXX466]",
+    "[XXXXX467]",
+    "[XXXXX468]",
+    "[XXXXX469]",
+    "[XXXXX470]",
+    "[XXXXX471]",
+    "[XXXXX472]",
+    "[XXXXX473]",
+    "[XXXXX474]",
+    "[XXXXX475]",
+    "[XXXXX476]",
+    "[XXXXX477]",
+    "[XXXXX478]",
+    "[XXXXX479]",
+    "[XXXXX480]",
+    "[XXXXX481]",
+    "[XXXXX482]",
+    "[XXXXX483]",
+    "[XXXXX484]",
+    "[XXXXX485]",
+    "[XXXXX486]",
+    "[XXXXX487]",
+    "[XXXXX488]",
+    "[XXXXX489]",
+    "[XXXXX490]",
+    "[XXXXX491]",
+    "[XXXXX492]",
+    "[XXXXX493]",
+    "[XXXXX494]",
+    "[XXXXX495]",
+    "[XXXXX496]",
+    "[XXXXX497]",
+    "[XXXXX498]",
+    "[XXXXX499]",
+    "[XXXXX500]",
+    "[XXXXX501]",
+    "[XXXXX502]",
+    "[XXXXX503]",
+    "[XXXXX504]",
+    "[XXXXX505]",
+    "[XXXXX506]",
+    "[XXXXX507]",
+    "[XXXXX508]",
+    "[XXXXX509]",
+    "[XXXXX510]",
+    "[XXXXX511]",
+    "[XXXXX512]",
+    "[XXXXX513]",
+    "[XXXXX514]",
+    "[XXXXX515]",
+    "[XXXXX516]",
+    "[XXXXX517]",
+    "[XXXXX518]",
+    "[XXXXX519]",
+    "[XXXXX520]",
+    "[XXXXX521]",
+    "[XXXXX522]",
+    "[XXXXX523]",
+    "[XXXXX524]",
+    "[XXXXX525]",
+    "[XXXXX526]",
+    "[XXXXX527]",
+    "[XXXXX528]",
+    "[XXXXX529]",
+    "[XXXXX530]",
+    "[XXXXX531]",
+    "[XXXXX532]",
+    "[XXXXX533]",
+    "[XXXXX534]",
+    "[XXXXX535]",
+    "[XXXXX536]",
+    "[XXXXX537]",
+    "[XXXXX538]",
+    "[XXXXX539]",
+    "[XXXXX540]",
+    "[XXXXX541]",
+    "[XXXXX542]",
+    "[XXXXX543]",
+    "[XXXXX544]",
+    "[XXXXX545]",
+    "[XXXXX546]",
+    "[XXXXX547]",
+    "[XXXXX548]",
+    "[XXXXX549]",
+    "[XXXXX550]",
+    "[XXXXX551]",
+    "[XXXXX552]",
+    "[XXXXX553]",
+    "[XXXXX554]",
+    "[XXXXX555]",
+    "[XXXXX556]",
+    "[XXXXX557]",
+    "[XXXXX558]",
+    "[XXXXX559]",
+    "[XXXXX560]",
+    "[XXXXX561]",
+    "[XXXXX562]",
+    "[XXXXX563]",
+    "[XXXXX564]",
+    "[XXXXX565]",
+    "[XXXXX566]",
+    "[XXXXX567]",
+    "[XXXXX568]",
+    "[XXXXX569]",
+    "[XXXXX570]",
+    "[XXXXX571]",
+    "[XXXXX572]",
+    "[XXXXX573]",
+    "[XXXXX574]",
+    "[XXXXX575]",
+    "[XXXXX576]",
+    "[XXXXX577]",
+    "[XXXXX578]",
+    "[XXXXX579]",
+    "[XXXXX580]",
+    "[XXXXX581]",
+    "[XXXXX582]",
+    "[XXXXX583]",
+    "[XXXXX584]",
+    "[XXXXX585]",
+    "[XXXXX586]",
+    "[XXXXX587]",
+    "[XXXXX588]",
+    "[XXXXX589]",
+    "[XXXXX590]",
+    "[XXXXX591]",
+    "[XXXXX592]",
+    "[XXXXX593]",
+    "[XXXXX594]",
+    "[XXXXX595]",
+    "[XXXXX596]",
+    "[XXXXX597]",
+    "[XXXXX598]",
+    "[XXXXX599]",
+    "[XXXXX600]",
+    "[XXXXX601]",
+    "[XXXXX602]",
+    "[XXXXX603]",
+    "[XXXXX604]",
+    "[XXXXX605]",
+    "[XXXXX606]",
+    "[XXXXX607]",
+    "[XXXXX608]",
+    "[XXXXX609]",
+    "[XXXXX610]",
+    "[XXXXX611]",
+    "[XXXXX612]",
+    "[XXXXX613]",
+    "[XXXXX614]",
+    "[XXXXX615]",
+    "[XXXXX616]",
+    "[XXXXX617]",
+    "[XXXXX618]",
+    "[XXXXX619]",
+    "[XXXXX620]",
+    "[XXXXX621]",
+    "[XXXXX622]",
+    "[XXXXX623]",
+    "[XXXXX624]",
+    "[XXXXX625]",
+    "[XXXXX626]",
+    "[XXXXX627]",
+    "[XXXXX628]",
+    "[XXXXX629]",
+    "[XXXXX630]",
+    "[XXXXX631]",
+    "[XXXXX632]",
+    "[XXXXX633]",
+    "[XXXXX634]",
+    "[XXXXX635]",
+    "[XXXXX636]",
+    "[XXXXX637]",
+    "[XXXXX638]",
+    "[XXXXX639]",
+    "[XXXXX640]",
+    "[XXXXX641]",
+    "[XXXXX642]",
+    "[XXXXX643]",
+    "[XXXXX644]",
+    "[XXXXX645]",
+    "[XXXXX646]",
+    "[XXXXX647]",
+    "[XXXXX648]",
+    "[XXXXX649]",
+    "[XXXXX650]",
+    "[XXXXX651]",
+    "[XXXXX652]",
+    "[XXXXX653]",
+    "[XXXXX654]",
+    "[XXXXX655]",
+    "[XXXXX656]",
+    "[XXXXX657]",
+    "[XXXXX658]",
+    "[XXXXX659]",
+    "[XXXXX660]",
+    "[XXXXX661]",
+    "[XXXXX662]",
+    "[XXXXX663]",
+    "[XXXXX664]",
+    "[XXXXX665]",
+    "[XXXXX666]",
+    "[XXXXX667]",
+    "[XXXXX668]",
+    "[XXXXX669]",
+    "[XXXXX670]",
+    "[XXXXX671]",
+    "[XXXXX672]",
+    "[XXXXX673]",
+    "[XXXXX674]",
+    "[XXXXX675]",
+    "[XXXXX676]",
+    "[XXXXX677]",
+    "[XXXXX678]",
+    "[XXXXX679]",
+    "[XXXXX680]",
+    "[XXXXX681]",
+    "[XXXXX682]",
+    "[XXXXX683]",
+    "[XXXXX684]",
+    "[XXXXX685]",
+    "[XXXXX686]",
+    "[XXXXX687]",
+    "[XXXXX688]",
+    "[XXXXX689]",
+    "[XXXXX690]",
+    "[XXXXX691]",
+    "[XXXXX692]",
+    "[XXXXX693]",
+    "[XXXXX694]",
+    "[XXXXX695]",
+    "[XXXXX696]",
+    "[XXXXX697]",
+    "[XXXXX698]",
+    "[XXXXX699]",
+    "[XXXXX700]",
+    "[XXXXX701]",
+    "[XXXXX702]",
+    "[XXXXX703]",
+    "[XXXXX704]",
+    "[XXXXX705]",
+    "[XXXXX706]",
+    "[XXXXX707]",
+    "[XXXXX708]",
+    "[XXXXX709]",
+    "[XXXXX710]",
+    "[XXXXX711]",
+    "[XXXXX712]",
+    "[XXXXX713]",
+    "[XXXXX714]",
+    "[XXXXX715]",
+    "[XXXXX716]",
+    "[XXXXX717]",
+    "[XXXXX718]",
+    "[XXXXX719]",
+    "[XXXXX720]",
+    "[XXXXX721]",
+    "[XXXXX722]",
+    "[XXXXX723]",
+    "[XXXXX724]",
+    "[XXXXX725]",
+    "[XXXXX726]",
+    "[XXXXX727]",
+    "[XXXXX728]",
+    "[XXXXX729]",
+    "[XXXXX730]",
+    "[XXXXX731]",
+    "[XXXXX732]",
+    "[XXXXX733]",
+    "[XXXXX734]",
+    "[XXXXX735]",
+    "[XXXXX736]",
+    "[XXXXX737]",
+    "[XXXXX738]",
+    "[XXXXX739]",
+    "[XXXXX740]",
+    "[XXXXX741]",
+    "[XXXXX742]",
+    "[XXXXX743]",
+    "[XXXXX744]",
+    "[XXXXX745]",
+    "[XXXXX746]",
+    "[XXXXX747]",
+    "[XXXXX748]",
+    "[XXXXX749]",
+    "[XXXXX750]",
+    "[XXXXX751]",
+    "[XXXXX752]",
+    "[XXXXX753]",
+    "[XXXXX754]",
+    "[XXXXX755]",
+    "[XXXXX756]",
+    "[XXXXX757]",
+    "[XXXXX758]",
+    "[XXXXX759]",
+    "[XXXXX760]",
+    "[XXXXX761]",
+    "[XXXXX762]",
+    "[XXXXX763]",
+    "[XXXXX764]",
+    "[XXXXX765]",
+    "[XXXXX766]",
+    "[XXXXX767]",
+    "[XXXXX768]",
+    "[XXXXX769]",
+    "[XXXXX770]",
+    "[XXXXX771]",
+    "[XXXXX772]",
+    "[XXXXX773]",
+    "[XXXXX774]",
+    "[XXXXX775]",
+    "[XXXXX776]",
+    "[XXXXX777]",
+    "[XXXXX778]",
+    "[XXXXX779]",
+    "[XXXXX780]",
+    "[XXXXX781]",
+    "[XXXXX782]",
+    "[XXXXX783]",
+    "[XXXXX784]",
+    "[XXXXX785]",
+    "[XXXXX786]",
+    "[XXXXX787]",
+    "[XXXXX788]",
+    "[XXXXX789]",
+    "[XXXXX790]",
+    "[XXXXX791]",
+    "[XXXXX792]",
+    "[XXXXX793]",
+    "[XXXXX794]",
+    "[XXXXX795]",
+    "[XXXXX796]",
+    "[XXXXX797]",
+    "[XXXXX798]",
+    "[XXXXX799]",
+    "[XXXXX800]",
+    "[XXXXX801]",
+    "[XXXXX802]",
+    "[XXXXX803]",
+    "[XXXXX804]",
+    "[XXXXX805]",
+    "[XXXXX806]",
+    "[XXXXX807]",
+    "[XXXXX808]",
+    "[XXXXX809]",
+    "[XXXXX810]",
+    "[XXXXX811]",
+    "[XXXXX812]",
+    "[XXXXX813]",
+    "[XXXXX814]",
+    "[XXXXX815]",
+    "[XXXXX816]",
+    "[XXXXX817]",
+    "[XXXXX818]",
+    "[XXXXX819]",
+    "[XXXXX820]",
+    "[XXXXX821]",
+    "[XXXXX822]",
+    "[XXXXX823]",
+    "[XXXXX824]",
+    "[XXXXX825]",
+    "[XXXXX826]",
+    "[XXXXX827]",
+    "[XXXXX828]",
+    "[XXXXX829]",
+    "[XXXXX830]",
+    "[XXXXX831]",
+    "[XXXXX832]",
+    "[XXXXX833]",
+    "[XXXXX834]",
+    "[XXXXX835]",
+    "[XXXXX836]",
+    "[XXXXX837]",
+    "[XXXXX838]",
+    "[XXXXX839]",
+    "[XXXXX840]",
+    "[XXXXX841]",
+    "[XXXXX842]",
+    "[XXXXX843]",
+    "[XXXXX844]",
+    "[XXXXX845]",
+    "[XXXXX846]",
+    "[XXXXX847]",
+    "[XXXXX848]",
+    "[XXXXX849]",
+    "[XXXXX850]",
+    "[XXXXX851]",
+    "[XXXXX852]",
+    "[XXXXX853]",
+    "[XXXXX854]",
+    "[XXXXX855]",
+    "[XXXXX856]",
+    "[XXXXX857]",
+    "[XXXXX858]",
+    "[XXXXX859]",
+    "[XXXXX860]",
+    "[XXXXX861]",
+    "[XXXXX862]",
+    "[XXXXX863]",
+    "[XXXXX864]",
+    "[XXXXX865]",
+    "[XXXXX866]",
+    "[XXXXX867]",
+    "[XXXXX868]",
+    "[XXXXX869]",
+    "[XXXXX870]",
+    "[XXXXX871]",
+    "[XXXXX872]",
+    "[XXXXX873]",
+    "[XXXXX874]",
+    "[XXXXX875]",
+    "[XXXXX876]",
+    "[XXXXX877]",
+    "[XXXXX878]",
+    "[XXXXX879]",
+    "[XXXXX880]",
+    "[XXXXX881]",
+    "[XXXXX882]",
+    "[XXXXX883]",
+    "[XXXXX884]",
+    "[XXXXX885]",
+    "[XXXXX886]",
+    "[XXXXX887]",
+    "[XXXXX888]",
+    "[XXXXX889]",
+    "[XXXXX890]",
+    "[XXXXX891]",
+    "[XXXXX892]",
+    "[XXXXX893]",
+    "[XXXXX894]",
+    "[XXXXX895]",
+    "[XXXXX896]",
+    "[XXXXX897]",
+    "[XXXXX898]",
+    "[XXXXX899]",
+    "[XXXXX900]",
+    "[XXXXX901]",
+    "[XXXXX902]",
+    "[XXXXX903]",
+    "[XXXXX904]",
+    "[XXXXX905]",
+    "[XXXXX906]",
+    "[XXXXX907]",
+    "[XXXXX908]",
+    "[XXXXX909]",
+    "[XXXXX910]",
+    "[XXXXX911]",
+    "[XXXXX912]",
+    "[XXXXX913]",
+    "[XXXXX914]",
+    "[XXXXX915]",
+    "[XXXXX916]",
+    "[XXXXX917]",
+    "[XXXXX918]",
+    "[XXXXX919]",
+    "[XXXXX920]",
+    "[XXXXX921]",
+    "[XXXXX922]",
+    "[XXXXX923]",
+    "[XXXXX924]",
+    "[XXXXX925]",
+    "[XXXXX926]",
+    "[XXXXX927]",
+    "[XXXXX928]",
+    "[XXXXX929]",
+    "[XXXXX930]",
+    "[XXXXX931]",
+    "[XXXXX932]",
+    "[XXXXX933]",
+    "[XXXXX934]",
+    "[XXXXX935]",
+    "[XXXXX936]",
+    "[XXXXX937]",
+    "[XXXXX938]",
+    "[XXXXX939]",
+    "[XXXXX940]",
+    "[XXXXX941]",
+    "[XXXXX942]",
+    "[XXXXX943]",
+    "[XXXXX944]",
+    "[XXXXX945]",
+    "[XXXXX946]",
+    "[XXXXX947]",
+    "[XXXXX948]",
+    "[XXXXX949]",
+    "[XXXXX950]",
+    "[XXXXX951]",
+    "[XXXXX952]",
+    "[XXXXX953]",
+    "[XXXXX954]",
+    "[XXXXX955]",
+    "[XXXXX956]",
+    "[XXXXX957]",
+    "[XXXXX958]",
+    "[XXXXX959]",
+    "[XXXXX960]",
+    "[XXXXX961]",
+    "[XXXXX962]",
+    "[XXXXX963]",
+    "[XXXXX964]",
+    "[XXXXX965]",
+    "[XXXXX966]",
+    "[XXXXX967]",
+    "[XXXXX968]",
+    "[XXXXX969]",
+    "[XXXXX970]",
+    "[XXXXX971]",
+    "[XXXXX972]",
+    "[XXXXX973]",
+    "[XXXXX974]",
+    "[XXXXX975]",
+    "[XXXXX976]",
+    "[XXXXX977]",
+    "[XXXXX978]",
+    "[XXXXX979]",
+    "[XXXXX980]",
+    "[XXXXX981]",
+    "[XXXXX982]",
+    "[XXXXX983]",
+    "[XXXXX984]",
+    "[XXXXX985]",
+    "[XXXXX986]",
+    "[XXXXX987]",
+    "[XXXXX988]",
+    "[XXXXX989]",
+    "[XXXXX990]",
+    "[XXXXX991]",
+    "[XXXXX992]",
+    "[XXXXX993]",
+    "[XXXXX994]",
+    "[XXXXX995]",
+    "[XXXXX996]",
+    "[XXXXX997]",
+    "[XXXXX998]",
+    "[XXXXX999]",
+    "[XXXXX1000]",
+    "[XXXXX1001]",
+    "[XXXXX1002]",
+    "[XXXXX1003]",
+    "[XXXXX1004]",
+    "[XXXXX1005]",
+    "[XXXXX1006]",
+    "[XXXXX1007]",
+    "[XXXXX1008]",
+    "[XXXXX1009]",
+    "[XXXXX1010]",
+    "[XXXXX1011]",
+    "[XXXXX1012]",
+    "[XXXXX1013]",
+    "[XXXXX1014]",
+    "[XXXXX1015]",
+    "[XXXXX1016]",
+    "[XXXXX1017]",
+    "[XXXXX1018]",
+    "[XXXXX1019]",
+    "[XXXXX1020]",
+    "[XXXXX1021]",
+    "[XXXXX1022]",
+    "[XXXXX1023]",
+    "[XXXXX1024]",
+    "[XXXXX1025]",
+    "[XXXXX1026]",
+    "[XXXXX1027]",
+    "[XXXXX1028]",
+    "[XXXXX1029]",
+    "[XXXXX1030]",
+    "[XXXXX1031]",
+    "[XXXXX1032]",
+    "[XXXXX1033]",
+    "[XXXXX1034]",
+    "[XXXXX1035]",
+    "[XXXXX1036]",
+    "[XXXXX1037]",
+    "[XXXXX1038]",
+    "[XXXXX1039]",
+    "[XXXXX1040]",
+    "[XXXXX1041]",
+    "[XXXXX1042]",
+    "[XXXXX1043]",
+    "[XXXXX1044]",
+    "[XXXXX1045]",
+    "[XXXXX1046]",
+    "[XXXXX1047]",
+    "[XXXXX1048]",
+    "[XXXXX1049]",
+    "[XXXXX1050]",
+    "[XXXXX1051]",
+    "[XXXXX1052]",
+    "[XXXXX1053]",
+    "[XXXXX1054]",
+    "[XXXXX1055]",
+    "[XXXXX1056]",
+    "[XXXXX1057]",
+    "[XXXXX1058]",
+    "[XXXXX1059]",
+    "[XXXXX1060]",
+    "[XXXXX1061]",
+    "[XXXXX1062]",
+    "[XXXXX1063]",
+    "[XXXXX1064]",
+    "[XXXXX1065]",
+    "[XXXXX1066]",
+    "[XXXXX1067]",
+    "[XXXXX1068]",
+    "[XXXXX1069]",
+    "[XXXXX1070]",
+    "[XXXXX1071]",
+    "[XXXXX1072]",
+    "[XXXXX1073]",
+    "[XXXXX1074]",
+    "[XXXXX1075]",
+    "[XXXXX1076]",
+    "[XXXXX1077]",
+    "[XXXXX1078]",
+    "[XXXXX1079]",
+    "[XXXXX1080]",
+    "[XXXXX1081]",
+    "[XXXXX1082]",
+    "[XXXXX1083]",
+    "[XXXXX1084]",
+    "[XXXXX1085]",
+    "[XXXXX1086]",
+    "[XXXXX1087]",
+    "[XXXXX1088]",
+    "[XXXXX1089]",
+    "[XXXXX1090]",
+    "[XXXXX1091]",
+    "[XXXXX1092]",
+    "[XXXXX1093]",
+    "[XXXXX1094]",
+    "[XXXXX1095]",
+    "[XXXXX1096]",
+    "[XXXXX1097]",
+    "[XXXXX1098]",
+    "[XXXXX1099]",
+    "[XXXXX1100]",
+    "[XXXXX1101]",
+    "[XXXXX1102]",
+    "[XXXXX1103]",
+    "[XXXXX1104]",
+    "[XXXXX1105]",
+    "[XXXXX1106]",
+    "[XXXXX1107]",
+    "[XXXXX1108]",
+    "[XXXXX1109]",
+    "[XXXXX1110]",
+    "[XXXXX1111]",
+    "[XXXXX1112]",
+    "[XXXXX1113]",
+    "[XXXXX1114]",
+    "[XXXXX1115]",
+    "[XXXXX1116]",
+    "[XXXXX1117]",
+    "[XXXXX1118]",
+    "[XXXXX1119]",
+    "[XXXXX1120]",
+    "[XXXXX1121]",
+    "[XXXXX1122]",
+    "[XXXXX1123]",
+    "[XXXXX1124]",
+    "[XXXXX1125]",
+    "[XXXXX1126]",
+    "[XXXXX1127]",
+    "[XXXXX1128]",
+    "[XXXXX1129]",
+    "[XXXXX1130]",
+    "[XXXXX1131]",
+    "[XXXXX1132]",
+    "[XXXXX1133]",
+    "[XXXXX1134]",
+    "[XXXXX1135]",
+    "[XXXXX1136]",
+    "[XXXXX1137]",
+    "[XXXXX1138]",
+    "[XXXXX1139]",
+    "[XXXXX1140]",
+    "[XXXXX1141]",
+    "[XXXXX1142]",
+    "[XXXXX1143]",
+    "[XXXXX1144]",
+    "[XXXXX1145]",
+    "[XXXXX1146]",
+    "[XXXXX1147]",
+    "[XXXXX1148]",
+    "[XXXXX1149]",
+    "[XXXXX1150]",
+    "[XXXXX1151]",
+    "[XXXXX1152]",
+    "[XXXXX1153]",
+    "[XXXXX1154]",
+    "[XXXXX1155]",
+    "[XXXXX1156]",
+    "[XXXXX1157]",
+    "[XXXXX1158]",
+    "[XXXXX1159]",
+    "[XXXXX1160]",
+    "[XXXXX1161]",
+    "[XXXXX1162]",
+    "[XXXXX1163]",
+    "[XXXXX1164]",
+    "[XXXXX1165]",
+    "[XXXXX1166]",
+    "[XXXXX1167]",
+    "[XXXXX1168]",
+    "[XXXXX1169]",
+    "[XXXXX1170]",
+    "[XXXXX1171]",
+    "[XXXXX1172]",
+    "[XXXXX1173]",
+    "[XXXXX1174]",
+    "[XXXXX1175]",
+    "[XXXXX1176]",
+    "[XXXXX1177]",
+    "[XXXXX1178]",
+    "[XXXXX1179]",
+    "[XXXXX1180]",
+    "[XXXXX1181]",
+    "[XXXXX1182]",
+    "[XXXXX1183]",
+    "[XXXXX1184]",
+    "[XXXXX1185]",
+    "[XXXXX1186]",
+    "[XXXXX1187]",
+    "[XXXXX1188]",
+    "[XXXXX1189]",
+    "[XXXXX1190]",
+    "[XXXXX1191]",
+    "[XXXXX1192]",
+    "[XXXXX1193]",
+    "[XXXXX1194]",
+    "[XXXXX1195]"
+  ],
+  "bos_token": {
+    "content": "[CLS]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "cls_token": {
+    "content": "[CLS]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "[SEP]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "mask_token": {
+    "content": "[MASK]",
+    "lstrip": true,
+    "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
+  }
+}

ud.py

@@ -0,0 +1,142 @@
+import numpy
+from transformers import TokenClassificationPipeline
+
+class BellmanFordTokenClassificationPipeline(TokenClassificationPipeline):
+  def __init__(self,**kwargs):
+    super().__init__(**kwargs)
+    x=self.model.config.label2id
+    y=[k for k in x if k.startswith("B-") or not (k.startswith("I-") or k.endswith("|root") or k.find("|l-")>0 or k.find("|r-")>0)]
+    self.transition=numpy.full((len(x),len(x)),numpy.nan)
+    for k,v in x.items():
+      for j in ["I-"+k[2:]] if k.startswith("B-") else [k]+y if k.startswith("I-") else y:
+        self.transition[v,x[j]]=0
+  def check_model_type(self,supported_models):
+    pass
+  def postprocess(self,model_outputs,**kwargs):
+    if "logits" not in model_outputs:
+      return self.postprocess(model_outputs[0],**kwargs)
+    m=model_outputs["logits"][0].numpy()
+    e=numpy.exp(m-numpy.max(m,axis=-1,keepdims=True))
+    z=e/e.sum(axis=-1,keepdims=True)
+    for i in range(m.shape[0]-1,0,-1):
+      m[i-1]+=numpy.nanmax(m[i]+self.transition,axis=1)
+    k=[numpy.nanargmax(m[0]+self.transition[0])]
+    for i in range(1,m.shape[0]):
+      k.append(numpy.nanargmax(m[i]+self.transition[k[-1]]))
+    w=[{"entity":self.model.config.id2label[j],"start":s,"end":e,"score":z[i,j]} for i,((s,e),j) in enumerate(zip(model_outputs["offset_mapping"][0].tolist(),k)) if s<e]
+    if "aggregation_strategy" in kwargs and kwargs["aggregation_strategy"]!="none":
+      for i,t in reversed(list(enumerate(w))):
+        p=t.pop("entity")
+        if p.startswith("I-"):
+          w[i-1]["score"]=min(w[i-1]["score"],t["score"])
+          w[i-1]["end"]=w.pop(i)["end"]
+        elif p.startswith("B-"):
+          t["entity_group"]=p[2:]
+        else:
+          t["entity_group"]=p
+    for t in w:
+      t["text"]=model_outputs["sentence"][t["start"]:t["end"]]
+    return w
+
+class UniversalDependenciesCausalPipeline(BellmanFordTokenClassificationPipeline):
+  def __init__(self,**kwargs):
+    kwargs["aggregation_strategy"]="simple"
+    super().__init__(**kwargs)
+    x=self.model.config.label2id
+    self.root=numpy.full((len(x)),numpy.nan)
+    self.left_arc=numpy.full((len(x)),numpy.nan)
+    self.right_arc=numpy.full((len(x)),numpy.nan)
+    for k,v in x.items():
+      if k.endswith("|root"):
+        self.root[v]=0
+      elif k.find("|l-")>0:
+        self.left_arc[v]=0
+      elif k.find("|r-")>0:
+        self.right_arc[v]=0
+  def postprocess(self,model_outputs,**kwargs):
+    import torch
+    if "logits" not in model_outputs:
+      return self.postprocess(model_outputs[0],**kwargs)
+    m=model_outputs["logits"][0].numpy()
+    for i in range(m.shape[0]-1,0,-1):
+      m[i-1]+=numpy.nanmax(m[i]+self.transition,axis=1)
+    k=[numpy.nanargmax(m[0]+self.transition[0])]
+    for i in range(1,m.shape[0]):
+      k.append(numpy.nanargmax(m[i]+self.transition[k[-1]]))
+    w=[{"entity":self.model.config.id2label[j],"start":s,"end":e} for i,((s,e),j) in enumerate(zip(model_outputs["offset_mapping"][0].tolist(),k)) if s<e]
+    for i,t in reversed(list(enumerate(w))):
+      p=t.pop("entity")
+      if p.startswith("I-"):
+        w[i-1]["end"]=max(w.pop(i)["end"],w[i-1]["end"])
+      elif i>0 and w[i-1]["end"]>w[i]["start"]:
+        w[i-1]["end"]=max(w.pop(i)["end"],w[i-1]["end"])
+      elif p.startswith("B-"):
+        t["entity_group"]=p[2:]
+      else:
+        t["entity_group"]=p
+    d=[model_outputs["sentence"][t["start"]:t["end"]] for t in w]
+    for i in range(len(d)-1,-1,-1):
+      if d[i].startswith(" "):
+        j=len(d[i])-len(d[i].lstrip())
+        d[i]=d[i].lstrip()
+        w[i]["start"]+=j
+      if d[i].endswith(" "):
+        j=len(d[i])-len(d[i].rstrip())
+        d[i]=d[i].rstrip()
+        w[i]["end"]-=j
+      if d[i].strip()=="":
+        d.pop(i)
+        w.pop(i)
+    v=self.tokenizer(d,add_special_tokens=False)
+    e=self.model.get_input_embeddings().weight
+    m=[]
+    for x in v["input_ids"]:
+      if x==[]:
+        x=[self.tokenizer.unk_token_id]
+      m.append(e[x,:].sum(axis=0))
+    m.append(e[self.tokenizer.sep_token_id,:])
+    m.append(e[self.tokenizer.pad_token_id,:])
+    m.append(e[self.tokenizer.cls_token_id,:])
+    m=torch.stack(m).to(self.device)
+    k=list(range(-1,len(d)+1))
+    e=[]
+    with torch.no_grad():
+      for i in range(len(d)):
+        e.append(self.model(inputs_embeds=torch.unsqueeze(m[k+list(range(i,len(d)))+[-2]*i,:],0)).logits[0,-len(d):,:])
+    e=torch.stack(e).cpu().numpy()
+    for i in range(len(d)):
+      for j in range(i):
+        e[-j-1,-i-1],e[-i-1,-j-1]=e[-i-1,i-j]+self.left_arc,e[-i-1,i-j]+self.right_arc
+      e[-i-1,-i-1]=e[-i-1,0]+self.root
+    m,p=numpy.nanmax(e,axis=2),numpy.nanargmax(e,axis=2)
+    h=self.chu_liu_edmonds(m)
+    z=[i for i,j in enumerate(h) if i==j]
+    if len(z)>1:
+      k,h=z[numpy.nanargmax(m[z,z])],numpy.nanmin(m)-numpy.nanmax(m)
+      m[:,z]+=[[0 if j in z and (i!=j or i==k) else h for i in z] for j in range(m.shape[0])]
+      h=self.chu_liu_edmonds(m)
+    q=[self.model.config.id2label[p[j,i]].split("|") for i,j in enumerate(h)]
+    t=model_outputs["sentence"].replace("\n"," ")
+    u="# text = "+t+"\n"
+    for i,j in enumerate(d):
+      u+="\t".join([str(i+1),j,"_",q[i][0],"_","_" if len(q[i])<3 else "|".join(q[i][1:-1]),str(0 if h[i]==i else h[i]+1),"root" if q[i][-1]=="root" else q[i][-1][2:],"_","_" if i+1<len(d) and w[i]["end"]<w[i+1]["start"] else "SpaceAfter=No"])+"\n"
+    return u+"\n"
+  def chu_liu_edmonds(self,matrix):
+    h=numpy.nanargmax(matrix,axis=0)
+    x=[-1 if i==j else j for i,j in enumerate(h)]
+    for b in [lambda x,i,j:-1 if i not in x else x[i],lambda x,i,j:-1 if j<0 else x[j]]:
+      y=[]
+      while x!=y:
+        y=list(x)
+        for i,j in enumerate(x):
+          x[i]=b(x,i,j)
+      if max(x)<0:
+        return h
+    y,x=[i for i,j in enumerate(x) if j==max(x)],[i for i,j in enumerate(x) if j<max(x)]
+    z=matrix-numpy.nanmax(matrix,axis=0)
+    m=numpy.block([[z[x,:][:,x],numpy.nanmax(z[x,:][:,y],axis=1).reshape(len(x),1)],[numpy.nanmax(z[y,:][:,x],axis=0),numpy.nanmax(z[y,y])]])
+    k=[j if i==len(x) else x[j] if j<len(x) else y[numpy.nanargmax(z[y,x[i]])] for i,j in enumerate(self.chu_liu_edmonds(m))]
+    h=[j if i in y else k[x.index(i)] for i,j in enumerate(h)]
+    i=y[numpy.nanargmax(z[x[k[-1]],y] if k[-1]<len(x) else z[y,y])]
+    h[i]=x[k[-1]] if k[-1]<len(x) else i
+    return h