initial release

README.md

@@ -0,0 +1,25 @@
+---
+language:
+- "bo"
+tags:
+- "tibetan"
+- "token-classification"
+- "pos"
+base_model: KoichiYasuoka/roberta-base-tibetan
+license: "cc-by-sa-4.0"
+pipeline_tag: "token-classification"
+---
+
+# roberta-base-tibetan-upos
+
+## Model Description
+
+This is a RoBERTa model for POS-tagging, derived from [roberta-base-tibetan](https://huggingface.co/KoichiYasuoka/roberta-base-tibetan). Every word is tagged by [UPOS](https://universaldependencies.org/u/pos/) (Universal Part-Of-Speech).
+
+## How to Use
+
+```py
+from transformers import pipeline
+nlp=pipeline("upos","KoichiYasuoka/roberta-base-tibetan-upos",trust_remote_code=True,aggregation_strategy="simple")
+```
+

config.json

@@ -0,0 +1,181 @@
+{
+  "architectures": [
+    "RobertaForTokenClassification"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "bos_token_id": 0,
+  "classifier_dropout": null,
+    "custom_pipelines": {
+    "upos": {
+      "impl": "upos.BellmanFordTokenClassificationPipeline",
+      "pt": "AutoModelForTokenClassification"
+    }
+  },
+  "eos_token_id": 2,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "id2label": {
+    "0": "ADJ",
+    "1": "ADP",
+    "2": "ADV",
+    "3": "AUX",
+    "4": "B-ADJ",
+    "5": "B-ADJ+ADP",
+    "6": "B-ADP",
+    "7": "B-ADV",
+    "8": "B-AUX",
+    "9": "B-AUX+PART",
+    "10": "B-AUX+SCONJ",
+    "11": "B-DET",
+    "12": "B-INTJ",
+    "13": "B-NOUN",
+    "14": "B-NOUN+ADP",
+    "15": "B-NUM",
+    "16": "B-NUM+ADP",
+    "17": "B-PART",
+    "18": "B-PRON",
+    "19": "B-PRON+ADP",
+    "20": "B-PROPN",
+    "21": "B-PROPN+ADP",
+    "22": "B-PROPN+SCONJ",
+    "23": "B-PUNCT",
+    "24": "B-PUNCT+NOUN",
+    "25": "B-SCONJ",
+    "26": "B-VERB",
+    "27": "B-VERB+ADP",
+    "28": "B-VERB+PART",
+    "29": "B-VERB+SCONJ",
+    "30": "B-X",
+    "31": "DET",
+    "32": "I-ADJ",
+    "33": "I-ADJ+ADP",
+    "34": "I-ADP",
+    "35": "I-ADV",
+    "36": "I-AUX",
+    "37": "I-AUX+PART",
+    "38": "I-AUX+SCONJ",
+    "39": "I-DET",
+    "40": "I-INTJ",
+    "41": "I-NOUN",
+    "42": "I-NOUN+ADP",
+    "43": "I-NUM",
+    "44": "I-NUM+ADP",
+    "45": "I-PART",
+    "46": "I-PRON",
+    "47": "I-PRON+ADP",
+    "48": "I-PROPN",
+    "49": "I-PROPN+ADP",
+    "50": "I-PROPN+SCONJ",
+    "51": "I-PUNCT",
+    "52": "I-PUNCT+NOUN",
+    "53": "I-SCONJ",
+    "54": "I-VERB",
+    "55": "I-VERB+ADP",
+    "56": "I-VERB+PART",
+    "57": "I-VERB+SCONJ",
+    "58": "I-X",
+    "59": "INTJ",
+    "60": "NOUN",
+    "61": "NUM",
+    "62": "PART",
+    "63": "PART+NOUN",
+    "64": "PART+PUNCT",
+    "65": "PRON",
+    "66": "PROPN",
+    "67": "PUNCT",
+    "68": "SCONJ",
+    "69": "SYM",
+    "70": "VERB",
+    "71": "X"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "label2id": {
+    "ADJ": 0,
+    "ADP": 1,
+    "ADV": 2,
+    "AUX": 3,
+    "B-ADJ": 4,
+    "B-ADJ+ADP": 5,
+    "B-ADP": 6,
+    "B-ADV": 7,
+    "B-AUX": 8,
+    "B-AUX+PART": 9,
+    "B-AUX+SCONJ": 10,
+    "B-DET": 11,
+    "B-INTJ": 12,
+    "B-NOUN": 13,
+    "B-NOUN+ADP": 14,
+    "B-NUM": 15,
+    "B-NUM+ADP": 16,
+    "B-PART": 17,
+    "B-PRON": 18,
+    "B-PRON+ADP": 19,
+    "B-PROPN": 20,
+    "B-PROPN+ADP": 21,
+    "B-PROPN+SCONJ": 22,
+    "B-PUNCT": 23,
+    "B-PUNCT+NOUN": 24,
+    "B-SCONJ": 25,
+    "B-VERB": 26,
+    "B-VERB+ADP": 27,
+    "B-VERB+PART": 28,
+    "B-VERB+SCONJ": 29,
+    "B-X": 30,
+    "DET": 31,
+    "I-ADJ": 32,
+    "I-ADJ+ADP": 33,
+    "I-ADP": 34,
+    "I-ADV": 35,
+    "I-AUX": 36,
+    "I-AUX+PART": 37,
+    "I-AUX+SCONJ": 38,
+    "I-DET": 39,
+    "I-INTJ": 40,
+    "I-NOUN": 41,
+    "I-NOUN+ADP": 42,
+    "I-NUM": 43,
+    "I-NUM+ADP": 44,
+    "I-PART": 45,
+    "I-PRON": 46,
+    "I-PRON+ADP": 47,
+    "I-PROPN": 48,
+    "I-PROPN+ADP": 49,
+    "I-PROPN+SCONJ": 50,
+    "I-PUNCT": 51,
+    "I-PUNCT+NOUN": 52,
+    "I-SCONJ": 53,
+    "I-VERB": 54,
+    "I-VERB+ADP": 55,
+    "I-VERB+PART": 56,
+    "I-VERB+SCONJ": 57,
+    "I-X": 58,
+    "INTJ": 59,
+    "NOUN": 60,
+    "NUM": 61,
+    "PART": 62,
+    "PART+NOUN": 63,
+    "PART+PUNCT": 64,
+    "PRON": 65,
+    "PROPN": 66,
+    "PUNCT": 67,
+    "SCONJ": 68,
+    "SYM": 69,
+    "VERB": 70,
+    "X": 71
+  },
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "roberta",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 1,
+  "position_embedding_type": "absolute",
+  "tokenizer_class": "BertTokenizerFast",
+  "torch_dtype": "float32",
+  "transformers_version": "4.39.3",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 8094
+}

maker.sh

@@ -0,0 +1,22 @@
+#! /bin/sh
+for D in classical-tibetan-corpus old-tibetan-corpus modern-tibetan-corpus
+do test -d $D || git clone --depth=1 https://github.com/tibetan-nlp/$D
+done
+( for F in *-tibetan-corpus/conllu/*.conllu
+  do case $F in
+     *-translated.conllu) : ;;
+     *) cat $F ;;
+     esac
+  done
+) | awk '
+{
+  if($0==""){
+    if(u!~/\tNOTAG\t/)
+      print u;
+    u="";
+  }
+  else
+    u=u$0"\n";
+}'> all.conllu
+python3 -m esupar.train KoichiYasuoka/roberta-base-tibetan KoichiYasuoka/roberta-base-tibetan-upos 24 /tmp all.conllu
+exit 0

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61e6c8c1160a2a0d9e7169b2c28ee602c3718a47b35017112021016edadb2b1a
+size 366957990

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 +1,62 @@
+{
+  "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,
+  "keep_accents": true,
+  "mask_token": "[MASK]",
+  "max_length": 510,
+  "model_max_length": 512,
+  "never_split": null,
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "stride": 0,
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizerFast",
+  "truncation_side": "right",
+  "truncation_strategy": "longest_first",
+  "unk_token": "[UNK]"
+}

upos.py

@@ -0,0 +1,46 @@
+from transformers import TokenClassificationPipeline
+
+class BellmanFordTokenClassificationPipeline(TokenClassificationPipeline):
+  def __init__(self,**kwargs):
+    import numpy
+    super().__init__(**kwargs)
+    x=self.model.config.label2id
+    y=[k for k in x if not k.startswith("I-")]
+    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):
+    import numpy
+    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
+    s=model_outputs["sentence"]
+    for i,t in enumerate(w):
+      if t["end"]<len(s):
+        if s[t["end"]] in {"\u0f0b","\u0f0c"}:
+          if len(w)-i==1 or t["end"]<w[i+1]["start"]:
+            t["end"]+=1
+      t["text"]=s[t["start"]:t["end"]]
+    return w
+