add model

amc.py

@@ -2,8 +2,7 @@ import copy
 
 from transformers import PreTrainedModel, PretrainedConfig, AutoModel, AutoTokenizer
 import torch.nn as nn
-import torch
-from typing import List
+
 
 
 class AMC_OPT_conf(PretrainedConfig):
@@ -18,7 +17,6 @@ class AMC_OPT_conf(PretrainedConfig):
         self.drop_out = drop_out
         self.emb_size = emb_size
         self.pretrained_model = pretrained_model
-        self.fine_tuned_model = "data/models/amc_opt_msmd_nocoadapt.pt"
         super().__init__(**kwargs)
 
 
@@ -53,10 +51,6 @@ class AMC_OPT(PreTrainedModel):
     def __init__(self, config):
         super().__init__(config)
         self.model = AMC_OPT_sub(config)
-        self.model.load_state_dict(torch.load(config.fine_tuned_model))
 
     def forward(self, labels, input_ids, attention_mask, **args):
         return self.model(labels, input_ids, attention_mask, **args)
-
-tokenizer = AutoTokenizer.from_pretrained("m-polignano-uniba/bert_uncased_L-12_H-768_A-12_italian_alb3rt0")
-tokenizer.model_max_length = 128

config.json

@@ -2,10 +2,13 @@
   "architectures": [
     "AMC_OPT"
   ],
+  "auto_map": {
+    "AutoConfig": "amc.AMC_OPT_conf",
+    "AutoModelForSequenceClassification": "amc.AMC_OPT"
+  },
   "drop_out": 0.1,
   "emb_size": 768,
-  "fine_tuned_model": "data/models/amc_opt_msmd_nocoadapt.pt",
-  "model_type": "bert-sa",
+  "model_type": "bert",
   "out_labels": 3,
   "pretrained_model": "m-polignano-uniba/bert_uncased_L-12_H-768_A-12_italian_alb3rt0",
   "torch_dtype": "float32",

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5e8e4602d1fabefb736c5ec5abd5de581f74a03fe58d653a958ec22b060dee6c
-size 737463021
+oid sha256:faacdd6b1fcee65d1a748e1e98e0f1c9ba3feb119224dcd9e07741e2f6d67155
+size 737460205