Fine-tuned Construction Receipt Model

README.md

@@ -16,7 +16,7 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model is a fine-tuned version of [DanSarm/receipt-core-model](https://huggingface.co/DanSarm/receipt-core-model) on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Loss: 1.2568
+- Loss: 0.7435
 
 ## Model description
 
@@ -50,21 +50,21 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch | Step | Validation Loss |
 |:-------------:|:-----:|:----:|:---------------:|
-| 2.911         | 1.0   | 2    | 2.5178          |
-| 2.0298        | 2.0   | 4    | 2.0128          |
-| 1.7239        | 3.0   | 6    | 1.7704          |
-| 1.5708        | 4.0   | 8    | 1.6107          |
-| 1.4319        | 5.0   | 10   | 1.5010          |
-| 1.3463        | 6.0   | 12   | 1.4165          |
-| 1.262         | 7.0   | 14   | 1.3495          |
-| 1.2137        | 8.0   | 16   | 1.3015          |
-| 1.1761        | 9.0   | 18   | 1.2705          |
-| 1.1369        | 10.0  | 20   | 1.2568          |
+| 4.0171        | 1.0   | 2    | 2.5465          |
+| 2.1463        | 2.0   | 4    | 1.9613          |
+| 1.6495        | 3.0   | 6    | 1.5790          |
+| 1.317         | 4.0   | 8    | 1.2996          |
+| 1.0697        | 5.0   | 10   | 1.0985          |
+| 0.9027        | 6.0   | 12   | 0.9597          |
+| 0.7894        | 7.0   | 14   | 0.8566          |
+| 0.7064        | 8.0   | 16   | 0.7940          |
+| 0.6629        | 9.0   | 18   | 0.7589          |
+| 0.629         | 10.0  | 20   | 0.7435          |
 
 
 ### Framework versions
 
 - Transformers 4.48.1
-- Pytorch 2.5.1+cu124
+- Pytorch 2.6.0+cu124
 - Datasets 3.2.0
 - Tokenizers 0.21.0

config.json

@@ -4,9 +4,9 @@
     "T5ForConditionalGeneration"
   ],
   "classifier_dropout": 0.0,
-  "d_ff": 2048,
+  "d_ff": 3072,
   "d_kv": 64,
-  "d_model": 512,
+  "d_model": 768,
   "decoder_start_token_id": 0,
   "dense_act_fn": "relu",
   "dropout_rate": 0.1,
@@ -18,9 +18,9 @@
   "layer_norm_epsilon": 1e-06,
   "model_type": "t5",
   "n_positions": 512,
-  "num_decoder_layers": 6,
-  "num_heads": 8,
-  "num_layers": 6,
+  "num_decoder_layers": 12,
+  "num_heads": 12,
+  "num_layers": 12,
   "output_past": true,
   "pad_token_id": 0,
   "relative_attention_max_distance": 128,

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2a419ec4234254612766ea7d422cebfc2c63ebe3c27896ee858ecacfe8ab7aca
-size 242041896
+oid sha256:dd619dc300dd887b3caffd241d334862f7c2da82b80bda3a02660d6877fac7d5
+size 891644712

tokenizer_config.json

@@ -928,12 +928,12 @@
     "<extra_id_98>",
     "<extra_id_99>"
   ],
-  "clean_up_tokenization_spaces": true,
+  "clean_up_tokenization_spaces": false,
   "eos_token": "</s>",
   "extra_ids": 100,
   "extra_special_tokens": {},
   "max_length": 128,
-  "model_max_length": 512,
+  "model_max_length": 1000000000000000019884624838656,
   "pad_to_multiple_of": null,
   "pad_token": "<pad>",
   "pad_token_type_id": 0,