model improved

config.json

@@ -33,267 +33,521 @@
   "hidden_activation": "gelu",
   "hidden_size": 768,
   "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": "CCONJ",
-    "34": "CCONJ|l-cc",
-    "35": "CCONJ|r-cc",
-    "36": "DET",
-    "37": "DET|l-det",
-    "38": "INTJ",
-    "39": "INTJ|l-discourse",
-    "40": "INTJ|r-discourse",
-    "41": "INTJ|root",
-    "42": "NOUN",
-    "43": "NOUN|Polarity=Neg",
-    "44": "NOUN|Polarity=Neg|l-obl",
-    "45": "NOUN|Polarity=Neg|root",
-    "46": "NOUN|l-acl",
-    "47": "NOUN|l-advcl",
-    "48": "NOUN|l-ccomp",
-    "49": "NOUN|l-compound",
-    "50": "NOUN|l-csubj",
-    "51": "NOUN|l-csubj:outer",
-    "52": "NOUN|l-nmod",
-    "53": "NOUN|l-nsubj",
-    "54": "NOUN|l-nsubj:outer",
-    "55": "NOUN|l-obj",
-    "56": "NOUN|l-obl",
-    "57": "NOUN|r-compound",
-    "58": "NOUN|r-nmod",
-    "59": "NOUN|r-nsubj",
-    "60": "NOUN|root",
-    "61": "NUM",
-    "62": "NUM|l-advcl",
-    "63": "NUM|l-compound",
-    "64": "NUM|l-nmod",
-    "65": "NUM|l-nsubj",
-    "66": "NUM|l-nsubj:outer",
-    "67": "NUM|l-nummod",
-    "68": "NUM|l-obj",
-    "69": "NUM|l-obl",
-    "70": "NUM|r-compound",
-    "71": "NUM|root",
-    "72": "PART",
-    "73": "PART|l-mark",
-    "74": "PART|r-mark",
-    "75": "PRON",
-    "76": "PRON|l-acl",
-    "77": "PRON|l-advcl",
-    "78": "PRON|l-nmod",
-    "79": "PRON|l-nsubj",
-    "80": "PRON|l-nsubj:outer",
-    "81": "PRON|l-obj",
-    "82": "PRON|l-obl",
-    "83": "PRON|root",
-    "84": "PROPN",
-    "85": "PROPN|l-acl",
-    "86": "PROPN|l-advcl",
-    "87": "PROPN|l-compound",
-    "88": "PROPN|l-nmod",
-    "89": "PROPN|l-nsubj",
-    "90": "PROPN|l-nsubj:outer",
-    "91": "PROPN|l-obj",
-    "92": "PROPN|l-obl",
-    "93": "PROPN|r-compound",
-    "94": "PROPN|r-nmod",
-    "95": "PROPN|root",
-    "96": "PUNCT",
-    "97": "PUNCT|l-punct",
-    "98": "PUNCT|r-punct",
-    "99": "SCONJ",
-    "100": "SCONJ|l-dep",
-    "101": "SCONJ|r-fixed",
-    "102": "SCONJ|r-mark",
-    "103": "SYM",
-    "104": "SYM|l-compound",
-    "105": "SYM|l-dep",
-    "106": "SYM|l-nmod",
-    "107": "SYM|l-obl",
-    "108": "SYM|r-compound",
-    "109": "SYM|r-dep",
-    "110": "VERB",
-    "111": "VERB|l-acl",
-    "112": "VERB|l-advcl",
-    "113": "VERB|l-ccomp",
-    "114": "VERB|l-compound",
-    "115": "VERB|l-csubj",
-    "116": "VERB|l-csubj:outer",
-    "117": "VERB|l-nmod",
-    "118": "VERB|l-obj",
-    "119": "VERB|l-obl",
-    "120": "VERB|r-acl",
-    "121": "VERB|r-advcl",
-    "122": "VERB|r-compound",
-    "123": "VERB|root",
-    "124": "X",
-    "125": "X|l-nmod",
-    "126": "X|r-dep",
-    "127": "X|r-goeswith"
+    "0": "ADJ|o",
+    "1": "ADJ|o|l-acl",
+    "2": "ADJ|o|l-advcl",
+    "3": "ADJ|o|l-amod",
+    "4": "ADJ|o|l-ccomp",
+    "5": "ADJ|o|l-csubj",
+    "6": "ADJ|o|l-csubj:outer",
+    "7": "ADJ|o|l-nmod",
+    "8": "ADJ|o|l-nsubj",
+    "9": "ADJ|o|l-obj",
+    "10": "ADJ|o|l-obl",
+    "11": "ADJ|o|r-acl",
+    "12": "ADJ|o|r-amod",
+    "13": "ADJ|o|r-dep",
+    "14": "ADJ|o|root",
+    "15": "ADJ|x",
+    "16": "ADJ|x|l-acl",
+    "17": "ADJ|x|l-advcl",
+    "18": "ADJ|x|l-amod",
+    "19": "ADJ|x|l-ccomp",
+    "20": "ADJ|x|l-csubj",
+    "21": "ADJ|x|l-csubj:outer",
+    "22": "ADJ|x|l-nmod",
+    "23": "ADJ|x|l-nsubj",
+    "24": "ADJ|x|l-obj",
+    "25": "ADJ|x|l-obl",
+    "26": "ADJ|x|r-acl",
+    "27": "ADJ|x|r-amod",
+    "28": "ADJ|x|r-dep",
+    "29": "ADJ|x|root",
+    "30": "ADP|o",
+    "31": "ADP|o|l-case",
+    "32": "ADP|o|r-case",
+    "33": "ADP|o|r-fixed",
+    "34": "ADP|x",
+    "35": "ADP|x|l-case",
+    "36": "ADP|x|r-case",
+    "37": "ADP|x|r-fixed",
+    "38": "ADV|o",
+    "39": "ADV|o|l-advcl",
+    "40": "ADV|o|l-advmod",
+    "41": "ADV|o|l-obj",
+    "42": "ADV|o|r-dep",
+    "43": "ADV|o|root",
+    "44": "ADV|x",
+    "45": "ADV|x|l-advcl",
+    "46": "ADV|x|l-advmod",
+    "47": "ADV|x|l-obj",
+    "48": "ADV|x|r-dep",
+    "49": "ADV|x|root",
+    "50": "AUX|o",
+    "51": "AUX|o|Polarity=Neg",
+    "52": "AUX|o|Polarity=Neg|r-aux",
+    "53": "AUX|o|Polarity=Neg|r-fixed",
+    "54": "AUX|o|r-aux",
+    "55": "AUX|o|r-cop",
+    "56": "AUX|o|r-fixed",
+    "57": "AUX|o|root",
+    "58": "AUX|x",
+    "59": "AUX|x|Polarity=Neg",
+    "60": "AUX|x|Polarity=Neg|r-aux",
+    "61": "AUX|x|Polarity=Neg|r-fixed",
+    "62": "AUX|x|r-aux",
+    "63": "AUX|x|r-cop",
+    "64": "AUX|x|r-fixed",
+    "65": "AUX|x|root",
+    "66": "CCONJ|o",
+    "67": "CCONJ|o|l-cc",
+    "68": "CCONJ|o|r-cc",
+    "69": "CCONJ|x",
+    "70": "CCONJ|x|l-cc",
+    "71": "CCONJ|x|r-cc",
+    "72": "DET|o",
+    "73": "DET|o|l-det",
+    "74": "DET|x",
+    "75": "DET|x|l-det",
+    "76": "INTJ|o",
+    "77": "INTJ|o|l-discourse",
+    "78": "INTJ|o|r-discourse",
+    "79": "INTJ|o|root",
+    "80": "INTJ|x",
+    "81": "INTJ|x|l-discourse",
+    "82": "INTJ|x|r-discourse",
+    "83": "INTJ|x|root",
+    "84": "NOUN|o",
+    "85": "NOUN|o|Polarity=Neg",
+    "86": "NOUN|o|Polarity=Neg|l-obl",
+    "87": "NOUN|o|Polarity=Neg|root",
+    "88": "NOUN|o|l-acl",
+    "89": "NOUN|o|l-advcl",
+    "90": "NOUN|o|l-ccomp",
+    "91": "NOUN|o|l-compound",
+    "92": "NOUN|o|l-csubj",
+    "93": "NOUN|o|l-csubj:outer",
+    "94": "NOUN|o|l-nmod",
+    "95": "NOUN|o|l-nsubj",
+    "96": "NOUN|o|l-nsubj:outer",
+    "97": "NOUN|o|l-obj",
+    "98": "NOUN|o|l-obl",
+    "99": "NOUN|o|r-compound",
+    "100": "NOUN|o|r-nmod",
+    "101": "NOUN|o|r-nsubj",
+    "102": "NOUN|o|root",
+    "103": "NOUN|x",
+    "104": "NOUN|x|Polarity=Neg",
+    "105": "NOUN|x|Polarity=Neg|l-obl",
+    "106": "NOUN|x|Polarity=Neg|root",
+    "107": "NOUN|x|l-acl",
+    "108": "NOUN|x|l-advcl",
+    "109": "NOUN|x|l-ccomp",
+    "110": "NOUN|x|l-compound",
+    "111": "NOUN|x|l-csubj",
+    "112": "NOUN|x|l-csubj:outer",
+    "113": "NOUN|x|l-nmod",
+    "114": "NOUN|x|l-nsubj",
+    "115": "NOUN|x|l-nsubj:outer",
+    "116": "NOUN|x|l-obj",
+    "117": "NOUN|x|l-obl",
+    "118": "NOUN|x|r-compound",
+    "119": "NOUN|x|r-nmod",
+    "120": "NOUN|x|r-nsubj",
+    "121": "NOUN|x|root",
+    "122": "NUM|o",
+    "123": "NUM|o|l-advcl",
+    "124": "NUM|o|l-compound",
+    "125": "NUM|o|l-nmod",
+    "126": "NUM|o|l-nsubj",
+    "127": "NUM|o|l-nsubj:outer",
+    "128": "NUM|o|l-nummod",
+    "129": "NUM|o|l-obj",
+    "130": "NUM|o|l-obl",
+    "131": "NUM|o|r-compound",
+    "132": "NUM|o|root",
+    "133": "NUM|x",
+    "134": "NUM|x|l-advcl",
+    "135": "NUM|x|l-compound",
+    "136": "NUM|x|l-nmod",
+    "137": "NUM|x|l-nsubj",
+    "138": "NUM|x|l-nsubj:outer",
+    "139": "NUM|x|l-nummod",
+    "140": "NUM|x|l-obj",
+    "141": "NUM|x|l-obl",
+    "142": "NUM|x|r-compound",
+    "143": "NUM|x|root",
+    "144": "PART|o",
+    "145": "PART|o|l-mark",
+    "146": "PART|o|r-mark",
+    "147": "PART|x",
+    "148": "PART|x|l-mark",
+    "149": "PART|x|r-mark",
+    "150": "PRON|o",
+    "151": "PRON|o|l-acl",
+    "152": "PRON|o|l-advcl",
+    "153": "PRON|o|l-nmod",
+    "154": "PRON|o|l-nsubj",
+    "155": "PRON|o|l-nsubj:outer",
+    "156": "PRON|o|l-obj",
+    "157": "PRON|o|l-obl",
+    "158": "PRON|o|root",
+    "159": "PRON|x",
+    "160": "PRON|x|l-acl",
+    "161": "PRON|x|l-advcl",
+    "162": "PRON|x|l-nmod",
+    "163": "PRON|x|l-nsubj",
+    "164": "PRON|x|l-nsubj:outer",
+    "165": "PRON|x|l-obj",
+    "166": "PRON|x|l-obl",
+    "167": "PRON|x|root",
+    "168": "PROPN|o",
+    "169": "PROPN|o|l-acl",
+    "170": "PROPN|o|l-advcl",
+    "171": "PROPN|o|l-compound",
+    "172": "PROPN|o|l-nmod",
+    "173": "PROPN|o|l-nsubj",
+    "174": "PROPN|o|l-nsubj:outer",
+    "175": "PROPN|o|l-obj",
+    "176": "PROPN|o|l-obl",
+    "177": "PROPN|o|r-compound",
+    "178": "PROPN|o|r-nmod",
+    "179": "PROPN|o|root",
+    "180": "PROPN|x",
+    "181": "PROPN|x|l-acl",
+    "182": "PROPN|x|l-advcl",
+    "183": "PROPN|x|l-compound",
+    "184": "PROPN|x|l-nmod",
+    "185": "PROPN|x|l-nsubj",
+    "186": "PROPN|x|l-nsubj:outer",
+    "187": "PROPN|x|l-obj",
+    "188": "PROPN|x|l-obl",
+    "189": "PROPN|x|r-compound",
+    "190": "PROPN|x|r-nmod",
+    "191": "PROPN|x|root",
+    "192": "PUNCT|o",
+    "193": "PUNCT|o|l-punct",
+    "194": "PUNCT|o|r-punct",
+    "195": "PUNCT|x",
+    "196": "PUNCT|x|l-punct",
+    "197": "PUNCT|x|r-punct",
+    "198": "SCONJ|o",
+    "199": "SCONJ|o|l-dep",
+    "200": "SCONJ|o|r-fixed",
+    "201": "SCONJ|o|r-mark",
+    "202": "SCONJ|x",
+    "203": "SCONJ|x|l-dep",
+    "204": "SCONJ|x|r-fixed",
+    "205": "SCONJ|x|r-mark",
+    "206": "SYM|o",
+    "207": "SYM|o|l-compound",
+    "208": "SYM|o|l-dep",
+    "209": "SYM|o|l-nmod",
+    "210": "SYM|o|l-obl",
+    "211": "SYM|o|r-compound",
+    "212": "SYM|o|r-dep",
+    "213": "SYM|x",
+    "214": "SYM|x|l-compound",
+    "215": "SYM|x|l-dep",
+    "216": "SYM|x|l-nmod",
+    "217": "SYM|x|l-obl",
+    "218": "SYM|x|r-compound",
+    "219": "SYM|x|r-dep",
+    "220": "VERB|o",
+    "221": "VERB|o|l-acl",
+    "222": "VERB|o|l-advcl",
+    "223": "VERB|o|l-ccomp",
+    "224": "VERB|o|l-compound",
+    "225": "VERB|o|l-csubj",
+    "226": "VERB|o|l-csubj:outer",
+    "227": "VERB|o|l-nmod",
+    "228": "VERB|o|l-obj",
+    "229": "VERB|o|l-obl",
+    "230": "VERB|o|r-acl",
+    "231": "VERB|o|r-advcl",
+    "232": "VERB|o|r-compound",
+    "233": "VERB|o|root",
+    "234": "VERB|x",
+    "235": "VERB|x|l-acl",
+    "236": "VERB|x|l-advcl",
+    "237": "VERB|x|l-ccomp",
+    "238": "VERB|x|l-compound",
+    "239": "VERB|x|l-csubj",
+    "240": "VERB|x|l-csubj:outer",
+    "241": "VERB|x|l-nmod",
+    "242": "VERB|x|l-obj",
+    "243": "VERB|x|l-obl",
+    "244": "VERB|x|r-acl",
+    "245": "VERB|x|r-advcl",
+    "246": "VERB|x|r-compound",
+    "247": "VERB|x|root",
+    "248": "X|o",
+    "249": "X|o|l-nmod",
+    "250": "X|o|r-dep",
+    "251": "X|x",
+    "252": "X|x|l-nmod",
+    "253": "X|x|r-dep",
+    "254": "X|x|r-goeswith"
   },
   "initializer_cutoff_factor": 2.0,
   "initializer_range": 0.02,
   "intermediate_size": 1152,
   "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,
-    "CCONJ": 33,
-    "CCONJ|l-cc": 34,
-    "CCONJ|r-cc": 35,
-    "DET": 36,
-    "DET|l-det": 37,
-    "INTJ": 38,
-    "INTJ|l-discourse": 39,
-    "INTJ|r-discourse": 40,
-    "INTJ|root": 41,
-    "NOUN": 42,
-    "NOUN|Polarity=Neg": 43,
-    "NOUN|Polarity=Neg|l-obl": 44,
-    "NOUN|Polarity=Neg|root": 45,
-    "NOUN|l-acl": 46,
-    "NOUN|l-advcl": 47,
-    "NOUN|l-ccomp": 48,
-    "NOUN|l-compound": 49,
-    "NOUN|l-csubj": 50,
-    "NOUN|l-csubj:outer": 51,
-    "NOUN|l-nmod": 52,
-    "NOUN|l-nsubj": 53,
-    "NOUN|l-nsubj:outer": 54,
-    "NOUN|l-obj": 55,
-    "NOUN|l-obl": 56,
-    "NOUN|r-compound": 57,
-    "NOUN|r-nmod": 58,
-    "NOUN|r-nsubj": 59,
-    "NOUN|root": 60,
-    "NUM": 61,
-    "NUM|l-advcl": 62,
-    "NUM|l-compound": 63,
-    "NUM|l-nmod": 64,
-    "NUM|l-nsubj": 65,
-    "NUM|l-nsubj:outer": 66,
-    "NUM|l-nummod": 67,
-    "NUM|l-obj": 68,
-    "NUM|l-obl": 69,
-    "NUM|r-compound": 70,
-    "NUM|root": 71,
-    "PART": 72,
-    "PART|l-mark": 73,
-    "PART|r-mark": 74,
-    "PRON": 75,
-    "PRON|l-acl": 76,
-    "PRON|l-advcl": 77,
-    "PRON|l-nmod": 78,
-    "PRON|l-nsubj": 79,
-    "PRON|l-nsubj:outer": 80,
-    "PRON|l-obj": 81,
-    "PRON|l-obl": 82,
-    "PRON|root": 83,
-    "PROPN": 84,
-    "PROPN|l-acl": 85,
-    "PROPN|l-advcl": 86,
-    "PROPN|l-compound": 87,
-    "PROPN|l-nmod": 88,
-    "PROPN|l-nsubj": 89,
-    "PROPN|l-nsubj:outer": 90,
-    "PROPN|l-obj": 91,
-    "PROPN|l-obl": 92,
-    "PROPN|r-compound": 93,
-    "PROPN|r-nmod": 94,
-    "PROPN|root": 95,
-    "PUNCT": 96,
-    "PUNCT|l-punct": 97,
-    "PUNCT|r-punct": 98,
-    "SCONJ": 99,
-    "SCONJ|l-dep": 100,
-    "SCONJ|r-fixed": 101,
-    "SCONJ|r-mark": 102,
-    "SYM": 103,
-    "SYM|l-compound": 104,
-    "SYM|l-dep": 105,
-    "SYM|l-nmod": 106,
-    "SYM|l-obl": 107,
-    "SYM|r-compound": 108,
-    "SYM|r-dep": 109,
-    "VERB": 110,
-    "VERB|l-acl": 111,
-    "VERB|l-advcl": 112,
-    "VERB|l-ccomp": 113,
-    "VERB|l-compound": 114,
-    "VERB|l-csubj": 115,
-    "VERB|l-csubj:outer": 116,
-    "VERB|l-nmod": 117,
-    "VERB|l-obj": 118,
-    "VERB|l-obl": 119,
-    "VERB|r-acl": 120,
-    "VERB|r-advcl": 121,
-    "VERB|r-compound": 122,
-    "VERB|root": 123,
-    "X": 124,
-    "X|l-nmod": 125,
-    "X|r-dep": 126,
-    "X|r-goeswith": 127
+    "ADJ|o": 0,
+    "ADJ|o|l-acl": 1,
+    "ADJ|o|l-advcl": 2,
+    "ADJ|o|l-amod": 3,
+    "ADJ|o|l-ccomp": 4,
+    "ADJ|o|l-csubj": 5,
+    "ADJ|o|l-csubj:outer": 6,
+    "ADJ|o|l-nmod": 7,
+    "ADJ|o|l-nsubj": 8,
+    "ADJ|o|l-obj": 9,
+    "ADJ|o|l-obl": 10,
+    "ADJ|o|r-acl": 11,
+    "ADJ|o|r-amod": 12,
+    "ADJ|o|r-dep": 13,
+    "ADJ|o|root": 14,
+    "ADJ|x": 15,
+    "ADJ|x|l-acl": 16,
+    "ADJ|x|l-advcl": 17,
+    "ADJ|x|l-amod": 18,
+    "ADJ|x|l-ccomp": 19,
+    "ADJ|x|l-csubj": 20,
+    "ADJ|x|l-csubj:outer": 21,
+    "ADJ|x|l-nmod": 22,
+    "ADJ|x|l-nsubj": 23,
+    "ADJ|x|l-obj": 24,
+    "ADJ|x|l-obl": 25,
+    "ADJ|x|r-acl": 26,
+    "ADJ|x|r-amod": 27,
+    "ADJ|x|r-dep": 28,
+    "ADJ|x|root": 29,
+    "ADP|o": 30,
+    "ADP|o|l-case": 31,
+    "ADP|o|r-case": 32,
+    "ADP|o|r-fixed": 33,
+    "ADP|x": 34,
+    "ADP|x|l-case": 35,
+    "ADP|x|r-case": 36,
+    "ADP|x|r-fixed": 37,
+    "ADV|o": 38,
+    "ADV|o|l-advcl": 39,
+    "ADV|o|l-advmod": 40,
+    "ADV|o|l-obj": 41,
+    "ADV|o|r-dep": 42,
+    "ADV|o|root": 43,
+    "ADV|x": 44,
+    "ADV|x|l-advcl": 45,
+    "ADV|x|l-advmod": 46,
+    "ADV|x|l-obj": 47,
+    "ADV|x|r-dep": 48,
+    "ADV|x|root": 49,
+    "AUX|o": 50,
+    "AUX|o|Polarity=Neg": 51,
+    "AUX|o|Polarity=Neg|r-aux": 52,
+    "AUX|o|Polarity=Neg|r-fixed": 53,
+    "AUX|o|r-aux": 54,
+    "AUX|o|r-cop": 55,
+    "AUX|o|r-fixed": 56,
+    "AUX|o|root": 57,
+    "AUX|x": 58,
+    "AUX|x|Polarity=Neg": 59,
+    "AUX|x|Polarity=Neg|r-aux": 60,
+    "AUX|x|Polarity=Neg|r-fixed": 61,
+    "AUX|x|r-aux": 62,
+    "AUX|x|r-cop": 63,
+    "AUX|x|r-fixed": 64,
+    "AUX|x|root": 65,
+    "CCONJ|o": 66,
+    "CCONJ|o|l-cc": 67,
+    "CCONJ|o|r-cc": 68,
+    "CCONJ|x": 69,
+    "CCONJ|x|l-cc": 70,
+    "CCONJ|x|r-cc": 71,
+    "DET|o": 72,
+    "DET|o|l-det": 73,
+    "DET|x": 74,
+    "DET|x|l-det": 75,
+    "INTJ|o": 76,
+    "INTJ|o|l-discourse": 77,
+    "INTJ|o|r-discourse": 78,
+    "INTJ|o|root": 79,
+    "INTJ|x": 80,
+    "INTJ|x|l-discourse": 81,
+    "INTJ|x|r-discourse": 82,
+    "INTJ|x|root": 83,
+    "NOUN|o": 84,
+    "NOUN|o|Polarity=Neg": 85,
+    "NOUN|o|Polarity=Neg|l-obl": 86,
+    "NOUN|o|Polarity=Neg|root": 87,
+    "NOUN|o|l-acl": 88,
+    "NOUN|o|l-advcl": 89,
+    "NOUN|o|l-ccomp": 90,
+    "NOUN|o|l-compound": 91,
+    "NOUN|o|l-csubj": 92,
+    "NOUN|o|l-csubj:outer": 93,
+    "NOUN|o|l-nmod": 94,
+    "NOUN|o|l-nsubj": 95,
+    "NOUN|o|l-nsubj:outer": 96,
+    "NOUN|o|l-obj": 97,
+    "NOUN|o|l-obl": 98,
+    "NOUN|o|r-compound": 99,
+    "NOUN|o|r-nmod": 100,
+    "NOUN|o|r-nsubj": 101,
+    "NOUN|o|root": 102,
+    "NOUN|x": 103,
+    "NOUN|x|Polarity=Neg": 104,
+    "NOUN|x|Polarity=Neg|l-obl": 105,
+    "NOUN|x|Polarity=Neg|root": 106,
+    "NOUN|x|l-acl": 107,
+    "NOUN|x|l-advcl": 108,
+    "NOUN|x|l-ccomp": 109,
+    "NOUN|x|l-compound": 110,
+    "NOUN|x|l-csubj": 111,
+    "NOUN|x|l-csubj:outer": 112,
+    "NOUN|x|l-nmod": 113,
+    "NOUN|x|l-nsubj": 114,
+    "NOUN|x|l-nsubj:outer": 115,
+    "NOUN|x|l-obj": 116,
+    "NOUN|x|l-obl": 117,
+    "NOUN|x|r-compound": 118,
+    "NOUN|x|r-nmod": 119,
+    "NOUN|x|r-nsubj": 120,
+    "NOUN|x|root": 121,
+    "NUM|o": 122,
+    "NUM|o|l-advcl": 123,
+    "NUM|o|l-compound": 124,
+    "NUM|o|l-nmod": 125,
+    "NUM|o|l-nsubj": 126,
+    "NUM|o|l-nsubj:outer": 127,
+    "NUM|o|l-nummod": 128,
+    "NUM|o|l-obj": 129,
+    "NUM|o|l-obl": 130,
+    "NUM|o|r-compound": 131,
+    "NUM|o|root": 132,
+    "NUM|x": 133,
+    "NUM|x|l-advcl": 134,
+    "NUM|x|l-compound": 135,
+    "NUM|x|l-nmod": 136,
+    "NUM|x|l-nsubj": 137,
+    "NUM|x|l-nsubj:outer": 138,
+    "NUM|x|l-nummod": 139,
+    "NUM|x|l-obj": 140,
+    "NUM|x|l-obl": 141,
+    "NUM|x|r-compound": 142,
+    "NUM|x|root": 143,
+    "PART|o": 144,
+    "PART|o|l-mark": 145,
+    "PART|o|r-mark": 146,
+    "PART|x": 147,
+    "PART|x|l-mark": 148,
+    "PART|x|r-mark": 149,
+    "PRON|o": 150,
+    "PRON|o|l-acl": 151,
+    "PRON|o|l-advcl": 152,
+    "PRON|o|l-nmod": 153,
+    "PRON|o|l-nsubj": 154,
+    "PRON|o|l-nsubj:outer": 155,
+    "PRON|o|l-obj": 156,
+    "PRON|o|l-obl": 157,
+    "PRON|o|root": 158,
+    "PRON|x": 159,
+    "PRON|x|l-acl": 160,
+    "PRON|x|l-advcl": 161,
+    "PRON|x|l-nmod": 162,
+    "PRON|x|l-nsubj": 163,
+    "PRON|x|l-nsubj:outer": 164,
+    "PRON|x|l-obj": 165,
+    "PRON|x|l-obl": 166,
+    "PRON|x|root": 167,
+    "PROPN|o": 168,
+    "PROPN|o|l-acl": 169,
+    "PROPN|o|l-advcl": 170,
+    "PROPN|o|l-compound": 171,
+    "PROPN|o|l-nmod": 172,
+    "PROPN|o|l-nsubj": 173,
+    "PROPN|o|l-nsubj:outer": 174,
+    "PROPN|o|l-obj": 175,
+    "PROPN|o|l-obl": 176,
+    "PROPN|o|r-compound": 177,
+    "PROPN|o|r-nmod": 178,
+    "PROPN|o|root": 179,
+    "PROPN|x": 180,
+    "PROPN|x|l-acl": 181,
+    "PROPN|x|l-advcl": 182,
+    "PROPN|x|l-compound": 183,
+    "PROPN|x|l-nmod": 184,
+    "PROPN|x|l-nsubj": 185,
+    "PROPN|x|l-nsubj:outer": 186,
+    "PROPN|x|l-obj": 187,
+    "PROPN|x|l-obl": 188,
+    "PROPN|x|r-compound": 189,
+    "PROPN|x|r-nmod": 190,
+    "PROPN|x|root": 191,
+    "PUNCT|o": 192,
+    "PUNCT|o|l-punct": 193,
+    "PUNCT|o|r-punct": 194,
+    "PUNCT|x": 195,
+    "PUNCT|x|l-punct": 196,
+    "PUNCT|x|r-punct": 197,
+    "SCONJ|o": 198,
+    "SCONJ|o|l-dep": 199,
+    "SCONJ|o|r-fixed": 200,
+    "SCONJ|o|r-mark": 201,
+    "SCONJ|x": 202,
+    "SCONJ|x|l-dep": 203,
+    "SCONJ|x|r-fixed": 204,
+    "SCONJ|x|r-mark": 205,
+    "SYM|o": 206,
+    "SYM|o|l-compound": 207,
+    "SYM|o|l-dep": 208,
+    "SYM|o|l-nmod": 209,
+    "SYM|o|l-obl": 210,
+    "SYM|o|r-compound": 211,
+    "SYM|o|r-dep": 212,
+    "SYM|x": 213,
+    "SYM|x|l-compound": 214,
+    "SYM|x|l-dep": 215,
+    "SYM|x|l-nmod": 216,
+    "SYM|x|l-obl": 217,
+    "SYM|x|r-compound": 218,
+    "SYM|x|r-dep": 219,
+    "VERB|o": 220,
+    "VERB|o|l-acl": 221,
+    "VERB|o|l-advcl": 222,
+    "VERB|o|l-ccomp": 223,
+    "VERB|o|l-compound": 224,
+    "VERB|o|l-csubj": 225,
+    "VERB|o|l-csubj:outer": 226,
+    "VERB|o|l-nmod": 227,
+    "VERB|o|l-obj": 228,
+    "VERB|o|l-obl": 229,
+    "VERB|o|r-acl": 230,
+    "VERB|o|r-advcl": 231,
+    "VERB|o|r-compound": 232,
+    "VERB|o|root": 233,
+    "VERB|x": 234,
+    "VERB|x|l-acl": 235,
+    "VERB|x|l-advcl": 236,
+    "VERB|x|l-ccomp": 237,
+    "VERB|x|l-compound": 238,
+    "VERB|x|l-csubj": 239,
+    "VERB|x|l-csubj:outer": 240,
+    "VERB|x|l-nmod": 241,
+    "VERB|x|l-obj": 242,
+    "VERB|x|l-obl": 243,
+    "VERB|x|r-acl": 244,
+    "VERB|x|r-advcl": 245,
+    "VERB|x|r-compound": 246,
+    "VERB|x|root": 247,
+    "X|o": 248,
+    "X|o|l-nmod": 249,
+    "X|o|r-dep": 250,
+    "X|x": 251,
+    "X|x|l-nmod": 252,
+    "X|x|r-dep": 253,
+    "X|x|r-goeswith": 254
   },
   "layer_norm_eps": 1e-05,
   "local_attention": 128,

maker.py

@@ -7,12 +7,12 @@ 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")
 class UDTriangularDataset(object):
-  def __init__(self,conllu,tokenizer,embeddings=None):
+  def __init__(self,conllu,tokenizer):
     self.conllu=open(conllu,"r",encoding="utf-8")
     self.tokenizer=tokenizer
     self.seeks=[0]
-    label=set(["SYM","X"])
-    dep=set(["X|r-goeswith"])
+    label=set(["SYM|x","X|x"])
+    dep=set(["X|x|r-goeswith"])
     s=self.conllu.readline()
     while s!="":
       if s=="\n":
@@ -21,9 +21,13 @@ class UDTriangularDataset(object):
         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])
+            p=w[3]
+            q="" if w[5]=="_" else "|"+w[5]
+            d=("|" if w[6]=="0" else "|l-" if int(w[0])<int(w[6]) else "|r-")+w[7]
+            label.add(p+"|o"+q)
+            label.add(p+"|x"+q)
+            dep.add(p+"|o"+q+d)
+            dep.add(p+"|x"+q+d)
       s=self.conllu.readline()
     lid={l:i for i,l in enumerate(sorted(label))}
     for i,d in enumerate(sorted(dep),len(lid)):
@@ -51,19 +55,27 @@ class UDTriangularDataset(object):
         c.insert(i+1,[c[i][0],"_","_","X","_","_",c[i][0],"goeswith","_","_"])
     y=["0"]+[t[0] for t in c]
     h=[i if t[6]=="0" else y.index(t[6]) for i,t in enumerate(c,1)]
-    p=[t[3] if t[5]=="_" else t[3]+"|"+t[5] for t in c]
+    x=["o" if k>i or sum([1 if j==i+1 else 0 for j in h[i+1:]])>0 else "x" for i,k in enumerate(h)]
+    z=[len(x)-i+1 if k=="o" else 0 for i,k in enumerate(x)]
+    w=sum(z)+1
+    for i,j in enumerate(z):
+      if j==0 and w+len(x)-i<8192:
+        z[i]=len(x)-i+1
+        w+=z[i]
+    p=[t[3]+"|"+x[i] if t[5]=="_" else t[3]+"|"+x[i]+"|"+t[5] for i,t in enumerate(c)]
     d=[t[7] if t[6]=="0" else "l-"+t[7] if int(t[0])<int(t[6]) else "r-"+t[7] for t in c]
     v=sum(v,[])
     ids=[self.tokenizer.cls_token_id]
-    upos=["SYM"]
+    upos=["SYM|x"]
     for i,k in enumerate(v):
-      ids.append(k)
-      upos.append(p[i]+"|"+d[i] if h[i]==i+1 else p[i])
-      for j in range(i+1,len(v)):
-        ids.append(v[j])
-        upos.append(p[j]+"|"+d[j] if h[j]==i+1 else p[i]+"|"+d[i] if h[i]==j+1 else p[j])
-      ids.append(self.tokenizer.sep_token_id)
-      upos.append("SYM")
+      if z[i]>0:
+        ids.append(k)
+        upos.append(p[i]+"|"+d[i] if h[i]==i+1 else p[i])
+        for j in range(i+1,len(v)):
+          ids.append(v[j])
+          upos.append(p[j]+"|"+d[j] if h[j]==i+1 else p[i]+"|"+d[i] if h[i]==j+1 else p[j])
+        ids.append(self.tokenizer.sep_token_id)
+        upos.append("SYM|x")
     return {"input_ids":ids,"labels":[self.label2id[p] for p in upos]}
 from transformers import AutoTokenizer,AutoConfig,AutoModelForTokenClassification,DataCollatorForTokenClassification,TrainingArguments,Trainer
 tkz=AutoTokenizer.from_pretrained(src)

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0150e7519d7c382222f5dc38320b305f3eed53f6cf737c9233ba964521c3cfdf
-size 643808050
+oid sha256:972383c5654a11a5ff63bf765edf9fbfecaff7769b24eeb85249895d185d2ed7
+size 644198706

ud.py

@@ -5,9 +5,9 @@ class UniversalDependenciesPipeline(TokenClassificationPipeline):
   def __init__(self,**kwargs):
     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)
+    self.root=numpy.full((len(x)),-numpy.inf)
+    self.left_arc=numpy.full((len(x)),-numpy.inf)
+    self.right_arc=numpy.full((len(x)),-numpy.inf)
     for k,v in x.items():
       if k.endswith("|root"):
         self.root[v]=0
@@ -15,41 +15,73 @@ class UniversalDependenciesPipeline(TokenClassificationPipeline):
         self.left_arc[v]=0
       elif k.find("|r-")>0:
         self.right_arc[v]=0
-  def _forward(self,model_inputs):
-    import torch
-    v=model_inputs["input_ids"][0].tolist()
-    with torch.no_grad():
-      e=self.model(input_ids=torch.tensor([sum([v]+[v[i:] for i in range(2,len(v)-1)],[])]).to(self.device))
-    return {"logits":e.logits,**model_inputs}
   def check_model_type(self,supported_models):
     pass
   def postprocess(self,model_outputs,**kwargs):
+    import torch
     if "logits" not in model_outputs:
       return "".join(self.postprocess(x,**kwargs) for x in model_outputs)
     m=model_outputs["logits"][0].cpu().numpy()
-    w=len(model_outputs["input_ids"][0])-2
-    e=numpy.zeros((w,w,m.shape[-1]))
+    k=numpy.argmax(m,axis=1).tolist()
+    x=[self.model.config.id2label[i].split("|")[1]=="o" for i in k[1:-1]]
+    v=model_outputs["input_ids"][0].tolist()
+    off=model_outputs["offset_mapping"][0].tolist()
+    for i,(s,e) in reversed(list(enumerate(off))):
+      if s<e:
+        d=model_outputs["sentence"][s:e]
+        j=len(d)-len(d.lstrip())
+        if j>0:
+          d=d.lstrip()
+          off[i][0]+=j
+        j=len(d)-len(d.rstrip())
+        if j>0:
+          d=d.rstrip()
+          off[i][1]-=j
+        if d.strip()=="":
+          off.pop(i)
+          v.pop(i)
+          x.pop(i-1)
+    if len(x)<127:
+      x=[True]*len(x)
+    else:
+      w=sum([len(x)-i+1 if b else 0 for i,b in enumerate(x)])+1
+      for i in numpy.argsort(numpy.max(m,axis=1)[1:-1]):
+        if x[i]==False and w+len(x)-i<8192:
+          x[i]=True
+          w+=len(x)-i+1
+    w=[self.tokenizer.cls_token_id]
+    for i,j in enumerate(x):
+      if j:
+        w+=v[i+1:]
+    with torch.no_grad():
+      e=self.model(input_ids=torch.tensor([w]).to(self.device))
+    m=e.logits[0].cpu().numpy()
+    w=len(v)-2
+    e=numpy.full((w,w,m.shape[-1]),m.min())
     k=1
     for i in range(w):
-      e[i,i]=m[k]+self.root
-      for j in range(1,w-i):
-        e[i+j,i]=m[k+j]+self.left_arc
-        e[i,i+j]=m[k+j]+self.right_arc
-      k+=w-i+1
-    g=self.model.config.label2id["X|r-goeswith"]
+      if x[i]:
+        e[i,i]=m[k]+self.root
+        k+=1
+        for j in range(1,w-i):
+          e[i+j,i]=m[k]+self.left_arc
+          e[i,i+j]=m[k]+self.right_arc
+          k+=1
+        k+=1
+    g=self.model.config.label2id["X|x|r-goeswith"]
     r=numpy.tri(e.shape[0])
     for i in range(e.shape[0]):
       for j in range(i+2,e.shape[1]):
-        r[i,j]=r[i,j-1] if numpy.nanargmax(e[i,j-1])==g else 1
-    e[:,:,g]+=numpy.where(r==0,0,numpy.nan)
-    m,p=numpy.nanmax(e,axis=2),numpy.nanargmax(e,axis=2)
+        r[i,j]=r[i,j-1] if numpy.argmax(e[i,j-1])==g else 1
+    e[:,:,g]+=numpy.where(r==0,0,-numpy.inf)
+    m,p=numpy.max(e,axis=2),numpy.argmax(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)
+      k,h=z[numpy.argmax(m[z,z])],numpy.min(m)-numpy.max(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)
-    v=[(s,e) for s,e in model_outputs["offset_mapping"][0].tolist() if s<e]
+    v=[(s,e) for s,e in off if s<e]
     q=[self.model.config.id2label[p[j,i]].split("|") for i,j in enumerate(h)]
     if "aggregation_strategy" in kwargs and kwargs["aggregation_strategy"]!="none":
       for i,j in reversed(list(enumerate(q[1:],1))):
@@ -64,11 +96,10 @@ class UniversalDependenciesPipeline(TokenClassificationPipeline):
     t=model_outputs["sentence"].replace("\n"," ")
     u="# text = "+t+"\n"
     for i,(s,e) in enumerate(v):
-      u+="\t".join([str(i+1),t[s:e],"_",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(v) and e<v[i+1][0] else "SpaceAfter=No"])+"\n"
+      u+="\t".join([str(i+1),t[s:e],"_",q[i][0],"_","_" if len(q[i])<4 else "|".join(q[i][2:-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(v) and e<v[i+1][0] else "SpaceAfter=No"])+"\n"
     return u+"\n"
   def chu_liu_edmonds(self,matrix):
-    import numpy
-    h=numpy.nanargmax(matrix,axis=0)
+    h=numpy.argmax(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=[]
@@ -79,10 +110,10 @@ class UniversalDependenciesPipeline(TokenClassificationPipeline):
       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))]
+    z=matrix-numpy.max(matrix,axis=0)
+    m=numpy.block([[z[x,:][:,x],numpy.max(z[x,:][:,y],axis=1).reshape(len(x),1)],[numpy.max(z[y,:][:,x],axis=0),numpy.max(z[y,y])]])
+    k=[j if i==len(x) else x[j] if j<len(x) else y[numpy.argmax(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])]
+    i=y[numpy.argmax(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