agentlans commited on
Commit
79cb92c
·
verified ·
1 Parent(s): 7da4ac3

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +56 -79
README.md CHANGED
@@ -4,106 +4,83 @@ language:
4
  license: apache-2.0
5
  base_model: google/flan-t5-small
6
  tags:
7
- - generated_from_trainer
 
 
8
  datasets:
9
- - sentence-paraphrases
10
- model-index:
11
- - name: flan-t5-small-simplifier
12
- results: []
13
  ---
 
14
 
15
- <!-- This model card has been generated automatically according to the information the Trainer had access to. You
16
- should probably proofread and complete it, then remove this comment. -->
17
 
18
- # flan-t5-small-simplifier
19
 
20
- For paraphrasing and simplifying English text.
 
 
21
 
22
- Fine-tuned version of [google/flan-t5-small](https://huggingface.co/google/flan-t5-small) on the agentlans/sentence-paraphrases dataset.
23
- It achieves the following results on the evaluation set:
24
- - Loss: 1.1518
25
- - Num Input Tokens Seen: 32939232
26
 
27
- ## Intended uses & limitations
 
 
 
28
 
29
- Works best on sentence length texts.
30
 
31
- ```
32
- import torch
33
- from transformers import pipeline
34
-
35
- # Check if GPU is available
36
- device = 0 if torch.cuda.is_available() else -1
37
-
38
- # Initialize the pipeline
39
- model_name = "agentlans/flan-t5-small-simplifier"
40
- flan_t5_pipeline = pipeline("text2text-generation", model=model_name, device=device)
41
-
42
- # Example input
43
- input_text = "While navigating the labyrinthine corridors of epistemological uncertainty, the precocious philosopher—whose seminal work on phenomenological interpretation had already garnered significant academic acclaim—paused momentarily to contemplate the intricate interplay between subjective perception and objective reality, ultimately recognizing that the boundaries of human understanding are perpetually fluid and dynamically reconstructed through continuous intellectual discourse and empirical investigation."
44
-
45
- # Generate output
46
- output = flan_t5_pipeline(input_text, max_length=1024)
47
-
48
- # Print the result
49
- print(output[0]["generated_text"])
50
- # The precocious philosopher, who had already been a major academic acclaim for his seminal work on phenomenological interpretation, paused momentarily to contemplate the intricate interplay between subjective perception and objective reality, recognizing that the boundaries of human understanding are perpetually fluid and dynamically reconstructed through continuous intellectual discourse and empirical investigation.
51
- ```
52
 
53
- Limitations:
54
- - English only
55
- - Doesn't handle mixed language texts well (for example, English with Greek letter words)
56
- - Might not be able to simplify some texts
57
 
58
- ## Training and evaluation data
59
 
60
- agentlans/sentence-paraphrases
61
-
62
- This dataset is a curated collection of sentence-length paraphrases derived from two primary sources:
63
-
64
- humarin/chatgpt-paraphrases
65
- xwjzds/paraphrase_collections.
66
-
67
- Dataset Details
68
- Dataset Description
69
-
70
- The dataset is structured to provide pairs of sentences from an original text and its paraphrase(s). For each entry:
71
 
72
- The "text" field contains the least readable paraphrase.
73
- The "paraphrase" field contains the most readable paraphrase.
 
74
 
75
- Readability was assessed using the agentlans/deberta-v3-xsmall-zyda-2-readability model.
76
 
77
- ## Training procedure
 
78
 
79
- ### Training hyperparameters
 
80
 
81
- The following hyperparameters were used during training:
82
- - learning_rate: 5e-05
83
- - train_batch_size: 8
84
- - eval_batch_size: 8
85
- - seed: 42
86
- - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
87
- - lr_scheduler_type: linear
88
- - num_epochs: 2.0
89
 
90
- ### Training results
 
 
91
 
92
- | Training Loss | Epoch | Step | Validation Loss | Input Tokens Seen |
93
- |:-------------:|:------:|:-----:|:---------------:|:-----------------:|
94
- | 1.4423 | 0.2224 | 10000 | 1.2431 | 3655312 |
95
- | 1.3884 | 0.4448 | 20000 | 1.2093 | 7331520 |
96
- | 1.3782 | 0.6673 | 30000 | 1.1859 | 10990432 |
97
- | 1.3595 | 0.8897 | 40000 | 1.1787 | 14653328 |
98
- | 1.3059 | 1.1121 | 50000 | 1.1665 | 18326104 |
99
- | 1.3298 | 1.3345 | 60000 | 1.1589 | 21991016 |
100
- | 1.2994 | 1.5569 | 70000 | 1.1562 | 25656600 |
101
- | 1.2952 | 1.7794 | 80000 | 1.1518 | 29314808 |
102
 
 
 
 
 
 
 
103
 
104
- ### Framework versions
105
 
106
  - Transformers 4.43.3
107
- - Pytorch 2.3.0+cu121
108
  - Datasets 3.2.0
109
- - Tokenizers 0.19.1
 
 
 
 
4
  license: apache-2.0
5
  base_model: google/flan-t5-small
6
  tags:
7
+ - text-simplification
8
+ - paraphrase
9
+ - natural-language-processing
10
  datasets:
11
+ - agentlans/sentence-paraphrases
 
 
 
12
  ---
13
+ # FLAN-T5 Small Simplifier
14
 
15
+ A fine-tuned text simplification and paraphrasing model based on Google's FLAN-T5 Small, designed to enhance text readability while preserving core semantic meaning.
 
16
 
17
+ ## Model Details
18
 
19
+ - **Base Model**: [google/flan-t5-small](https://huggingface.co/google/flan-t5-small)
20
+ - **Task**: Text Simplification and Paraphrasing
21
+ - **Languages**: English
22
 
23
+ ## Capabilities
 
 
 
24
 
25
+ The model is specialized in:
26
+ - Reducing text complexity
27
+ - Generating more readable paraphrases
28
+ - Maintaining original semantic content
29
 
30
+ ## Intended Use
31
 
32
+ **Primary Use Cases**:
33
+ - Academic writing simplification
34
+ - Technical document readability enhancement
35
+ - Content adaptation for diverse audiences
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
36
 
37
+ **Limitations**:
38
+ - Optimized for English language texts
39
+ - Best performance on sentence-length inputs
40
+ - May struggle with highly specialized or mixed-language texts
41
 
42
+ ## Usage Example
43
 
44
+ ```python
45
+ from transformers import pipeline
 
 
 
 
 
 
 
 
 
46
 
47
+ simplifier = pipeline(
48
+ "text2text-generation", model="agentlans/flan-t5-small-simplifier"
49
+ )
50
 
51
+ complex_text = "While navigating the labyrinthine corridors of epistemological uncertainty, the precocious philosopher paused to contemplate the intricate interplay between subjective perception and objective reality."
52
 
53
+ simplified_text = simplifier(complex_text, max_length=128)[0]["generated_text"]
54
+ print(simplified_text)
55
 
56
+ # The precocious philosopher paused to contemplate the complex interplay between subjective perception and objective reality while navigating the labyrinthine corridors of epistemological uncertainty.
57
+ ```
58
 
59
+ ## Training Details
 
 
 
 
 
 
 
60
 
61
+ **Dataset**: [agentlans/sentence-paraphrases](https://huggingface.co/datasets/agentlans/sentence-paraphrases)
62
+ - Source: Curated paraphrase collections
63
+ - Readability assessment using a finetuned [DeBERTa v3 XSmall](https://huggingface.co/agentlans/deberta-v3-xsmall-zyda-2-readability)
64
 
65
+ **Training Hyperparameters**:
66
+ - Learning Rate: 5e-05
67
+ - Batch Size: 8
68
+ - Optimizer: Adam
69
+ - Epochs: 2.0
 
 
 
 
 
70
 
71
+ **Performance Metrics**:
72
+ | Epoch | Training Loss | Validation Loss |
73
+ |:-----:|:-------------:|:---------------:|
74
+ | 0.22 | 1.4423 | 1.2431 |
75
+ | 0.89 | 1.3595 | 1.1787 |
76
+ | 1.78 | 1.2952 | 1.1518 |
77
 
78
+ ## Framework
79
 
80
  - Transformers 4.43.3
81
+ - PyTorch 2.3.0+cu121
82
  - Datasets 3.2.0
83
+
84
+ ## Ethical Considerations
85
+
86
+ Users should review generated text for accuracy and appropriateness, as the model may inherit biases from training data.