Hugging Face's logo

fodmrldg
/
news_summarization_model

Transformers
Safetensors
t5
text2text-generation
Generated from Trainer
text-generation-inference
Model card Files
xet
 Community
fodmrldg commited on
Commit
2204b04
·
verified ·
1 Parent(s): d0cdf24

End of training

Browse files
Files changed (7) hide show
  1. README.md +14 -18
  2. config.json +5 -5
  3. generation_config.json +0 -1
  4. merges.txt +0 -0
  5. model.safetensors +2 -2
  6. training_args.bin +1 -1
  7. vocab.json +0 -0
README.md CHANGED
@@ -1,6 +1,6 @@
1
  ---
2
  library_name: transformers
3
- base_model: eenzeenee/t5-small-korean-summarization
4
  tags:
5
  - generated_from_trainer
6
  metrics:
@@ -15,14 +15,14 @@ should probably proofread and complete it, then remove this comment. -->
15
 
16
  # news_summarization_model
17
 
18
- This model is a fine-tuned version of [eenzeenee/t5-small-korean-summarization](https://huggingface.co/eenzeenee/t5-small-korean-summarization) on the None dataset.
19
  It achieves the following results on the evaluation set:
20
- - Loss: 0.2389
21
- - Rouge1: 0.5209
22
- - Rouge2: 0.3501
23
- - Rougel: 0.5158
24
- - Rougelsum: 0.5203
25
- - Gen Len: 21.8986
26
 
27
  ## Model description
28
 
@@ -47,22 +47,18 @@ The following hyperparameters were used during training:
47
  - seed: 42
48
  - optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
49
  - lr_scheduler_type: linear
50
- - lr_scheduler_warmup_steps: 500
51
- - num_epochs: 3
52
  - mixed_precision_training: Native AMP
53
 
54
  ### Training results
55
 
56
  | Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Rougelsum | Gen Len |
57
  |:-------------:|:------:|:----:|:---------------:|:------:|:------:|:------:|:---------:|:-------:|
58
- | 0.393 | 0.3623 | 50 | 0.2685 | 0.5214 | 0.3586 | 0.5161 | 0.5197 | 22.1522 |
59
- | 0.3787 | 0.7246 | 100 | 0.2603 | 0.5080 | 0.3478 | 0.5039 | 0.5089 | 22.0507 |
60
- | 0.3528 | 1.0870 | 150 | 0.2572 | 0.5164 | 0.3515 | 0.5121 | 0.5157 | 22.0290 |
61
- | 0.2731 | 1.4493 | 200 | 0.2494 | 0.5195 | 0.3514 | 0.5140 | 0.5185 | 22.0 |
62
- | 0.3165 | 1.8116 | 250 | 0.2475 | 0.5149 | 0.3513 | 0.5108 | 0.5143 | 21.9348 |
63
- | 0.2831 | 2.1739 | 300 | 0.2375 | 0.5216 | 0.3614 | 0.5164 | 0.5216 | 21.8913 |
64
- | 0.2427 | 2.5362 | 350 | 0.2393 | 0.5159 | 0.3631 | 0.5114 | 0.5162 | 21.8478 |
65
- | 0.2733 | 2.8986 | 400 | 0.2401 | 0.5122 | 0.3471 | 0.5077 | 0.5130 | 21.9058 |
66
 
67
 
68
  ### Framework versions
 
1
  ---
2
  library_name: transformers
3
+ base_model: eenzeenee/t5-base-korean-summarization
4
  tags:
5
  - generated_from_trainer
6
  metrics:
 
15
 
16
  # news_summarization_model
17
 
18
+ This model is a fine-tuned version of [eenzeenee/t5-base-korean-summarization](https://huggingface.co/eenzeenee/t5-base-korean-summarization) on the None dataset.
19
  It achieves the following results on the evaluation set:
20
+ - Loss: 0.2221
21
+ - Rouge1: 0.6031
22
+ - Rouge2: 0.4193
23
+ - Rougel: 0.5896
24
+ - Rougelsum: 0.5892
25
+ - Gen Len: 42.2714
26
 
27
  ## Model description
28
 
 
47
  - seed: 42
48
  - optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
49
  - lr_scheduler_type: linear
50
+ - lr_scheduler_warmup_steps: 400
51
+ - num_epochs: 2
52
  - mixed_precision_training: Native AMP
53
 
54
  ### Training results
55
 
56
  | Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Rougelsum | Gen Len |
57
  |:-------------:|:------:|:----:|:---------------:|:------:|:------:|:------:|:---------:|:-------:|
58
+ | 0.523 | 0.4762 | 100 | 0.3847 | 0.4833 | 0.2872 | 0.4627 | 0.4654 | 41.7571 |
59
+ | 0.3377 | 0.9524 | 200 | 0.2590 | 0.5709 | 0.3929 | 0.5581 | 0.5564 | 41.7810 |
60
+ | 0.2641 | 1.4286 | 300 | 0.2334 | 0.5889 | 0.4018 | 0.5785 | 0.5781 | 42.5048 |
61
+ | 0.2361 | 1.9048 | 400 | 0.2221 | 0.6031 | 0.4193 | 0.5896 | 0.5892 | 42.2714 |
 
 
 
 
62
 
63
 
64
  ### Framework versions
config.json CHANGED
@@ -3,9 +3,9 @@
3
  "T5ForConditionalGeneration"
4
  ],
5
  "classifier_dropout": 0.0,
6
- "d_ff": 1024,
7
  "d_kv": 64,
8
- "d_model": 512,
9
  "decoder_start_token_id": 0,
10
  "dense_act_fn": "gelu_new",
11
  "dropout_rate": 0.1,
@@ -17,9 +17,9 @@
17
  "layer_norm_epsilon": 1e-06,
18
  "max_length": null,
19
  "model_type": "t5",
20
- "num_decoder_layers": 8,
21
- "num_heads": 6,
22
- "num_layers": 8,
23
  "output_past": true,
24
  "pad_token_id": 0,
25
  "relative_attention_max_distance": 128,
 
3
  "T5ForConditionalGeneration"
4
  ],
5
  "classifier_dropout": 0.0,
6
+ "d_ff": 2048,
7
  "d_kv": 64,
8
+ "d_model": 768,
9
  "decoder_start_token_id": 0,
10
  "dense_act_fn": "gelu_new",
11
  "dropout_rate": 0.1,
 
17
  "layer_norm_epsilon": 1e-06,
18
  "max_length": null,
19
  "model_type": "t5",
20
+ "num_decoder_layers": 12,
21
+ "num_heads": 12,
22
+ "num_layers": 12,
23
  "output_past": true,
24
  "pad_token_id": 0,
25
  "relative_attention_max_distance": 128,
generation_config.json CHANGED
@@ -1,5 +1,4 @@
1
  {
2
- "_from_model_config": true,
3
  "decoder_start_token_id": 0,
4
  "eos_token_id": 1,
5
  "max_length": 128,
 
1
  {
 
2
  "decoder_start_token_id": 0,
3
  "eos_token_id": 1,
4
  "max_length": 128,
merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ab345c7cc1efa33438911fc82448a28a5f4ac8dbafe203d451c1ceb2ec7a0076
3
- size 382537128
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:064c35d02b6c433c6704f2d4b8cce362ada1f38147ff483598f1d08ac494296f
3
+ size 1102350184
training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5844f8b022e5f9a86867f3753ee99c3e80e51cc5ec22a692fdc1edb59a08a1d2
3
  size 5432
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f4dd0cac597e48dccb9af31955f7458fcb1cc354c87694f889b7ee012ed16cc9
3
  size 5432
vocab.json ADDED
The diff for this file is too large to render. See raw diff