Update README.md
Browse files
README.md
CHANGED
@@ -1,14 +1,12 @@
|
|
1 |
# SliceX AI™ ELM (Efficient Language Models)
|
2 |
This repository contains code to run our ELM models.
|
3 |
|
4 |
-
Models are located in the "models" folder. ELM models in this repository comes in three sizes (elm-1.0, elm-0.75 and elm-0.25) and supports the following use-
|
5 |
-
- news_classification
|
6 |
-
- toxicity_detection
|
7 |
- news_content_generation
|
8 |
|
9 |
## Download ELM repo
|
10 |
```bash
|
11 |
-
git clone [email protected]:slicexai/elm-
|
12 |
sudo apt-get intall git-lfs
|
13 |
git lfs install
|
14 |
```
|
@@ -20,11 +18,11 @@ PATH=$PATH:/<absolute-path>/git-lfs-3.2.0/
|
|
20 |
git lfs install
|
21 |
```
|
22 |
|
23 |
-
## Download ELM
|
24 |
```bash
|
25 |
-
cd elm-
|
26 |
-
git lfs pull -I models/elm-
|
27 |
-
git lfs pull -I models/elm-0.
|
28 |
git lfs pull -I models/elm-0.25_news_content_generation/ckpt.pt
|
29 |
```
|
30 |
|
@@ -36,13 +34,13 @@ pip install -r requirements.txt
|
|
36 |
## How to use - Run ELM on a sample task (e.g., news classification)
|
37 |
```bash
|
38 |
python run.py <elm-model-directory>
|
39 |
-
E.g. python run.py models/elm-0.
|
40 |
```
|
41 |
-
Prompts for the specific tasks can be found in the corresponding checkpoint directory. See an example below in the form of `models/elm-0.
|
42 |
```json
|
43 |
{
|
44 |
-
"inputs": ["
|
45 |
-
"template": "[INST]
|
46 |
}
|
47 |
```
|
48 |
|
@@ -50,7 +48,7 @@ Running the above command returns the following response
|
|
50 |
|
51 |
```json
|
52 |
{
|
53 |
-
"prompt": "[INST]
|
54 |
-
"response": "
|
55 |
}
|
56 |
```
|
|
|
1 |
# SliceX AI™ ELM (Efficient Language Models)
|
2 |
This repository contains code to run our ELM models.
|
3 |
|
4 |
+
Models are located in the "models" folder. ELM models in this repository comes in three sizes (elm-1.0, elm-0.75 and elm-0.25) and supports the following use-case.
|
|
|
|
|
5 |
- news_content_generation
|
6 |
|
7 |
## Download ELM repo
|
8 |
```bash
|
9 |
+
git clone [email protected]:slicexai/elm-v0.1_news_content_generation
|
10 |
sudo apt-get intall git-lfs
|
11 |
git lfs install
|
12 |
```
|
|
|
18 |
git lfs install
|
19 |
```
|
20 |
|
21 |
+
## Download ELM model checkpoints
|
22 |
```bash
|
23 |
+
cd elm-v0.1_news_content_generation
|
24 |
+
git lfs pull -I models/elm-1.0_news_content_generation/ckpt.pt
|
25 |
+
git lfs pull -I models/elm-0.75_news_content_generation/ckpt.pt
|
26 |
git lfs pull -I models/elm-0.25_news_content_generation/ckpt.pt
|
27 |
```
|
28 |
|
|
|
34 |
## How to use - Run ELM on a sample task (e.g., news classification)
|
35 |
```bash
|
36 |
python run.py <elm-model-directory>
|
37 |
+
E.g. python run.py models/elm-0.75_news_content_generation
|
38 |
```
|
39 |
+
Prompts for the specific tasks can be found in the corresponding checkpoint directory. See an example below in the form of `models/elm-0.75_news_content_generation/example_prompts.json`.
|
40 |
```json
|
41 |
{
|
42 |
+
"inputs": ["Scientists Invent 'Invisible' Metamaterial With Bonus Reflect Mode"],
|
43 |
+
"template": "[INST]The following headline is the headline of a news report. Please write the content of the news passage based on only this headline.\n\nHeadline: {input} \n\nContent:[/INST]"
|
44 |
}
|
45 |
```
|
46 |
|
|
|
48 |
|
49 |
```json
|
50 |
{
|
51 |
+
"prompt": "[INST]The following headline is the headline of a news report. Please write the content of the news passage based on only this headline.\n\nHeadline: Scientists Invent 'Invisible' Metamaterial With Bonus Reflect Mode \n\nContent:[/INST]",
|
52 |
+
"response": "A team of scientists have created an invisible material that can make objects disappear. It is made of a special material that creates a layer of nanoscale dots that allow light to enter from the material, directing it to a layer of gas that allows light to enter from the material. The material is able to levitate and roll off its surface without leaving the material. This technology could have many future applications in battery technology, microelectronics, and more. A video demonstrating the material is available in the article."
|
53 |
}
|
54 |
```
|