Update README.md
Browse files
README.md
CHANGED
@@ -18,122 +18,128 @@ tags:
|
|
18 |
---
|
19 |
**model_name: 169Pi/generic_slm**
|
20 |
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
**
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
---
|
19 |
**model_name: 169Pi/generic_slm**
|
20 |
|
21 |
+
## Model Description
|
22 |
+
|
23 |
+
The **169Pi/generic_slm** is a fine-tuned version of the Meta-Llama-3.1-8B-bnb-4bit model, designed to deliver high-quality educational content. Leveraging techniques such as LoRA, PEFT, and RSLoRA, it aims to provide engaging, accurate, and contextually appropriate educational materials for students and educators.
|
24 |
+
|
25 |
+
## Tags
|
26 |
+
|
27 |
+
- transformers
|
28 |
+
- llama
|
29 |
+
- education
|
30 |
+
- fine-tuning
|
31 |
+
- LoRA
|
32 |
+
- PEFT
|
33 |
+
- RSLoRA
|
34 |
+
- quantized
|
35 |
+
|
36 |
+
## Uses
|
37 |
+
|
38 |
+
### Direct Use
|
39 |
+
- Summarizing chapters or concepts
|
40 |
+
- Answering curriculum-aligned questions
|
41 |
+
- Generating practice questions and explanations
|
42 |
+
- Recommending study materials
|
43 |
+
|
44 |
+
### Downstream Use
|
45 |
+
- Interactive learning tools
|
46 |
+
- Educational chatbots
|
47 |
+
- Personalized study guides
|
48 |
+
- Automated assessment materials
|
49 |
+
|
50 |
+
### Out of Scope
|
51 |
+
- Legal or financial decision-making
|
52 |
+
- Generating non-educational content
|
53 |
+
- Applications requiring high precision in non-educational contexts
|
54 |
+
|
55 |
+
## Training Details
|
56 |
+
|
57 |
+
### Dataset
|
58 |
+
Proprietary dataset by 169Pi
|
59 |
+
|
60 |
+
### Preprocessing Steps
|
61 |
+
- Removed duplicates
|
62 |
+
- Cleaned noisy and irrelevant data
|
63 |
+
- Normalized text for consistency
|
64 |
+
|
65 |
+
### Parameter Size
|
66 |
+
4.65 billion (quantized to 4-bit)
|
67 |
+
|
68 |
+
### Hyperparameters
|
69 |
+
- **learning_rate**: 5e-5
|
70 |
+
- **lr_scheduler_type**: cosine
|
71 |
+
- **batch_size_per_device**: 32
|
72 |
+
- **gradient_accumulation_steps**: 4
|
73 |
+
- **num_epochs**: 3
|
74 |
+
- **fp16**: True
|
75 |
+
- **bf16**: True
|
76 |
+
- **optimizer**: adamw_8bit
|
77 |
+
- **weight_decay**: 0.05
|
78 |
+
- **warmup_steps**: 1000
|
79 |
+
- **logging_steps**: 1000
|
80 |
+
- **evaluation_strategy**: steps
|
81 |
+
- **eval_steps**: 1000
|
82 |
+
- **save_strategy**: steps
|
83 |
+
- **save_steps**: 1000
|
84 |
+
|
85 |
+
## Architecture
|
86 |
+
|
87 |
+
### Base Model
|
88 |
+
Meta-Llama-3.1-8B
|
89 |
+
|
90 |
+
### Quantization
|
91 |
+
4-bit
|
92 |
+
|
93 |
+
### Techniques
|
94 |
+
- LoRA
|
95 |
+
- PEFT
|
96 |
+
- RSLoRA
|
97 |
+
|
98 |
+
## Bias, Risks, and Limitations
|
99 |
+
|
100 |
+
### Known Biases
|
101 |
+
Potential biases in educational content sources, including cultural or linguistic preferences.
|
102 |
+
|
103 |
+
### Risks
|
104 |
+
Model may generate incorrect or general responses for ambiguous queries.
|
105 |
+
|
106 |
+
### Recommendations
|
107 |
+
Use cautiously in critical contexts. Regularly evaluate outputs for accuracy and bias.
|
108 |
+
|
109 |
+
## Technical Specifications
|
110 |
+
|
111 |
+
### Model Architecture
|
112 |
+
Transformer-based architecture with multi-head self-attention, enhanced using LoRA, PEFT, and RSLoRA. Optimized for educational tasks.
|
113 |
+
|
114 |
+
### Objective
|
115 |
+
Generate high-quality educational content, including summarization, question-answering, and study material generation.
|
116 |
+
|
117 |
+
## Evaluation
|
118 |
+
|
119 |
+
### Metrics
|
120 |
+
- **Primary**: Loss during training
|
121 |
+
- **Secondary**: Accuracy and relevance through manual evaluation
|
122 |
+
|
123 |
+
### Results
|
124 |
+
Achieved low validation loss during training, demonstrating generalization capability.
|
125 |
+
|
126 |
+
## Environmental Impact
|
127 |
+
|
128 |
+
- **Hardware**: NVIDIA A100
|
129 |
+
- **Training Duration**: 26 hours
|
130 |
+
|
131 |
+
## Citation
|
132 |
+
|
133 |
+
```bibtex
|
134 |
+
@misc{169Pi_generic_slm,
|
135 |
+
title={169Pi/generic_slm: Fine-Tuned Educational Model},
|
136 |
+
author={169Pi},
|
137 |
+
year={2024},
|
138 |
+
publisher={Hugging Face},
|
139 |
+
url={https://huggingface.co/169Pi/generic_slm}
|
140 |
+
}
|
141 |
+
|
142 |
+
## Contact
|
143 |
+
|
144 |
+
- **Developer**: 169Pi AI
|
145 |
+
- **Email**: [[email protected]](mailto:[email protected])
|