Update README.md
Browse files
README.md
CHANGED
@@ -4,7 +4,7 @@ GPT3-like T5 model trained to generate text in multiple languages.
|
|
4 |
|
5 |
## Motivation
|
6 |
|
7 |
-
- GPT models are expensive run.
|
8 |
- GPT models are monolingual.
|
9 |
|
10 |
## Solution
|
@@ -15,7 +15,23 @@ GPT3-like T5 model trained to generate text in multiple languages.
|
|
15 |
I fine-tuned T5 on multiple languages (π¬π§ English, π©πͺ German, π«π· French) and multiple academic text snippets from
|
16 |
various domains like tech, law, finance and science etc. to generate text, just like GPT models do.
|
17 |
|
18 |
-
## Usage
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
|
20 |
- Provide some text e.g `"Italy, officially the Italian Republic is a country consisting of"`
|
21 |
- Tell Cheapity3 how many words you want to generate e.g `15` -- π Yes, you can control the length.
|
@@ -50,17 +66,19 @@ outputs = model.generate(
|
|
50 |
|
51 |
for i in range(4):
|
52 |
print(tokenizer.decode(outputs[i], skip_special_tokens=True, clean_up_tokenization_spaces=True))
|
|
|
53 |
|
54 |
-
|
55 |
|
56 |
-
|
57 |
-
|
|
|
58 |
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
```
|
65 |
|
66 |
## Pretty decent right?
|
@@ -69,9 +87,9 @@ Hence, whenever you feel like GPT3 is too expensive, Cheapity3 comes to the resc
|
|
69 |
|
70 |
## Model Training FYI
|
71 |
- T5-base model
|
72 |
-
- Trained on
|
73 |
-
- Mostly text from
|
74 |
- Learning rate: 0.00003
|
75 |
- 2 epochs
|
76 |
- Max input: 512 tokens
|
77 |
-
- Max output: 128 tokens
|
|
|
4 |
|
5 |
## Motivation
|
6 |
|
7 |
+
- GPT models are expensive to run.
|
8 |
- GPT models are monolingual.
|
9 |
|
10 |
## Solution
|
|
|
15 |
I fine-tuned T5 on multiple languages (π¬π§ English, π©πͺ German, π«π· French) and multiple academic text snippets from
|
16 |
various domains like tech, law, finance and science etc. to generate text, just like GPT models do.
|
17 |
|
18 |
+
## Usage - [NLPlayStore](https://github.com/flexudy/NLPlayStore)π
|
19 |
+
|
20 |
+
```python
|
21 |
+
from store.service_management import ServiceManager
|
22 |
+
|
23 |
+
service_manager = ServiceManager().get_service("cheapity3")
|
24 |
+
|
25 |
+
service.install()
|
26 |
+
|
27 |
+
service = service.launch()
|
28 |
+
|
29 |
+
input_text = "The mechanical engineering field requires an understanding of core areas including mechanics, dynamics, thermodynamics, materials science, structural analysis, and electricity."
|
30 |
+
|
31 |
+
generated_texts = service.play(input_text, 15)
|
32 |
+
```
|
33 |
+
|
34 |
+
## Usage - Hugging Face Transformers π€
|
35 |
|
36 |
- Provide some text e.g `"Italy, officially the Italian Republic is a country consisting of"`
|
37 |
- Tell Cheapity3 how many words you want to generate e.g `15` -- π Yes, you can control the length.
|
|
|
66 |
|
67 |
for i in range(4):
|
68 |
print(tokenizer.decode(outputs[i], skip_special_tokens=True, clean_up_tokenization_spaces=True))
|
69 |
+
```
|
70 |
|
71 |
+
**INPUT: The mechanical engineering field requires an understanding of core areas including mechanics, dynamics, thermodynamics, materials science, structural analysis, and electricity.**
|
72 |
|
73 |
+
```
|
74 |
+
> Cheapity3 continues with beam search:
|
75 |
+
... The field of mechanical engineering is a broad field that includes many core areas of engineering.
|
76 |
|
77 |
+
> Cheapity3 continues with sampling and top_k=50:
|
78 |
+
... Developing the knowledge base for these core areas will enable engineers to build their capabilities rapidly and efficiently. ...
|
79 |
+
... The field of mechanics offers a variety and broad range for applications throughout the engineering/technological fields. ...
|
80 |
+
... Mechanics generally is not understood by students. While they can be employed in the field, mechanical engineering ...
|
81 |
+
... Introduction to mechanical engineering and core fields including chemical products, materials science, structural analysis, and geomatics ...
|
82 |
```
|
83 |
|
84 |
## Pretty decent right?
|
|
|
87 |
|
88 |
## Model Training FYI
|
89 |
- T5-base model
|
90 |
+
- Trained on ONLY 1M sentences from English, French and German text
|
91 |
+
- Mostly text from Wikipedia, arxiv and QA datasets
|
92 |
- Learning rate: 0.00003
|
93 |
- 2 epochs
|
94 |
- Max input: 512 tokens
|
95 |
+
- Max output: 128 tokens
|