KoichiYasuoka commited on
Commit
5896126
1 Parent(s): 7c2dad0

initial release

Browse files
config.json ADDED
@@ -0,0 +1,163 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForTokenClassification"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "auto_map": {
8
+ "AutoModelForTokenClassification": "upos.LlamaForTokenClassification"
9
+ },
10
+ "bos_token_id": 1,
11
+ "custom_pipelines": {
12
+ "upos": {
13
+ "impl": "upos.BellmanFordTokenClassificationPipeline",
14
+ "pt": "AutoModelForTokenClassification"
15
+ }
16
+ },
17
+ "eos_token_id": 2,
18
+ "hidden_act": "silu",
19
+ "hidden_size": 4096,
20
+ "id2label": {
21
+ "0": "ADJ",
22
+ "1": "B-ADJ",
23
+ "2": "I-ADJ",
24
+ "3": "ADJ|Polarity=Neg",
25
+ "4": "B-ADJ|Polarity=Neg",
26
+ "5": "I-ADJ|Polarity=Neg",
27
+ "6": "ADP",
28
+ "7": "B-ADP",
29
+ "8": "I-ADP",
30
+ "9": "ADV",
31
+ "10": "B-ADV",
32
+ "11": "I-ADV",
33
+ "12": "AUX",
34
+ "13": "B-AUX",
35
+ "14": "I-AUX",
36
+ "15": "AUX|Polarity=Neg",
37
+ "16": "B-AUX|Polarity=Neg",
38
+ "17": "I-AUX|Polarity=Neg",
39
+ "18": "CCONJ",
40
+ "19": "B-CCONJ",
41
+ "20": "I-CCONJ",
42
+ "21": "DET",
43
+ "22": "B-DET",
44
+ "23": "I-DET",
45
+ "24": "INTJ",
46
+ "25": "B-INTJ",
47
+ "26": "I-INTJ",
48
+ "27": "NOUN",
49
+ "28": "B-NOUN",
50
+ "29": "I-NOUN",
51
+ "30": "NOUN|Polarity=Neg",
52
+ "31": "B-NOUN|Polarity=Neg",
53
+ "32": "I-NOUN|Polarity=Neg",
54
+ "33": "NUM",
55
+ "34": "B-NUM",
56
+ "35": "I-NUM",
57
+ "36": "PART",
58
+ "37": "B-PART",
59
+ "38": "I-PART",
60
+ "39": "PRON",
61
+ "40": "B-PRON",
62
+ "41": "I-PRON",
63
+ "42": "PROPN",
64
+ "43": "B-PROPN",
65
+ "44": "I-PROPN",
66
+ "45": "PUNCT",
67
+ "46": "B-PUNCT",
68
+ "47": "I-PUNCT",
69
+ "48": "SCONJ",
70
+ "49": "B-SCONJ",
71
+ "50": "I-SCONJ",
72
+ "51": "SYM",
73
+ "52": "B-SYM",
74
+ "53": "I-SYM",
75
+ "54": "VERB",
76
+ "55": "B-VERB",
77
+ "56": "I-VERB",
78
+ "57": "X",
79
+ "58": "B-X",
80
+ "59": "I-X"
81
+ },
82
+ "initializer_range": 0.02,
83
+ "intermediate_size": 11008,
84
+ "label2id": {
85
+ "ADJ": 0,
86
+ "ADJ|Polarity=Neg": 3,
87
+ "ADP": 6,
88
+ "ADV": 9,
89
+ "AUX": 12,
90
+ "AUX|Polarity=Neg": 15,
91
+ "B-ADJ": 1,
92
+ "B-ADJ|Polarity=Neg": 4,
93
+ "B-ADP": 7,
94
+ "B-ADV": 10,
95
+ "B-AUX": 13,
96
+ "B-AUX|Polarity=Neg": 16,
97
+ "B-CCONJ": 19,
98
+ "B-DET": 22,
99
+ "B-INTJ": 25,
100
+ "B-NOUN": 28,
101
+ "B-NOUN|Polarity=Neg": 31,
102
+ "B-NUM": 34,
103
+ "B-PART": 37,
104
+ "B-PRON": 40,
105
+ "B-PROPN": 43,
106
+ "B-PUNCT": 46,
107
+ "B-SCONJ": 49,
108
+ "B-SYM": 52,
109
+ "B-VERB": 55,
110
+ "B-X": 58,
111
+ "CCONJ": 18,
112
+ "DET": 21,
113
+ "I-ADJ": 2,
114
+ "I-ADJ|Polarity=Neg": 5,
115
+ "I-ADP": 8,
116
+ "I-ADV": 11,
117
+ "I-AUX": 14,
118
+ "I-AUX|Polarity=Neg": 17,
119
+ "I-CCONJ": 20,
120
+ "I-DET": 23,
121
+ "I-INTJ": 26,
122
+ "I-NOUN": 29,
123
+ "I-NOUN|Polarity=Neg": 32,
124
+ "I-NUM": 35,
125
+ "I-PART": 38,
126
+ "I-PRON": 41,
127
+ "I-PROPN": 44,
128
+ "I-PUNCT": 47,
129
+ "I-SCONJ": 50,
130
+ "I-SYM": 53,
131
+ "I-VERB": 56,
132
+ "I-X": 59,
133
+ "INTJ": 24,
134
+ "NOUN": 27,
135
+ "NOUN|Polarity=Neg": 30,
136
+ "NUM": 33,
137
+ "PART": 36,
138
+ "PRON": 39,
139
+ "PROPN": 42,
140
+ "PUNCT": 45,
141
+ "SCONJ": 48,
142
+ "SYM": 51,
143
+ "VERB": 54,
144
+ "X": 57
145
+ },
146
+ "max_position_embeddings": 4096,
147
+ "max_sequence_length": 4096,
148
+ "model_type": "llama",
149
+ "num_attention_heads": 32,
150
+ "num_hidden_layers": 32,
151
+ "num_key_value_heads": 32,
152
+ "pad_token_id": 0,
153
+ "pretraining_tp": 1,
154
+ "rms_norm_eps": 1e-05,
155
+ "rope_scaling": null,
156
+ "rope_theta": 10000.0,
157
+ "tie_word_embeddings": false,
158
+ "tokenizer_class": "LlamaTokenizerFast",
159
+ "torch_dtype": "float32",
160
+ "transformers_version": "4.38.1",
161
+ "use_cache": true,
162
+ "vocab_size": 43744
163
+ }
maker.sh ADDED
@@ -0,0 +1,177 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #! /bin/sh
2
+ test -f ja_gsd_modern.conllu || curl -LO https://github.com/KoichiYasuoka/SuPar-UniDic/raw/main/suparunidic/suparmodels/ja_gsd_modern.conllu
3
+ test -f JapaneseCoreKanji.txt || curl -LO https://www.unicode.org/wg2/iso10646/edition6/data/JapaneseCoreKanji.txt
4
+
5
+ if [ ! -d exSwallow-7b-plus-hf ]
6
+ then TMPA=./maker$$a.py
7
+ cat << 'EOF' > $TMPA
8
+ #! /usr/bin/python3
9
+ src="tokyotech-llm/Swallow-7b-plus-hf"
10
+ tgt="exSwallow-7b-plus-hf"
11
+ import json,torch,unicodedata
12
+ from transformers import LlamaTokenizerFast,LlamaForCausalLM
13
+ with open("JapaneseCoreKanji.txt","r",encoding="utf-8") as r:
14
+ cjk=[chr(int(t,16)) for t in r.read().strip().split("\n") if not t.startswith("#")]
15
+ with open("ja_gsd_modern.conllu","r",encoding="utf-8") as r:
16
+ for s in r:
17
+ t=s.split("\t")
18
+ if len(t)==10:
19
+ for c in t[1]:
20
+ if unicodedata.name(c).startswith("CJK "):
21
+ cjk.append(c)
22
+ cjk=list(set(cjk))
23
+ tkz=LlamaTokenizerFast.from_pretrained(src,cls_token="<s>",sep_token="<s>",mask_token="<unk>",pad_token="</s>")
24
+ c={i:j[2:] for i,j in zip(cjk,tkz(cjk)["input_ids"]) if len(j)>3}
25
+ d=json.loads(tkz.backend_tokenizer.to_str())
26
+ for i,j in enumerate(c,len(tkz)):
27
+ d["model"]["vocab"][j]=i
28
+ tkz.backend_tokenizer.from_str(json.dumps(d)).save("tokenizer.json")
29
+ mdl=LlamaForCausalLM.from_pretrained(src)
30
+ tkz=LlamaTokenizerFast(tokenizer_file="tokenizer.json",model_max_length=mdl.config.max_position_embeddings,cls_token="<s>",sep_token="<s>",mask_token="<unk>",pad_token="</s>")
31
+ e=mdl.resize_token_embeddings(len(tkz))
32
+ f=mdl.get_output_embeddings()
33
+ with torch.no_grad():
34
+ for k,v in c.items():
35
+ e.weight[d["model"]["vocab"][k],:]=e.weight[v,:].sum(0)
36
+ f.weight[d["model"]["vocab"][k],:]=f.weight[v,:].sum(0)
37
+ mdl.set_input_embeddings(e)
38
+ mdl.set_output_embeddings(f)
39
+ mdl.save_pretrained(tgt)
40
+ tkz.save_pretrained(tgt)
41
+ EOF
42
+ chmod 755 $TMPA
43
+ $TMPA
44
+ fi
45
+
46
+ TMPB=./maker$$b.py
47
+ cat << 'EOF' > $TMPB
48
+ #! /usr/bin/env deepspeed
49
+ src="exSwallow-7b-plus-hf"
50
+ tgt="KoichiYasuoka/Swallow-7b-plus-upos"
51
+ from transformers import LlamaTokenizerFast,LlamaModel,LlamaPreTrainedModel,AutoConfig,DataCollatorForTokenClassification,TrainingArguments,Trainer
52
+ from transformers.modeling_outputs import TokenClassifierOutput
53
+ from tokenizers.normalizers import Replace
54
+
55
+ class LlamaForTokenClassification(LlamaPreTrainedModel):
56
+ def __init__(self,config):
57
+ from torch import nn
58
+ super().__init__(config)
59
+ self.num_labels=config.num_labels
60
+ self.model=LlamaModel(config)
61
+ if hasattr(config,"classifier_dropout") and config.classifier_dropout is not None:
62
+ classifier_dropout=config.classifier_dropout
63
+ elif hasattr(config,"hidden_dropout") and config.hidden_dropout is not None:
64
+ classifier_dropout=config.hidden_dropout
65
+ else:
66
+ classifier_dropout=0.1
67
+ self.dropout=nn.Dropout(classifier_dropout)
68
+ self.classifier=nn.Linear(config.hidden_size,config.num_labels)
69
+ self.post_init()
70
+ def get_input_embeddings(self):
71
+ return self.model.embed_tokens
72
+ def set_input_embeddings(self,value):
73
+ self.model.embed_tokens=value
74
+ def forward(self,input_ids=None,past_key_values=None,attention_mask=None,position_ids=None,inputs_embeds=None,labels=None,use_cache=None,output_attentions=None,output_hidden_states=None,return_dict=None):
75
+ return_dict=return_dict if return_dict is not None else self.config.use_return_dict
76
+ transformer_outputs=self.model(input_ids,past_key_values=past_key_values,attention_mask=attention_mask,position_ids=position_ids,inputs_embeds=inputs_embeds,use_cache=use_cache,output_attentions=output_attentions,output_hidden_states=output_hidden_states,return_dict=return_dict)
77
+ hidden_states=transformer_outputs[0]
78
+ hidden_states=self.dropout(hidden_states)
79
+ logits=self.classifier(hidden_states)
80
+ loss=None
81
+ if labels is not None:
82
+ from torch import nn
83
+ loss_fct=nn.CrossEntropyLoss()
84
+ loss=loss_fct(logits.view(-1,self.num_labels),labels.view(-1))
85
+ if not return_dict:
86
+ output=(logits,)+transformer_outputs[1:]
87
+ return ((loss,)+output) if loss is not None else output
88
+ return TokenClassifierOutput(loss=loss,logits=logits,hidden_states=transformer_outputs.hidden_states,attentions=transformer_outputs.attentions)
89
+
90
+ class UPOSFileDataset(object):
91
+ def __init__(self,conllu,tokenizer):
92
+ self.conllu=open(conllu,"r",encoding="utf-8")
93
+ self.tokenizer=tokenizer
94
+ self.seeks=[0]
95
+ self.multiword={}
96
+ label=set(["SYM"])
97
+ s=self.conllu.readline()
98
+ while s!="":
99
+ if s=="\n":
100
+ self.seeks.append(self.conllu.tell())
101
+ else:
102
+ w=s.split("\t")
103
+ if len(w)==10:
104
+ if w[0].isdecimal():
105
+ label.add(w[3] if w[5]=="_" else w[3]+"|"+w[5])
106
+ elif w[0].find("-")>0:
107
+ t=w[0].split("-")
108
+ f,j,k=w[1],[],[]
109
+ for i in range(int(t[0]),int(t[1])+1):
110
+ w=self.conllu.readline().split("\t")
111
+ j.append(w[3] if w[5]=="_" else w[3]+"|"+w[5])
112
+ k.append(w[1])
113
+ p="+".join(j)
114
+ label.add(p)
115
+ if p in self.multiword:
116
+ self.multiword[p][f]=list(k)
117
+ else:
118
+ self.multiword[p]={f:list(k)}
119
+ s=self.conllu.readline()
120
+ lid={}
121
+ for i,l in enumerate(sorted(label)):
122
+ lid[l],lid["B-"+l],lid["I-"+l]=i*3,i*3+1,i*3+2
123
+ self.label2id=lid
124
+ def __call__(*args):
125
+ lid={l:i for i,l in enumerate(sorted(set(sum([list(t.label2id) for t in args],[]))))}
126
+ for t in args:
127
+ t.label2id=lid
128
+ return lid
129
+ def __del__(self):
130
+ self.conllu.close()
131
+ __len__=lambda self:len(self.seeks)-1
132
+ def __getitem__(self,i):
133
+ self.conllu.seek(self.seeks[i])
134
+ form,upos=[],[]
135
+ while self.conllu.tell()<self.seeks[i+1]:
136
+ w=self.conllu.readline().split("\t")
137
+ if len(w)==10:
138
+ form.append(w[1])
139
+ if w[0].isdecimal():
140
+ upos.append(w[3] if w[5]=="_" else w[3]+"|"+w[5])
141
+ elif w[0].find("-")>0:
142
+ t=w[0].split("-")
143
+ u=[]
144
+ for j in range(int(t[0]),int(t[1])+1):
145
+ k=self.conllu.readline().split("\t")
146
+ u.append(k[3] if k[5]=="_" else k[3]+"|"+k[5])
147
+ upos.append("+".join(u))
148
+ v=self.tokenizer(form,add_special_tokens=False)
149
+ i,u=[],[]
150
+ for j,(x,y) in enumerate(zip(v["input_ids"],upos)):
151
+ if x!=[]:
152
+ i+=x
153
+ u+=[y] if len(x)==1 else ["B-"+y]+["I-"+y]*(len(x)-1)
154
+ if len(i)<self.tokenizer.model_max_length-3:
155
+ ids=[self.tokenizer.cls_token_id]+i+[self.tokenizer.sep_token_id]
156
+ upos=["SYM"]+u+["SYM"]
157
+ else:
158
+ ids=i[0:self.tokenizer.model_max_length-2]
159
+ upos=u[0:self.tokenizer.model_max_length-2]
160
+ return {"input_ids":ids,"labels":[self.label2id[t] for t in upos]}
161
+
162
+ tkz=LlamaTokenizerFast.from_pretrained(src)
163
+ tkz.backend_tokenizer.normalizer=Replace(" ","\u2581")
164
+ tkz.backend_tokenizer.model.byte_fallback=False
165
+ trainDS=UPOSFileDataset("ja_gsd_modern.conllu",tkz)
166
+ lid=trainDS.label2id
167
+ cfg=AutoConfig.from_pretrained(src,num_labels=len(lid),label2id=lid,id2label={i:l for l,i in lid.items()},ignore_mismatched_sizes=True)
168
+ dsp={"fp16":{"enabled":"auto"},"optimizer":{"type":"AdamW"},"scheduler":{"type":"WarmupLR","params":{}},"train_batch_size":"auto","train_micro_batch_size_per_gpu":"auto","zero_optimization":{"stage":3,"offload_optimizer":{"device":"cpu","pin_memory":True},"offload_param":{"device":"cpu","pin_memory":True},"overlap_comm":True,"contiguous_gradients":True,"reduce_bucket_size":"auto","stage3_prefetch_bucket_size":"auto","stage3_param_persistence_threshold":"auto","stage3_gather_16bit_weights_on_model_save":True}}
169
+ arg=TrainingArguments(num_train_epochs=3,per_device_train_batch_size=8,deepspeed=dsp,output_dir=tgt,overwrite_output_dir=True,save_total_limit=2,learning_rate=5e-05,warmup_ratio=0.1,save_safetensors=False)
170
+ trn=Trainer(args=arg,data_collator=DataCollatorForTokenClassification(tkz),model=LlamaForTokenClassification.from_pretrained(src,config=cfg,ignore_mismatched_sizes=True),train_dataset=trainDS)
171
+ trn.train()
172
+ trn.save_model(tgt)
173
+ tkz.save_pretrained(tgt)
174
+ EOF
175
+ chmod 755 $TMPB
176
+ $TMPB
177
+ exit
pytorch_model-00001-of-00006.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:febbf717e4d094a9f32b65b84a0dc1e76d671b6c09dd22e1729ea30cc1e72f26
3
+ size 4965712452
pytorch_model-00002-of-00006.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0df6860eb51bbd2ab136351a39f295d4a2853a74a7719e1ea83a296b0af3e9d1
3
+ size 4924328556
pytorch_model-00003-of-00006.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e9680466f58012f08f43e5f12372f3d48769f70e387debe22aaefba6d154682f
3
+ size 4857219294
pytorch_model-00004-of-00006.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:161461dcdc9717b80393bfa008d903aaa92122f584fb287047d4e78b2c68433a
3
+ size 4857219294
pytorch_model-00005-of-00006.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5dbfcec466897e051cbcf40dff7961a632a3836e68ae494edb2ce90e94a920f0
3
+ size 4857219294
pytorch_model-00006-of-00006.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1c75d142cb23ea767435988d9cff5daca87f8f22b2755b4424cfc9aa9a7f78df
3
+ size 2161173694
pytorch_model.bin.index.json ADDED
@@ -0,0 +1,299 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_size": 26622771440
4
+ },
5
+ "weight_map": {
6
+ "classifier.bias": "pytorch_model-00006-of-00006.bin",
7
+ "classifier.weight": "pytorch_model-00006-of-00006.bin",
8
+ "model.embed_tokens.weight": "pytorch_model-00001-of-00006.bin",
9
+ "model.layers.0.input_layernorm.weight": "pytorch_model-00001-of-00006.bin",
10
+ "model.layers.0.mlp.down_proj.weight": "pytorch_model-00001-of-00006.bin",
11
+ "model.layers.0.mlp.gate_proj.weight": "pytorch_model-00001-of-00006.bin",
12
+ "model.layers.0.mlp.up_proj.weight": "pytorch_model-00001-of-00006.bin",
13
+ "model.layers.0.post_attention_layernorm.weight": "pytorch_model-00001-of-00006.bin",
14
+ "model.layers.0.self_attn.k_proj.weight": "pytorch_model-00001-of-00006.bin",
15
+ "model.layers.0.self_attn.o_proj.weight": "pytorch_model-00001-of-00006.bin",
16
+ "model.layers.0.self_attn.q_proj.weight": "pytorch_model-00001-of-00006.bin",
17
+ "model.layers.0.self_attn.v_proj.weight": "pytorch_model-00001-of-00006.bin",
18
+ "model.layers.1.input_layernorm.weight": "pytorch_model-00001-of-00006.bin",
19
+ "model.layers.1.mlp.down_proj.weight": "pytorch_model-00001-of-00006.bin",
20
+ "model.layers.1.mlp.gate_proj.weight": "pytorch_model-00001-of-00006.bin",
21
+ "model.layers.1.mlp.up_proj.weight": "pytorch_model-00001-of-00006.bin",
22
+ "model.layers.1.post_attention_layernorm.weight": "pytorch_model-00001-of-00006.bin",
23
+ "model.layers.1.self_attn.k_proj.weight": "pytorch_model-00001-of-00006.bin",
24
+ "model.layers.1.self_attn.o_proj.weight": "pytorch_model-00001-of-00006.bin",
25
+ "model.layers.1.self_attn.q_proj.weight": "pytorch_model-00001-of-00006.bin",
26
+ "model.layers.1.self_attn.v_proj.weight": "pytorch_model-00001-of-00006.bin",
27
+ "model.layers.10.input_layernorm.weight": "pytorch_model-00002-of-00006.bin",
28
+ "model.layers.10.mlp.down_proj.weight": "pytorch_model-00002-of-00006.bin",
29
+ "model.layers.10.mlp.gate_proj.weight": "pytorch_model-00002-of-00006.bin",
30
+ "model.layers.10.mlp.up_proj.weight": "pytorch_model-00002-of-00006.bin",
31
+ "model.layers.10.post_attention_layernorm.weight": "pytorch_model-00002-of-00006.bin",
32
+ "model.layers.10.self_attn.k_proj.weight": "pytorch_model-00002-of-00006.bin",
33
+ "model.layers.10.self_attn.o_proj.weight": "pytorch_model-00002-of-00006.bin",
34
+ "model.layers.10.self_attn.q_proj.weight": "pytorch_model-00002-of-00006.bin",
35
+ "model.layers.10.self_attn.v_proj.weight": "pytorch_model-00002-of-00006.bin",
36
+ "model.layers.11.input_layernorm.weight": "pytorch_model-00003-of-00006.bin",
37
+ "model.layers.11.mlp.down_proj.weight": "pytorch_model-00003-of-00006.bin",
38
+ "model.layers.11.mlp.gate_proj.weight": "pytorch_model-00003-of-00006.bin",
39
+ "model.layers.11.mlp.up_proj.weight": "pytorch_model-00003-of-00006.bin",
40
+ "model.layers.11.post_attention_layernorm.weight": "pytorch_model-00003-of-00006.bin",
41
+ "model.layers.11.self_attn.k_proj.weight": "pytorch_model-00002-of-00006.bin",
42
+ "model.layers.11.self_attn.o_proj.weight": "pytorch_model-00002-of-00006.bin",
43
+ "model.layers.11.self_attn.q_proj.weight": "pytorch_model-00002-of-00006.bin",
44
+ "model.layers.11.self_attn.v_proj.weight": "pytorch_model-00002-of-00006.bin",
45
+ "model.layers.12.input_layernorm.weight": "pytorch_model-00003-of-00006.bin",
46
+ "model.layers.12.mlp.down_proj.weight": "pytorch_model-00003-of-00006.bin",
47
+ "model.layers.12.mlp.gate_proj.weight": "pytorch_model-00003-of-00006.bin",
48
+ "model.layers.12.mlp.up_proj.weight": "pytorch_model-00003-of-00006.bin",
49
+ "model.layers.12.post_attention_layernorm.weight": "pytorch_model-00003-of-00006.bin",
50
+ "model.layers.12.self_attn.k_proj.weight": "pytorch_model-00003-of-00006.bin",
51
+ "model.layers.12.self_attn.o_proj.weight": "pytorch_model-00003-of-00006.bin",
52
+ "model.layers.12.self_attn.q_proj.weight": "pytorch_model-00003-of-00006.bin",
53
+ "model.layers.12.self_attn.v_proj.weight": "pytorch_model-00003-of-00006.bin",
54
+ "model.layers.13.input_layernorm.weight": "pytorch_model-00003-of-00006.bin",
55
+ "model.layers.13.mlp.down_proj.weight": "pytorch_model-00003-of-00006.bin",
56
+ "model.layers.13.mlp.gate_proj.weight": "pytorch_model-00003-of-00006.bin",
57
+ "model.layers.13.mlp.up_proj.weight": "pytorch_model-00003-of-00006.bin",
58
+ "model.layers.13.post_attention_layernorm.weight": "pytorch_model-00003-of-00006.bin",
59
+ "model.layers.13.self_attn.k_proj.weight": "pytorch_model-00003-of-00006.bin",
60
+ "model.layers.13.self_attn.o_proj.weight": "pytorch_model-00003-of-00006.bin",
61
+ "model.layers.13.self_attn.q_proj.weight": "pytorch_model-00003-of-00006.bin",
62
+ "model.layers.13.self_attn.v_proj.weight": "pytorch_model-00003-of-00006.bin",
63
+ "model.layers.14.input_layernorm.weight": "pytorch_model-00003-of-00006.bin",
64
+ "model.layers.14.mlp.down_proj.weight": "pytorch_model-00003-of-00006.bin",
65
+ "model.layers.14.mlp.gate_proj.weight": "pytorch_model-00003-of-00006.bin",
66
+ "model.layers.14.mlp.up_proj.weight": "pytorch_model-00003-of-00006.bin",
67
+ "model.layers.14.post_attention_layernorm.weight": "pytorch_model-00003-of-00006.bin",
68
+ "model.layers.14.self_attn.k_proj.weight": "pytorch_model-00003-of-00006.bin",
69
+ "model.layers.14.self_attn.o_proj.weight": "pytorch_model-00003-of-00006.bin",
70
+ "model.layers.14.self_attn.q_proj.weight": "pytorch_model-00003-of-00006.bin",
71
+ "model.layers.14.self_attn.v_proj.weight": "pytorch_model-00003-of-00006.bin",
72
+ "model.layers.15.input_layernorm.weight": "pytorch_model-00003-of-00006.bin",
73
+ "model.layers.15.mlp.down_proj.weight": "pytorch_model-00003-of-00006.bin",
74
+ "model.layers.15.mlp.gate_proj.weight": "pytorch_model-00003-of-00006.bin",
75
+ "model.layers.15.mlp.up_proj.weight": "pytorch_model-00003-of-00006.bin",
76
+ "model.layers.15.post_attention_layernorm.weight": "pytorch_model-00003-of-00006.bin",
77
+ "model.layers.15.self_attn.k_proj.weight": "pytorch_model-00003-of-00006.bin",
78
+ "model.layers.15.self_attn.o_proj.weight": "pytorch_model-00003-of-00006.bin",
79
+ "model.layers.15.self_attn.q_proj.weight": "pytorch_model-00003-of-00006.bin",
80
+ "model.layers.15.self_attn.v_proj.weight": "pytorch_model-00003-of-00006.bin",
81
+ "model.layers.16.input_layernorm.weight": "pytorch_model-00003-of-00006.bin",
82
+ "model.layers.16.mlp.down_proj.weight": "pytorch_model-00003-of-00006.bin",
83
+ "model.layers.16.mlp.gate_proj.weight": "pytorch_model-00003-of-00006.bin",
84
+ "model.layers.16.mlp.up_proj.weight": "pytorch_model-00003-of-00006.bin",
85
+ "model.layers.16.post_attention_layernorm.weight": "pytorch_model-00003-of-00006.bin",
86
+ "model.layers.16.self_attn.k_proj.weight": "pytorch_model-00003-of-00006.bin",
87
+ "model.layers.16.self_attn.o_proj.weight": "pytorch_model-00003-of-00006.bin",
88
+ "model.layers.16.self_attn.q_proj.weight": "pytorch_model-00003-of-00006.bin",
89
+ "model.layers.16.self_attn.v_proj.weight": "pytorch_model-00003-of-00006.bin",
90
+ "model.layers.17.input_layernorm.weight": "pytorch_model-00004-of-00006.bin",
91
+ "model.layers.17.mlp.down_proj.weight": "pytorch_model-00004-of-00006.bin",
92
+ "model.layers.17.mlp.gate_proj.weight": "pytorch_model-00004-of-00006.bin",
93
+ "model.layers.17.mlp.up_proj.weight": "pytorch_model-00004-of-00006.bin",
94
+ "model.layers.17.post_attention_layernorm.weight": "pytorch_model-00004-of-00006.bin",
95
+ "model.layers.17.self_attn.k_proj.weight": "pytorch_model-00003-of-00006.bin",
96
+ "model.layers.17.self_attn.o_proj.weight": "pytorch_model-00003-of-00006.bin",
97
+ "model.layers.17.self_attn.q_proj.weight": "pytorch_model-00003-of-00006.bin",
98
+ "model.layers.17.self_attn.v_proj.weight": "pytorch_model-00003-of-00006.bin",
99
+ "model.layers.18.input_layernorm.weight": "pytorch_model-00004-of-00006.bin",
100
+ "model.layers.18.mlp.down_proj.weight": "pytorch_model-00004-of-00006.bin",
101
+ "model.layers.18.mlp.gate_proj.weight": "pytorch_model-00004-of-00006.bin",
102
+ "model.layers.18.mlp.up_proj.weight": "pytorch_model-00004-of-00006.bin",
103
+ "model.layers.18.post_attention_layernorm.weight": "pytorch_model-00004-of-00006.bin",
104
+ "model.layers.18.self_attn.k_proj.weight": "pytorch_model-00004-of-00006.bin",
105
+ "model.layers.18.self_attn.o_proj.weight": "pytorch_model-00004-of-00006.bin",
106
+ "model.layers.18.self_attn.q_proj.weight": "pytorch_model-00004-of-00006.bin",
107
+ "model.layers.18.self_attn.v_proj.weight": "pytorch_model-00004-of-00006.bin",
108
+ "model.layers.19.input_layernorm.weight": "pytorch_model-00004-of-00006.bin",
109
+ "model.layers.19.mlp.down_proj.weight": "pytorch_model-00004-of-00006.bin",
110
+ "model.layers.19.mlp.gate_proj.weight": "pytorch_model-00004-of-00006.bin",
111
+ "model.layers.19.mlp.up_proj.weight": "pytorch_model-00004-of-00006.bin",
112
+ "model.layers.19.post_attention_layernorm.weight": "pytorch_model-00004-of-00006.bin",
113
+ "model.layers.19.self_attn.k_proj.weight": "pytorch_model-00004-of-00006.bin",
114
+ "model.layers.19.self_attn.o_proj.weight": "pytorch_model-00004-of-00006.bin",
115
+ "model.layers.19.self_attn.q_proj.weight": "pytorch_model-00004-of-00006.bin",
116
+ "model.layers.19.self_attn.v_proj.weight": "pytorch_model-00004-of-00006.bin",
117
+ "model.layers.2.input_layernorm.weight": "pytorch_model-00001-of-00006.bin",
118
+ "model.layers.2.mlp.down_proj.weight": "pytorch_model-00001-of-00006.bin",
119
+ "model.layers.2.mlp.gate_proj.weight": "pytorch_model-00001-of-00006.bin",
120
+ "model.layers.2.mlp.up_proj.weight": "pytorch_model-00001-of-00006.bin",
121
+ "model.layers.2.post_attention_layernorm.weight": "pytorch_model-00001-of-00006.bin",
122
+ "model.layers.2.self_attn.k_proj.weight": "pytorch_model-00001-of-00006.bin",
123
+ "model.layers.2.self_attn.o_proj.weight": "pytorch_model-00001-of-00006.bin",
124
+ "model.layers.2.self_attn.q_proj.weight": "pytorch_model-00001-of-00006.bin",
125
+ "model.layers.2.self_attn.v_proj.weight": "pytorch_model-00001-of-00006.bin",
126
+ "model.layers.20.input_layernorm.weight": "pytorch_model-00004-of-00006.bin",
127
+ "model.layers.20.mlp.down_proj.weight": "pytorch_model-00004-of-00006.bin",
128
+ "model.layers.20.mlp.gate_proj.weight": "pytorch_model-00004-of-00006.bin",
129
+ "model.layers.20.mlp.up_proj.weight": "pytorch_model-00004-of-00006.bin",
130
+ "model.layers.20.post_attention_layernorm.weight": "pytorch_model-00004-of-00006.bin",
131
+ "model.layers.20.self_attn.k_proj.weight": "pytorch_model-00004-of-00006.bin",
132
+ "model.layers.20.self_attn.o_proj.weight": "pytorch_model-00004-of-00006.bin",
133
+ "model.layers.20.self_attn.q_proj.weight": "pytorch_model-00004-of-00006.bin",
134
+ "model.layers.20.self_attn.v_proj.weight": "pytorch_model-00004-of-00006.bin",
135
+ "model.layers.21.input_layernorm.weight": "pytorch_model-00004-of-00006.bin",
136
+ "model.layers.21.mlp.down_proj.weight": "pytorch_model-00004-of-00006.bin",
137
+ "model.layers.21.mlp.gate_proj.weight": "pytorch_model-00004-of-00006.bin",
138
+ "model.layers.21.mlp.up_proj.weight": "pytorch_model-00004-of-00006.bin",
139
+ "model.layers.21.post_attention_layernorm.weight": "pytorch_model-00004-of-00006.bin",
140
+ "model.layers.21.self_attn.k_proj.weight": "pytorch_model-00004-of-00006.bin",
141
+ "model.layers.21.self_attn.o_proj.weight": "pytorch_model-00004-of-00006.bin",
142
+ "model.layers.21.self_attn.q_proj.weight": "pytorch_model-00004-of-00006.bin",
143
+ "model.layers.21.self_attn.v_proj.weight": "pytorch_model-00004-of-00006.bin",
144
+ "model.layers.22.input_layernorm.weight": "pytorch_model-00004-of-00006.bin",
145
+ "model.layers.22.mlp.down_proj.weight": "pytorch_model-00004-of-00006.bin",
146
+ "model.layers.22.mlp.gate_proj.weight": "pytorch_model-00004-of-00006.bin",
147
+ "model.layers.22.mlp.up_proj.weight": "pytorch_model-00004-of-00006.bin",
148
+ "model.layers.22.post_attention_layernorm.weight": "pytorch_model-00004-of-00006.bin",
149
+ "model.layers.22.self_attn.k_proj.weight": "pytorch_model-00004-of-00006.bin",
150
+ "model.layers.22.self_attn.o_proj.weight": "pytorch_model-00004-of-00006.bin",
151
+ "model.layers.22.self_attn.q_proj.weight": "pytorch_model-00004-of-00006.bin",
152
+ "model.layers.22.self_attn.v_proj.weight": "pytorch_model-00004-of-00006.bin",
153
+ "model.layers.23.input_layernorm.weight": "pytorch_model-00005-of-00006.bin",
154
+ "model.layers.23.mlp.down_proj.weight": "pytorch_model-00005-of-00006.bin",
155
+ "model.layers.23.mlp.gate_proj.weight": "pytorch_model-00005-of-00006.bin",
156
+ "model.layers.23.mlp.up_proj.weight": "pytorch_model-00005-of-00006.bin",
157
+ "model.layers.23.post_attention_layernorm.weight": "pytorch_model-00005-of-00006.bin",
158
+ "model.layers.23.self_attn.k_proj.weight": "pytorch_model-00004-of-00006.bin",
159
+ "model.layers.23.self_attn.o_proj.weight": "pytorch_model-00004-of-00006.bin",
160
+ "model.layers.23.self_attn.q_proj.weight": "pytorch_model-00004-of-00006.bin",
161
+ "model.layers.23.self_attn.v_proj.weight": "pytorch_model-00004-of-00006.bin",
162
+ "model.layers.24.input_layernorm.weight": "pytorch_model-00005-of-00006.bin",
163
+ "model.layers.24.mlp.down_proj.weight": "pytorch_model-00005-of-00006.bin",
164
+ "model.layers.24.mlp.gate_proj.weight": "pytorch_model-00005-of-00006.bin",
165
+ "model.layers.24.mlp.up_proj.weight": "pytorch_model-00005-of-00006.bin",
166
+ "model.layers.24.post_attention_layernorm.weight": "pytorch_model-00005-of-00006.bin",
167
+ "model.layers.24.self_attn.k_proj.weight": "pytorch_model-00005-of-00006.bin",
168
+ "model.layers.24.self_attn.o_proj.weight": "pytorch_model-00005-of-00006.bin",
169
+ "model.layers.24.self_attn.q_proj.weight": "pytorch_model-00005-of-00006.bin",
170
+ "model.layers.24.self_attn.v_proj.weight": "pytorch_model-00005-of-00006.bin",
171
+ "model.layers.25.input_layernorm.weight": "pytorch_model-00005-of-00006.bin",
172
+ "model.layers.25.mlp.down_proj.weight": "pytorch_model-00005-of-00006.bin",
173
+ "model.layers.25.mlp.gate_proj.weight": "pytorch_model-00005-of-00006.bin",
174
+ "model.layers.25.mlp.up_proj.weight": "pytorch_model-00005-of-00006.bin",
175
+ "model.layers.25.post_attention_layernorm.weight": "pytorch_model-00005-of-00006.bin",
176
+ "model.layers.25.self_attn.k_proj.weight": "pytorch_model-00005-of-00006.bin",
177
+ "model.layers.25.self_attn.o_proj.weight": "pytorch_model-00005-of-00006.bin",
178
+ "model.layers.25.self_attn.q_proj.weight": "pytorch_model-00005-of-00006.bin",
179
+ "model.layers.25.self_attn.v_proj.weight": "pytorch_model-00005-of-00006.bin",
180
+ "model.layers.26.input_layernorm.weight": "pytorch_model-00005-of-00006.bin",
181
+ "model.layers.26.mlp.down_proj.weight": "pytorch_model-00005-of-00006.bin",
182
+ "model.layers.26.mlp.gate_proj.weight": "pytorch_model-00005-of-00006.bin",
183
+ "model.layers.26.mlp.up_proj.weight": "pytorch_model-00005-of-00006.bin",
184
+ "model.layers.26.post_attention_layernorm.weight": "pytorch_model-00005-of-00006.bin",
185
+ "model.layers.26.self_attn.k_proj.weight": "pytorch_model-00005-of-00006.bin",
186
+ "model.layers.26.self_attn.o_proj.weight": "pytorch_model-00005-of-00006.bin",
187
+ "model.layers.26.self_attn.q_proj.weight": "pytorch_model-00005-of-00006.bin",
188
+ "model.layers.26.self_attn.v_proj.weight": "pytorch_model-00005-of-00006.bin",
189
+ "model.layers.27.input_layernorm.weight": "pytorch_model-00005-of-00006.bin",
190
+ "model.layers.27.mlp.down_proj.weight": "pytorch_model-00005-of-00006.bin",
191
+ "model.layers.27.mlp.gate_proj.weight": "pytorch_model-00005-of-00006.bin",
192
+ "model.layers.27.mlp.up_proj.weight": "pytorch_model-00005-of-00006.bin",
193
+ "model.layers.27.post_attention_layernorm.weight": "pytorch_model-00005-of-00006.bin",
194
+ "model.layers.27.self_attn.k_proj.weight": "pytorch_model-00005-of-00006.bin",
195
+ "model.layers.27.self_attn.o_proj.weight": "pytorch_model-00005-of-00006.bin",
196
+ "model.layers.27.self_attn.q_proj.weight": "pytorch_model-00005-of-00006.bin",
197
+ "model.layers.27.self_attn.v_proj.weight": "pytorch_model-00005-of-00006.bin",
198
+ "model.layers.28.input_layernorm.weight": "pytorch_model-00005-of-00006.bin",
199
+ "model.layers.28.mlp.down_proj.weight": "pytorch_model-00005-of-00006.bin",
200
+ "model.layers.28.mlp.gate_proj.weight": "pytorch_model-00005-of-00006.bin",
201
+ "model.layers.28.mlp.up_proj.weight": "pytorch_model-00005-of-00006.bin",
202
+ "model.layers.28.post_attention_layernorm.weight": "pytorch_model-00005-of-00006.bin",
203
+ "model.layers.28.self_attn.k_proj.weight": "pytorch_model-00005-of-00006.bin",
204
+ "model.layers.28.self_attn.o_proj.weight": "pytorch_model-00005-of-00006.bin",
205
+ "model.layers.28.self_attn.q_proj.weight": "pytorch_model-00005-of-00006.bin",
206
+ "model.layers.28.self_attn.v_proj.weight": "pytorch_model-00005-of-00006.bin",
207
+ "model.layers.29.input_layernorm.weight": "pytorch_model-00006-of-00006.bin",
208
+ "model.layers.29.mlp.down_proj.weight": "pytorch_model-00006-of-00006.bin",
209
+ "model.layers.29.mlp.gate_proj.weight": "pytorch_model-00006-of-00006.bin",
210
+ "model.layers.29.mlp.up_proj.weight": "pytorch_model-00006-of-00006.bin",
211
+ "model.layers.29.post_attention_layernorm.weight": "pytorch_model-00006-of-00006.bin",
212
+ "model.layers.29.self_attn.k_proj.weight": "pytorch_model-00005-of-00006.bin",
213
+ "model.layers.29.self_attn.o_proj.weight": "pytorch_model-00005-of-00006.bin",
214
+ "model.layers.29.self_attn.q_proj.weight": "pytorch_model-00005-of-00006.bin",
215
+ "model.layers.29.self_attn.v_proj.weight": "pytorch_model-00005-of-00006.bin",
216
+ "model.layers.3.input_layernorm.weight": "pytorch_model-00001-of-00006.bin",
217
+ "model.layers.3.mlp.down_proj.weight": "pytorch_model-00001-of-00006.bin",
218
+ "model.layers.3.mlp.gate_proj.weight": "pytorch_model-00001-of-00006.bin",
219
+ "model.layers.3.mlp.up_proj.weight": "pytorch_model-00001-of-00006.bin",
220
+ "model.layers.3.post_attention_layernorm.weight": "pytorch_model-00001-of-00006.bin",
221
+ "model.layers.3.self_attn.k_proj.weight": "pytorch_model-00001-of-00006.bin",
222
+ "model.layers.3.self_attn.o_proj.weight": "pytorch_model-00001-of-00006.bin",
223
+ "model.layers.3.self_attn.q_proj.weight": "pytorch_model-00001-of-00006.bin",
224
+ "model.layers.3.self_attn.v_proj.weight": "pytorch_model-00001-of-00006.bin",
225
+ "model.layers.30.input_layernorm.weight": "pytorch_model-00006-of-00006.bin",
226
+ "model.layers.30.mlp.down_proj.weight": "pytorch_model-00006-of-00006.bin",
227
+ "model.layers.30.mlp.gate_proj.weight": "pytorch_model-00006-of-00006.bin",
228
+ "model.layers.30.mlp.up_proj.weight": "pytorch_model-00006-of-00006.bin",
229
+ "model.layers.30.post_attention_layernorm.weight": "pytorch_model-00006-of-00006.bin",
230
+ "model.layers.30.self_attn.k_proj.weight": "pytorch_model-00006-of-00006.bin",
231
+ "model.layers.30.self_attn.o_proj.weight": "pytorch_model-00006-of-00006.bin",
232
+ "model.layers.30.self_attn.q_proj.weight": "pytorch_model-00006-of-00006.bin",
233
+ "model.layers.30.self_attn.v_proj.weight": "pytorch_model-00006-of-00006.bin",
234
+ "model.layers.31.input_layernorm.weight": "pytorch_model-00006-of-00006.bin",
235
+ "model.layers.31.mlp.down_proj.weight": "pytorch_model-00006-of-00006.bin",
236
+ "model.layers.31.mlp.gate_proj.weight": "pytorch_model-00006-of-00006.bin",
237
+ "model.layers.31.mlp.up_proj.weight": "pytorch_model-00006-of-00006.bin",
238
+ "model.layers.31.post_attention_layernorm.weight": "pytorch_model-00006-of-00006.bin",
239
+ "model.layers.31.self_attn.k_proj.weight": "pytorch_model-00006-of-00006.bin",
240
+ "model.layers.31.self_attn.o_proj.weight": "pytorch_model-00006-of-00006.bin",
241
+ "model.layers.31.self_attn.q_proj.weight": "pytorch_model-00006-of-00006.bin",
242
+ "model.layers.31.self_attn.v_proj.weight": "pytorch_model-00006-of-00006.bin",
243
+ "model.layers.4.input_layernorm.weight": "pytorch_model-00001-of-00006.bin",
244
+ "model.layers.4.mlp.down_proj.weight": "pytorch_model-00001-of-00006.bin",
245
+ "model.layers.4.mlp.gate_proj.weight": "pytorch_model-00001-of-00006.bin",
246
+ "model.layers.4.mlp.up_proj.weight": "pytorch_model-00001-of-00006.bin",
247
+ "model.layers.4.post_attention_layernorm.weight": "pytorch_model-00001-of-00006.bin",
248
+ "model.layers.4.self_attn.k_proj.weight": "pytorch_model-00001-of-00006.bin",
249
+ "model.layers.4.self_attn.o_proj.weight": "pytorch_model-00001-of-00006.bin",
250
+ "model.layers.4.self_attn.q_proj.weight": "pytorch_model-00001-of-00006.bin",
251
+ "model.layers.4.self_attn.v_proj.weight": "pytorch_model-00001-of-00006.bin",
252
+ "model.layers.5.input_layernorm.weight": "pytorch_model-00002-of-00006.bin",
253
+ "model.layers.5.mlp.down_proj.weight": "pytorch_model-00002-of-00006.bin",
254
+ "model.layers.5.mlp.gate_proj.weight": "pytorch_model-00002-of-00006.bin",
255
+ "model.layers.5.mlp.up_proj.weight": "pytorch_model-00002-of-00006.bin",
256
+ "model.layers.5.post_attention_layernorm.weight": "pytorch_model-00002-of-00006.bin",
257
+ "model.layers.5.self_attn.k_proj.weight": "pytorch_model-00001-of-00006.bin",
258
+ "model.layers.5.self_attn.o_proj.weight": "pytorch_model-00002-of-00006.bin",
259
+ "model.layers.5.self_attn.q_proj.weight": "pytorch_model-00001-of-00006.bin",
260
+ "model.layers.5.self_attn.v_proj.weight": "pytorch_model-00001-of-00006.bin",
261
+ "model.layers.6.input_layernorm.weight": "pytorch_model-00002-of-00006.bin",
262
+ "model.layers.6.mlp.down_proj.weight": "pytorch_model-00002-of-00006.bin",
263
+ "model.layers.6.mlp.gate_proj.weight": "pytorch_model-00002-of-00006.bin",
264
+ "model.layers.6.mlp.up_proj.weight": "pytorch_model-00002-of-00006.bin",
265
+ "model.layers.6.post_attention_layernorm.weight": "pytorch_model-00002-of-00006.bin",
266
+ "model.layers.6.self_attn.k_proj.weight": "pytorch_model-00002-of-00006.bin",
267
+ "model.layers.6.self_attn.o_proj.weight": "pytorch_model-00002-of-00006.bin",
268
+ "model.layers.6.self_attn.q_proj.weight": "pytorch_model-00002-of-00006.bin",
269
+ "model.layers.6.self_attn.v_proj.weight": "pytorch_model-00002-of-00006.bin",
270
+ "model.layers.7.input_layernorm.weight": "pytorch_model-00002-of-00006.bin",
271
+ "model.layers.7.mlp.down_proj.weight": "pytorch_model-00002-of-00006.bin",
272
+ "model.layers.7.mlp.gate_proj.weight": "pytorch_model-00002-of-00006.bin",
273
+ "model.layers.7.mlp.up_proj.weight": "pytorch_model-00002-of-00006.bin",
274
+ "model.layers.7.post_attention_layernorm.weight": "pytorch_model-00002-of-00006.bin",
275
+ "model.layers.7.self_attn.k_proj.weight": "pytorch_model-00002-of-00006.bin",
276
+ "model.layers.7.self_attn.o_proj.weight": "pytorch_model-00002-of-00006.bin",
277
+ "model.layers.7.self_attn.q_proj.weight": "pytorch_model-00002-of-00006.bin",
278
+ "model.layers.7.self_attn.v_proj.weight": "pytorch_model-00002-of-00006.bin",
279
+ "model.layers.8.input_layernorm.weight": "pytorch_model-00002-of-00006.bin",
280
+ "model.layers.8.mlp.down_proj.weight": "pytorch_model-00002-of-00006.bin",
281
+ "model.layers.8.mlp.gate_proj.weight": "pytorch_model-00002-of-00006.bin",
282
+ "model.layers.8.mlp.up_proj.weight": "pytorch_model-00002-of-00006.bin",
283
+ "model.layers.8.post_attention_layernorm.weight": "pytorch_model-00002-of-00006.bin",
284
+ "model.layers.8.self_attn.k_proj.weight": "pytorch_model-00002-of-00006.bin",
285
+ "model.layers.8.self_attn.o_proj.weight": "pytorch_model-00002-of-00006.bin",
286
+ "model.layers.8.self_attn.q_proj.weight": "pytorch_model-00002-of-00006.bin",
287
+ "model.layers.8.self_attn.v_proj.weight": "pytorch_model-00002-of-00006.bin",
288
+ "model.layers.9.input_layernorm.weight": "pytorch_model-00002-of-00006.bin",
289
+ "model.layers.9.mlp.down_proj.weight": "pytorch_model-00002-of-00006.bin",
290
+ "model.layers.9.mlp.gate_proj.weight": "pytorch_model-00002-of-00006.bin",
291
+ "model.layers.9.mlp.up_proj.weight": "pytorch_model-00002-of-00006.bin",
292
+ "model.layers.9.post_attention_layernorm.weight": "pytorch_model-00002-of-00006.bin",
293
+ "model.layers.9.self_attn.k_proj.weight": "pytorch_model-00002-of-00006.bin",
294
+ "model.layers.9.self_attn.o_proj.weight": "pytorch_model-00002-of-00006.bin",
295
+ "model.layers.9.self_attn.q_proj.weight": "pytorch_model-00002-of-00006.bin",
296
+ "model.layers.9.self_attn.v_proj.weight": "pytorch_model-00002-of-00006.bin",
297
+ "model.norm.weight": "pytorch_model-00006-of-00006.bin"
298
+ }
299
+ }
special_tokens_map.json ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<s>",
4
+ "lstrip": false,
5
+ "normalized": true,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "cls_token": {
10
+ "content": "<s>",
11
+ "lstrip": false,
12
+ "normalized": true,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "eos_token": {
17
+ "content": "</s>",
18
+ "lstrip": false,
19
+ "normalized": true,
20
+ "rstrip": false,
21
+ "single_word": false
22
+ },
23
+ "mask_token": {
24
+ "content": "<unk>",
25
+ "lstrip": false,
26
+ "normalized": true,
27
+ "rstrip": false,
28
+ "single_word": false
29
+ },
30
+ "pad_token": {
31
+ "content": "</s>",
32
+ "lstrip": false,
33
+ "normalized": true,
34
+ "rstrip": false,
35
+ "single_word": false
36
+ },
37
+ "sep_token": {
38
+ "content": "<s>",
39
+ "lstrip": false,
40
+ "normalized": true,
41
+ "rstrip": false,
42
+ "single_word": false
43
+ },
44
+ "unk_token": {
45
+ "content": "<unk>",
46
+ "lstrip": false,
47
+ "normalized": true,
48
+ "rstrip": false,
49
+ "single_word": false
50
+ }
51
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": true,
3
+ "add_eos_token": false,
4
+ "added_tokens_decoder": {
5
+ "0": {
6
+ "content": "<unk>",
7
+ "lstrip": false,
8
+ "normalized": true,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "1": {
14
+ "content": "<s>",
15
+ "lstrip": false,
16
+ "normalized": true,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "2": {
22
+ "content": "</s>",
23
+ "lstrip": false,
24
+ "normalized": true,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ }
29
+ },
30
+ "bos_token": "<s>",
31
+ "clean_up_tokenization_spaces": false,
32
+ "cls_token": "<s>",
33
+ "eos_token": "</s>",
34
+ "mask_token": "<unk>",
35
+ "model_max_length": 4096,
36
+ "pad_token": "</s>",
37
+ "sep_token": "<s>",
38
+ "tokenizer_class": "LlamaTokenizerFast",
39
+ "unk_token": "<unk>",
40
+ "use_default_system_prompt": false
41
+ }
upos.py ADDED
@@ -0,0 +1,76 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from transformers import TokenClassificationPipeline,LlamaModel,LlamaPreTrainedModel
2
+ from transformers.modeling_outputs import TokenClassifierOutput
3
+
4
+ class BellmanFordTokenClassificationPipeline(TokenClassificationPipeline):
5
+ def __init__(self,**kwargs):
6
+ import numpy
7
+ super().__init__(**kwargs)
8
+ x=self.model.config.label2id
9
+ y=[k for k in x if not k.startswith("I-")]
10
+ self.transition=numpy.full((len(x),len(x)),numpy.nan)
11
+ for k,v in x.items():
12
+ for j in ["I-"+k[2:]] if k.startswith("B-") else [k]+y if k.startswith("I-") else y:
13
+ self.transition[v,x[j]]=0
14
+ def check_model_type(self,supported_models):
15
+ pass
16
+ def postprocess(self,model_outputs,**kwargs):
17
+ import numpy
18
+ if "logits" not in model_outputs:
19
+ return self.postprocess(model_outputs[0],**kwargs)
20
+ m=model_outputs["logits"][0].numpy()
21
+ e=numpy.exp(m-numpy.max(m,axis=-1,keepdims=True))
22
+ z=e/e.sum(axis=-1,keepdims=True)
23
+ for i in range(m.shape[0]-1,0,-1):
24
+ m[i-1]+=numpy.nanmax(m[i]+self.transition,axis=1)
25
+ k=[numpy.nanargmax(m[0])]
26
+ for i in range(1,m.shape[0]):
27
+ k.append(numpy.nanargmax(m[i]+self.transition[k[-1]]))
28
+ 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]
29
+ if "aggregation_strategy" in kwargs and kwargs["aggregation_strategy"]!="none":
30
+ for i,t in reversed(list(enumerate(w))):
31
+ p=t.pop("entity")
32
+ if p.startswith("I-"):
33
+ w[i-1]["score"]=min(w[i-1]["score"],t["score"])
34
+ w[i-1]["end"]=w.pop(i)["end"]
35
+ elif p.startswith("B-"):
36
+ t["entity_group"]=p[2:]
37
+ else:
38
+ t["entity_group"]=p
39
+ for t in w:
40
+ t["text"]=model_outputs["sentence"][t["start"]:t["end"]]
41
+ return w
42
+
43
+ class LlamaForTokenClassification(LlamaPreTrainedModel):
44
+ def __init__(self,config):
45
+ from torch import nn
46
+ super().__init__(config)
47
+ self.num_labels=config.num_labels
48
+ self.model=LlamaModel(config)
49
+ if hasattr(config,"classifier_dropout") and config.classifier_dropout is not None:
50
+ classifier_dropout=config.classifier_dropout
51
+ elif hasattr(config,"hidden_dropout") and config.hidden_dropout is not None:
52
+ classifier_dropout=config.hidden_dropout
53
+ else:
54
+ classifier_dropout=0.1
55
+ self.dropout=nn.Dropout(classifier_dropout)
56
+ self.classifier=nn.Linear(config.hidden_size,config.num_labels)
57
+ self.post_init()
58
+ def get_input_embeddings(self):
59
+ return self.model.embed_tokens
60
+ def set_input_embeddings(self,value):
61
+ self.model.embed_tokens=value
62
+ def forward(self,input_ids=None,past_key_values=None,attention_mask=None,position_ids=None,inputs_embeds=None,labels=None,use_cache=None,output_attentions=None,output_hidden_states=None,return_dict=None):
63
+ return_dict=return_dict if return_dict is not None else self.config.use_return_dict
64
+ transformer_outputs=self.model(input_ids,past_key_values=past_key_values,attention_mask=attention_mask,position_ids=position_ids,inputs_embeds=inputs_embeds,use_cache=use_cache,output_attentions=output_attentions,output_hidden_states=output_hidden_states,return_dict=return_dict)
65
+ hidden_states=transformer_outputs[0]
66
+ hidden_states=self.dropout(hidden_states)
67
+ logits=self.classifier(hidden_states)
68
+ loss=None
69
+ if labels is not None:
70
+ from torch import nn
71
+ loss_fct=nn.CrossEntropyLoss()
72
+ loss=loss_fct(logits.view(-1,self.num_labels),labels.view(-1))
73
+ if not return_dict:
74
+ output=(logits,)+transformer_outputs[1:]
75
+ return ((loss,)+output) if loss is not None else output
76
+ return TokenClassifierOutput(loss=loss,logits=logits,hidden_states=transformer_outputs.hidden_states,attentions=transformer_outputs.attentions)