cpv2280 commited on
Commit
763b6bb
ยท
verified ยท
1 Parent(s): 75e9ce0

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +19 -17
README.md CHANGED
@@ -10,7 +10,7 @@ pinned: false
10
  license: apache-2.0
11
  short_description: Gradio
12
  ---
13
- GPT-2 Fine-Tuned TinyStories Project
14
 
15
  Overview
16
 
@@ -18,19 +18,19 @@ This project fine-tunes a GPT-2 model on the TinyStories dataset to generate str
18
 
19
  Features
20
 
21
- Story Generation: Produces coherent, child-friendly short stories.
22
 
23
- Bias Monitoring: Ensures balanced gender and cultural representation.
24
 
25
- Efficient Training: Fine-tuned on 200,000 training samples and 20,000 test samples.
26
 
27
- Grammar & Readability Improvements: Integrated grammar-checking tools and text refinement.
28
 
29
- Optimized Model Performance: Uses loss tracking, sampling techniques, and bias mitigation strategies.
30
 
31
- System Architecture
32
 
33
- The model is designed for easy interaction via Hugging Face Spaces and follows this pipeline:
34
 
35
  Data Preprocessing & Cleaning
36
 
@@ -60,7 +60,7 @@ Tracked using Weights & Biases (W&B) and TensorBoard.
60
 
61
  Analyzed validation loss and coherence checks.
62
 
63
- Getting Started
64
 
65
  Accessing the Model
66
 
@@ -76,7 +76,7 @@ Click Generate to receive an AI-generated short story.
76
 
77
  Modify the prompt and settings (temperature, top-k, top-p) for different results.
78
 
79
- Training Details
80
 
81
  Model: GPT-2 (124M)
82
 
@@ -90,7 +90,7 @@ Training Loss: 3.08 โ†’ 2.86
90
 
91
  Validation Loss: 1.46 โ†’ 1.40
92
 
93
- Evaluation & Observations
94
 
95
  Perplexity improved from 8.12 โ†’ 2.09, indicating better text fluency.
96
 
@@ -98,7 +98,7 @@ Validation loss decreased consistently, suggesting effective generalization.
98
 
99
  Human evaluation highlighted minor inconsistencies, such as abrupt scene shifts and simplistic narratives.
100
 
101
- Ethical Considerations
102
 
103
  Bias Monitoring: Pronoun analysis and diversity checks to ensure fairness.
104
 
@@ -106,7 +106,7 @@ Harmful Content Mitigation: Manually reviewed outputs for stereotypes.
106
 
107
  Text Processing Issues: UTF-8 encoding applied to prevent character errors.
108
 
109
- Future Improvements
110
 
111
  Enhancing Creativity: Fine-tune temperature and randomness settings.
112
 
@@ -114,17 +114,17 @@ Genre-Specific Training: Introduce theme-based datasets.
114
 
115
  Larger Model Training: Experiment with GPT-2 (355M) for richer storytelling.
116
 
117
- Contributors
118
 
119
  Charla Pia Vella (Project Developer)
120
 
121
  Affiliation: ARI3333 Generative AI
122
 
123
- License
124
 
125
  This project is released under the Apache-2.0 License.
126
 
127
- Acknowledgments
128
 
129
  OpenAI for GPT-2
130
 
@@ -132,7 +132,9 @@ Hugging Face for the fine-tuning framework
132
 
133
  Ronen Eldan for the TinyStories dataset
134
 
135
- For more details, visit the Hugging Face Space.
 
 
136
 
137
  Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
138
 
 
10
  license: apache-2.0
11
  short_description: Gradio
12
  ---
13
+ GPT-2 TinyStories Generator ๐Ÿข - FableWeaver AI
14
 
15
  Overview
16
 
 
18
 
19
  Features
20
 
21
+ โœ… Story Generation: Produces structured, child-friendly short stories.
22
 
23
+ โœ… Bias Monitoring: Ensures balanced gender and cultural representation.
24
 
25
+ โœ… Efficient Training: Fine-tuned on 200,000 training samples and 20,000 test samples.
26
 
27
+ โœ… Grammar & Readability Enhancements: Integrated grammar-checking tools and text refinement.
28
 
29
+ โœ… Optimized Performance: Uses loss tracking, sampling techniques, and bias mitigation strategies.
30
 
31
+ ๐Ÿš€ System Architecture
32
 
33
+ The model is designed for easy interaction via Hugging Face Spaces and follows this workflow:
34
 
35
  Data Preprocessing & Cleaning
36
 
 
60
 
61
  Analyzed validation loss and coherence checks.
62
 
63
+ ๐Ÿ”น Getting Started
64
 
65
  Accessing the Model
66
 
 
76
 
77
  Modify the prompt and settings (temperature, top-k, top-p) for different results.
78
 
79
+ ๐Ÿ“Š Training Details
80
 
81
  Model: GPT-2 (124M)
82
 
 
90
 
91
  Validation Loss: 1.46 โ†’ 1.40
92
 
93
+ ๐Ÿ“ Evaluation & Observations
94
 
95
  Perplexity improved from 8.12 โ†’ 2.09, indicating better text fluency.
96
 
 
98
 
99
  Human evaluation highlighted minor inconsistencies, such as abrupt scene shifts and simplistic narratives.
100
 
101
+ โš–๏ธ Ethical Considerations
102
 
103
  Bias Monitoring: Pronoun analysis and diversity checks to ensure fairness.
104
 
 
106
 
107
  Text Processing Issues: UTF-8 encoding applied to prevent character errors.
108
 
109
+ ๐Ÿ”ฎ Future Improvements
110
 
111
  Enhancing Creativity: Fine-tune temperature and randomness settings.
112
 
 
114
 
115
  Larger Model Training: Experiment with GPT-2 (355M) for richer storytelling.
116
 
117
+ ๐Ÿค Contributors
118
 
119
  Charla Pia Vella (Project Developer)
120
 
121
  Affiliation: ARI3333 Generative AI
122
 
123
+ ๐Ÿ“œ License
124
 
125
  This project is released under the Apache-2.0 License.
126
 
127
+ ๐ŸŽ“ Acknowledgments
128
 
129
  OpenAI for GPT-2
130
 
 
132
 
133
  Ronen Eldan for the TinyStories dataset
134
 
135
+ ๐Ÿ“Œ For more details, visit the Hugging Face Space.
136
+
137
+
138
 
139
  Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
140