End of training

README.md

@@ -1,27 +1,21 @@
 ---
 tags:
 - generated_from_trainer
-- trocr
 model-index:
 - name: khmer-trocr-base-printed
   results: []
-license: mit
-language:
-- km
-library_name: transformers
-pipeline_tag: image-to-text
 ---
 
 <!-- This model card has been generated automatically according to the information the Trainer had access to. You
 should probably proofread and complete it, then remove this comment. -->
 
-[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/rayranger/huggingface/runs/hpngjitd)
+[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/rayranger/huggingface/runs/hddo3082)
 # khmer-trocr-base-printed
 
 This model is a fine-tuned version of [](https://huggingface.co/) on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.4627
-- Cer: 0.7359
+- Loss: 0.1980
+- Cer: 0.5955
 
 ## Model description
 
@@ -46,24 +40,23 @@ The following hyperparameters were used during training:
 - seed: 42
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: linear
-- num_epochs: 3
+- num_epochs: 5
 - mixed_precision_training: Native AMP
 
 ### Training results
 
 | Training Loss | Epoch  | Step  | Validation Loss | Cer    |
 |:-------------:|:------:|:-----:|:---------------:|:------:|
-| 1.0771        | 0.2612 | 1000  | 1.0644          | 0.8631 |
-| 1.0182        | 0.5223 | 2000  | 0.9849          | 0.8791 |
-| 0.9409        | 0.7835 | 3000  | 0.9387          | 0.8704 |
-| 0.9053        | 1.0447 | 4000  | 0.8917          | 0.8646 |
-| 0.8428        | 1.3058 | 5000  | 0.8280          | 0.8600 |
-| 0.729         | 1.5670 | 6000  | 0.7200          | 0.8192 |
-| 0.6011        | 1.8282 | 7000  | 0.6203          | 0.7893 |
-| 0.5268        | 2.0893 | 8000  | 0.5568          | 0.7621 |
-| 0.4819        | 2.3505 | 9000  | 0.5150          | 0.7554 |
-| 0.4362        | 2.6116 | 10000 | 0.4799          | 0.7412 |
-| 0.4555        | 2.8728 | 11000 | 0.4627          | 0.7359 |
+| 1.1417        | 0.4876 | 1000  | 1.0816          | 0.8567 |
+| 1.031         | 0.9751 | 2000  | 0.9824          | 0.8652 |
+| 0.9015        | 1.4627 | 3000  | 0.8875          | 0.8421 |
+| 0.7111        | 1.9503 | 4000  | 0.6645          | 0.7871 |
+| 0.5049        | 2.4378 | 5000  | 0.4831          | 0.7234 |
+| 0.4108        | 2.9254 | 6000  | 0.3594          | 0.6712 |
+| 0.2876        | 3.4130 | 7000  | 0.3076          | 0.6458 |
+| 0.2163        | 3.9005 | 8000  | 0.2418          | 0.6214 |
+| 0.1862        | 4.3881 | 9000  | 0.2119          | 0.5998 |
+| 0.1933        | 4.8757 | 10000 | 0.1980          | 0.5955 |
 
 
 ### Framework versions
@@ -71,4 +64,4 @@ The following hyperparameters were used during training:
 - Transformers 4.42.3
 - Pytorch 2.1.2
 - Datasets 2.20.0
-- Tokenizers 0.19.1
+- Tokenizers 0.19.1

config.json

@@ -3,7 +3,7 @@
     "VisionEncoderDecoderModel"
   ],
   "decoder": {
-    "_name_or_path": "channudam/khmer-roberta-base",
+    "_name_or_path": "/kaggle/input/khmerrobertamlm/KhmerRobertaMLM",
     "add_cross_attention": true,
     "architectures": [
       "RobertaForMaskedLM"
@@ -85,7 +85,7 @@
   "decoder_start_token_id": 0,
   "early_stopping": true,
   "encoder": {
-    "_name_or_path": "facebook/deit-base-distilled-patch16-224",
+    "_name_or_path": "facebook/deit-base-distilled-patch16-384",
     "add_cross_attention": false,
     "architectures": [
       "DeiTForImageClassificationWithTeacher"
@@ -1112,7 +1112,7 @@
       "998": "ear, spike, capitulum",
       "999": "toilet tissue, toilet paper, bathroom tissue"
     },
-    "image_size": 224,
+    "image_size": 384,
     "initializer_range": 0.02,
     "intermediate_size": 3072,
     "is_decoder": false,
@@ -2161,7 +2161,7 @@
   "eos_token_id": 264,
   "is_encoder_decoder": true,
   "length_penalty": 2.0,
-  "max_length": 128,
+  "max_length": 145,
   "model_type": "vision-encoder-decoder",
   "no_repeat_ngram_size": 3,
   "num_beams": 4,

generation_config.json

@@ -4,7 +4,7 @@
   "early_stopping": true,
   "eos_token_id": 264,
   "length_penalty": 2.0,
-  "max_length": 128,
+  "max_length": 145,
   "no_repeat_ngram_size": 3,
   "num_beams": 4,
   "pad_token_id": 1,

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:55cbbf1b978618866b4a4f2ecdd35c982f1fb5c1a3607bc7f09bfcea817f06b2
-size 607168248
+oid sha256:91c126cc91712d1c91445786293cf542cbd0ba9ac2d40c950a5d640885803615
+size 608335608

tokenizer_config.json

@@ -48,8 +48,8 @@
   "eos_token": "</s>",
   "errors": "replace",
   "mask_token": "<mask>",
-  "max_len": 128,
-  "model_max_length": 128,
+  "max_len": 150,
+  "model_max_length": 150,
   "pad_token": "<pad>",
   "sep_token": "</s>",
   "tokenizer_class": "RobertaTokenizer",

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b4d9857cda86526b31e08dfdbae7f5e3b6f0f0537ae19f0407f84c6ebd8097b9
+oid sha256:016c3c28dbc67f7a85bbc1e4279553dae73e15da096e6a19b0186beb0a625b8f
 size 5240