Add new LinkTransformer model.

.gitattributes

@@ -34,3 +34,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 model.safetensors filter=lfs diff=lfs merge=lfs -text
+.git/lfs/objects/50/c5/50c55d19867e26fc83db98c43d59b1d3af837a78e24a9d131a4953a572074379 filter=lfs diff=lfs merge=lfs -text

1_Pooling/config.json

@@ -3,5 +3,7 @@
   "pooling_mode_cls_token": false,
   "pooling_mode_mean_tokens": true,
   "pooling_mode_max_tokens": false,
-  "pooling_mode_mean_sqrt_len_tokens": false
+  "pooling_mode_mean_sqrt_len_tokens": false,
+  "pooling_mode_weightedmean_tokens": false,
+  "pooling_mode_lasttoken": false
 }

LT_training_config.json

@@ -1,18 +1,18 @@
 {
   "model_save_dir": "models",
-  "model_save_name": "lt-historicjapanesecompanies-comp-prod-ind_onlinecontrastive_full",
-  "opt_model_description": "This model was trained on a dataset of historic Japanese companies, products, industry, addresses, and shareholders. Take a look at our paper for more details. The task is to link indices of japanese companies",
+  "model_save_name": "lt-wikidata-comp-prod-ind-ja",
+  "opt_model_description": "This is a (Modern) Japanese Link Transformer model  - trained on Company <SEP> Product <SEP> Industry from wiki data.",
   "opt_model_lang": "ja",
   "train_batch_size": 64,
-  "num_epochs": 50,
+  "num_epochs": 70,
   "warm_up_perc": 1,
   "learning_rate": 2e-05,
-  "loss_type": "onlinecontrastive",
+  "loss_type": "supcon",
   "val_perc": 0.2,
   "wandb_names": {
-    "id": "econabhishek",
-    "run": "lt-historicjapanesecompanies-comp-prod-ind_onlinecontrastive_full",
     "project": "linkage",
+    "id": "econabhishek",
+    "run": "lt-wikidata-comp-prod-ind-ja",
     "entity": "econabhishek"
   },
   "add_pooling_layer": false,
@@ -22,8 +22,8 @@
   "save_val_test_pickles": true,
   "val_query_prop": 0.5,
   "loss_params": {},
-  "eval_type": "classification",
+  "eval_type": "retrieval",
   "training_dataset": "dataframe",
   "base_model_path": "oshizo/sbert-jsnli-luke-japanese-base-lite",
-  "best_model_path": "models/lt-historicjapanesecompanies-comp-prod-ind_onlinecontrastive_full"
+  "best_model_path": "models/lt-wikidata-comp-prod-ind-ja"
 }

README.md

@@ -22,7 +22,7 @@ Take a look at the documentation of [sentence-transformers](https://www.sbert.ne
 This model has been fine-tuned on the model : oshizo/sbert-jsnli-luke-japanese-base-lite. It is pretrained for the language : - ja.
 
 
-This model was trained on a dataset of historic Japanese companies, products, industry, addresses, and shareholders. Take a look at our paper for more details. The task is to link indices of japanese companies
+This is a (Modern) Japanese Link Transformer model  - trained on Company <SEP> Product <SEP> Industry from wiki data.
 
 ## Usage (LinkTransformer)
 
@@ -95,20 +95,20 @@ The model was trained with the parameters:
 
 **DataLoader**:
 
-`torch.utils.data.dataloader.DataLoader` of length 45 with parameters:
+`torch.utils.data.dataloader.DataLoader` of length 57 with parameters:
 ```
-{'batch_size': 64, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
+{'batch_size': 64, 'sampler': 'torch.utils.data.dataloader._InfiniteConstantSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
 ```
 
 **Loss**:
 
-`linktransformer.modified_sbert.losses.OnlineContrastiveLoss_wandb` 
+`linktransformer.modified_sbert.losses.SupConLoss_wandb` 
 
 Parameters of the fit()-Method:
 ```
 {
-    "epochs": 50,
-    "evaluation_steps": 23,
+    "epochs": 70,
+    "evaluation_steps": 29,
     "evaluator": "sentence_transformers.evaluation.SequentialEvaluator.SequentialEvaluator",
     "max_grad_norm": 1,
     "optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
@@ -117,7 +117,7 @@ Parameters of the fit()-Method:
     },
     "scheduler": "WarmupLinear",
     "steps_per_epoch": null,
-    "warmup_steps": 2250,
+    "warmup_steps": 3990,
     "weight_decay": 0.01
 }
 ```
@@ -127,7 +127,7 @@ Parameters of the fit()-Method:
 
 LinkTransformer(
   (0): Transformer({'max_seq_length': 128, 'do_lower_case': False}) with Transformer model: LukeModel 
-  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
+  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False})
 )
 ```
 

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "/mnt/122a7683-fa4b-45dd-9f13-b18cc4f4a187/deeprecordlinkage/linktransformer/models/lt-historicjapanesecompanies-comp-prod-ind_onlinecontrastive_full/",
+  "_name_or_path": "models/lt-wikidata-comp-prod-ind-ja",
   "architectures": [
     "LukeModel"
   ],

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:50c55d19867e26fc83db98c43d59b1d3af837a78e24a9d131a4953a572074379
+oid sha256:1d4a325fec886bed0d30b8007d07638a460b83f1ce43ab481dd762e350b0f13b
 size 532299592