abhaskumarsinha
commited on
Commit
•
31a038a
1
Parent(s):
2eede07
Create README.md
Browse files
README.md
CHANGED
@@ -1,13 +1,65 @@
|
|
1 |
---
|
2 |
-
title: MinimalGPT-Felis Catus
|
3 |
-
emoji: 🏢
|
4 |
-
colorFrom: green
|
5 |
-
colorTo: yellow
|
6 |
-
sdk: gradio
|
7 |
-
sdk_version: 3.34.0
|
8 |
-
app_file: app.py
|
9 |
-
pinned: false
|
10 |
license: mit
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
---
|
12 |
|
13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
license: mit
|
3 |
+
title: 'MinimalGPT: Felis Catus'
|
4 |
+
sdk: gradio
|
5 |
+
emoji: 😻
|
6 |
+
colorFrom: gray
|
7 |
+
colorTo: blue
|
8 |
+
pinned: true
|
9 |
---
|
10 |
|
11 |
+
# MinimalGPT: Felis Catus
|
12 |
+
|
13 |
+
[[`MinimalGPT`](https://github.com/abhaskumarsinha/MinimalGPT)] [[`Project Gutenberg Dataset`](https://www.kaggle.com/datasets/shubchat/1002-short-stories-from-project-guttenberg)]
|
14 |
+
|
15 |
+
|
16 |
+
This HuggingFace space serves as an illustrative application of the GitHub Repository: [MinimalGPT](https://github.com/abhaskumarsinha/MinimalGPT), which embodies a departure from conventional GPT models that undergo scaling and training on high-performance computing systems and clusters. The primary objective of the MinimalGPT project was to explore the extent to which a GPT model could be minimized in size.
|
17 |
+
|
18 |
+
Within this HF space, we introduce a diminutive GPT model named [Felis Catus](https://en.wikipedia.org/wiki/Cat) (stray Cat), which boasts a mere 15 million parameters. What distinguishes this model is its training process, which was executed on a standard home computer CPU (specifically, an AMD Ryzen 5) without any reliance on GPU acceleration. Remarkably, the training duration lasted a mere 15 minutes, utilizing a dataset comprising a meager ~150,000 tokens of text.
|
19 |
+
|
20 |
+
At present, the Felis Catus model exhibits the capacity to generate a concise story excerpt consisting of 70 tokens, requiring a mere 5 to 7 words as input. The model's dictionary encompasses a modest 12,000 words. Moreover, we are presently engaged in endeavors to further scale the model in our forthcoming project.
|
21 |
+
|
22 |
+
## Model Specifications
|
23 |
+
|
24 |
+
```
|
25 |
+
Model: "model"
|
26 |
+
_________________________________________________________________
|
27 |
+
Layer (type) Output Shape Param #
|
28 |
+
=================================================================
|
29 |
+
input_1 (InputLayer) [(None, 10)] 0
|
30 |
+
|
31 |
+
embedding (Embedding) (None, 10, 128) 1597184
|
32 |
+
|
33 |
+
positional_embedding (Posit (None, 10, 128) 0
|
34 |
+
ionalEmbedding)
|
35 |
+
|
36 |
+
decoder (Decoder) (None, 10, 128) 71208
|
37 |
+
|
38 |
+
flatten (Flatten) (None, 1280) 0
|
39 |
+
|
40 |
+
dense (Dense) (None, 12479) 15985599
|
41 |
+
|
42 |
+
tf.nn.softmax (TFOpLambda) (None, 12479) 0
|
43 |
+
|
44 |
+
=================================================================
|
45 |
+
Total params: 17,653,991
|
46 |
+
Trainable params: 17,653,991
|
47 |
+
Non-trainable params: 0
|
48 |
+
_________________________________________________________________
|
49 |
+
```
|
50 |
+
|
51 |
+
## Hyperparameters
|
52 |
+
|
53 |
+
```
|
54 |
+
gpt_input: 10 [Max input size, d_k]
|
55 |
+
d_model: 128 [Embedding size, d_model]
|
56 |
+
h: 8 [Number of multiheads, h]
|
57 |
+
decoder_stacks: 1 [Number of decoder stacks, stack]
|
58 |
+
GPT_attention: True [Attention Layer implementation type - OpenAI style]
|
59 |
+
```
|
60 |
+
|
61 |
+
## References
|
62 |
+
1. Vaswani, Ashish, et al. "Attention is all you need." Advances in neural information processing systems 30 (2017).
|
63 |
+
2. Radford, Alec, et al. "Language models are unsupervised multitask learners." OpenAI blog 1.8 (2019): 9.
|
64 |
+
3. Project Gutenberg. (n.d.). Retrieved FebruApril 20, 2023, from www.gutenberg.org.
|
65 |
+
4. Abadi, Martın, et al. "TensorFlow: Large-scale machine learning on heterogeneous systems, software available from tensorflow. org (2015)." URL https://www.tensorflow.org (2015).
|