Update README.md
Browse files
README.md
CHANGED
@@ -1,78 +1,78 @@
|
|
1 |
-
---
|
2 |
-
inference: false
|
3 |
-
license:
|
4 |
-
---
|
5 |
-
|
6 |
-
# Model Card
|
7 |
-
|
8 |
-
<p align="center">
|
9 |
-
<img src="./icon.png" alt="Logo" width="350">
|
10 |
-
</p>
|
11 |
-
|
12 |
-
π [Technical report]() | π [Code]() | π° [Demo](http://bunny.dataoptim.org/)
|
13 |
-
This is Owlet-Phi-2.
|
14 |
-
|
15 |
-
Owlet is a family of lightweight but powerful multimodal models.
|
16 |
-
|
17 |
-
We provide Owlet-phi-2, which is built upon [SigLIP](https://huggingface.co/google/siglip-so400m-patch14-384) and [Phi-2](https://huggingface.co/microsoft/phi-2).
|
18 |
-
|
19 |
-
|
20 |
-
# Quickstart
|
21 |
-
|
22 |
-
Here we show a code snippet to show you how to use the model with transformers.
|
23 |
-
|
24 |
-
Before running the snippet, you need to install the following dependencies:
|
25 |
-
|
26 |
-
```shell
|
27 |
-
pip install torch transformers accelerate pillow decord
|
28 |
-
```
|
29 |
-
|
30 |
-
```python
|
31 |
-
import torch
|
32 |
-
import transformers
|
33 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer
|
34 |
-
from PIL import Image
|
35 |
-
import warnings
|
36 |
-
|
37 |
-
|
38 |
-
# disable some warnings
|
39 |
-
transformers.logging.set_verbosity_error()
|
40 |
-
transformers.logging.disable_progress_bar()
|
41 |
-
warnings.filterwarnings('ignore')
|
42 |
-
|
43 |
-
# set device
|
44 |
-
device = 'cuda' # or cpu
|
45 |
-
torch.set_default_device(device)
|
46 |
-
|
47 |
-
# create model
|
48 |
-
print('Loading the model...')
|
49 |
-
model = AutoModelForCausalLM.from_pretrained(
|
50 |
-
'phronetic-ai/owlet-phi-2',
|
51 |
-
torch_dtype=torch.float16, # float32 for cpu
|
52 |
-
device_map='auto',
|
53 |
-
trust_remote_code=True)
|
54 |
-
tokenizer = AutoTokenizer.from_pretrained(
|
55 |
-
'phronetic-ai/owlet-phi-2',
|
56 |
-
trust_remote_code=True)
|
57 |
-
|
58 |
-
print('Model loaded. Processing the query...')
|
59 |
-
# text prompt
|
60 |
-
prompt = 'What is happening in the video?'
|
61 |
-
text = f"A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: <image>\n{prompt} ASSISTANT:"
|
62 |
-
text_chunks = [tokenizer(chunk).input_ids for chunk in text.split('<image>')]
|
63 |
-
input_ids = torch.tensor(text_chunks[0] + [-200] + text_chunks[1], dtype=torch.long).unsqueeze(0).to(device)
|
64 |
-
|
65 |
-
# image or video file path
|
66 |
-
file_path = 'sample.mp4'
|
67 |
-
input_tensor = model.process(file_path, model.config).to(model.device, dtype=model.dtype)
|
68 |
-
|
69 |
-
# generate
|
70 |
-
output_ids = model.generate(
|
71 |
-
input_ids,
|
72 |
-
images=input_tensor,
|
73 |
-
max_new_tokens=100,
|
74 |
-
use_cache=True)[0]
|
75 |
-
|
76 |
-
print(f'Response: {tokenizer.decode(output_ids[input_ids.shape[1]:], skip_special_tokens=True).strip()}')
|
77 |
-
|
78 |
-
```
|
|
|
1 |
+
---
|
2 |
+
inference: false
|
3 |
+
license: cc-by-4.0
|
4 |
+
---
|
5 |
+
|
6 |
+
# Model Card
|
7 |
+
|
8 |
+
<p align="center">
|
9 |
+
<img src="./icon.png" alt="Logo" width="350">
|
10 |
+
</p>
|
11 |
+
|
12 |
+
π [Technical report]() | π [Code]() | π° [Demo](http://bunny.dataoptim.org/)
|
13 |
+
This is Owlet-Phi-2.
|
14 |
+
|
15 |
+
Owlet is a family of lightweight but powerful multimodal models.
|
16 |
+
|
17 |
+
We provide Owlet-phi-2, which is built upon [SigLIP](https://huggingface.co/google/siglip-so400m-patch14-384) and [Phi-2](https://huggingface.co/microsoft/phi-2).
|
18 |
+
|
19 |
+
|
20 |
+
# Quickstart
|
21 |
+
|
22 |
+
Here we show a code snippet to show you how to use the model with transformers.
|
23 |
+
|
24 |
+
Before running the snippet, you need to install the following dependencies:
|
25 |
+
|
26 |
+
```shell
|
27 |
+
pip install torch transformers accelerate pillow decord
|
28 |
+
```
|
29 |
+
|
30 |
+
```python
|
31 |
+
import torch
|
32 |
+
import transformers
|
33 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
34 |
+
from PIL import Image
|
35 |
+
import warnings
|
36 |
+
|
37 |
+
|
38 |
+
# disable some warnings
|
39 |
+
transformers.logging.set_verbosity_error()
|
40 |
+
transformers.logging.disable_progress_bar()
|
41 |
+
warnings.filterwarnings('ignore')
|
42 |
+
|
43 |
+
# set device
|
44 |
+
device = 'cuda' # or cpu
|
45 |
+
torch.set_default_device(device)
|
46 |
+
|
47 |
+
# create model
|
48 |
+
print('Loading the model...')
|
49 |
+
model = AutoModelForCausalLM.from_pretrained(
|
50 |
+
'phronetic-ai/owlet-phi-2',
|
51 |
+
torch_dtype=torch.float16, # float32 for cpu
|
52 |
+
device_map='auto',
|
53 |
+
trust_remote_code=True)
|
54 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
55 |
+
'phronetic-ai/owlet-phi-2',
|
56 |
+
trust_remote_code=True)
|
57 |
+
|
58 |
+
print('Model loaded. Processing the query...')
|
59 |
+
# text prompt
|
60 |
+
prompt = 'What is happening in the video?'
|
61 |
+
text = f"A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: <image>\n{prompt} ASSISTANT:"
|
62 |
+
text_chunks = [tokenizer(chunk).input_ids for chunk in text.split('<image>')]
|
63 |
+
input_ids = torch.tensor(text_chunks[0] + [-200] + text_chunks[1], dtype=torch.long).unsqueeze(0).to(device)
|
64 |
+
|
65 |
+
# image or video file path
|
66 |
+
file_path = 'sample.mp4'
|
67 |
+
input_tensor = model.process(file_path, model.config).to(model.device, dtype=model.dtype)
|
68 |
+
|
69 |
+
# generate
|
70 |
+
output_ids = model.generate(
|
71 |
+
input_ids,
|
72 |
+
images=input_tensor,
|
73 |
+
max_new_tokens=100,
|
74 |
+
use_cache=True)[0]
|
75 |
+
|
76 |
+
print(f'Response: {tokenizer.decode(output_ids[input_ids.shape[1]:], skip_special_tokens=True).strip()}')
|
77 |
+
|
78 |
+
```
|