gokaygokay
commited on
Commit
•
770ad2a
1
Parent(s):
3ec40c7
Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,51 @@
|
|
1 |
-
---
|
2 |
-
license: apache-2.0
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
datasets:
|
4 |
+
- google/docci
|
5 |
+
- gokaygokay/random_instruct_docci
|
6 |
+
language:
|
7 |
+
- en
|
8 |
+
pipeline_tag: image-text-to-text
|
9 |
+
---
|
10 |
+
|
11 |
+
Fine tuned version of [moondream2](https://huggingface.co/vikhyatk/moondream2) model using [gokaygokay/random_instruct_docci](https://huggingface.co/datasets/gokaygokay/random_instruct_docci) dataset. Which gives extremely detailed captions of the images.
|
12 |
+
|
13 |
+
```
|
14 |
+
pip install transformers timm einops bitsandbytes accelerate flash-attn
|
15 |
+
```
|
16 |
+
|
17 |
+
```python
|
18 |
+
import torch
|
19 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
20 |
+
from PIL import Image
|
21 |
+
|
22 |
+
DEVICE = "cuda"
|
23 |
+
DTYPE = (
|
24 |
+
torch.float32 if DEVICE == "cpu" else torch.float16
|
25 |
+
) # CPU doesn't support float16
|
26 |
+
revision = "3ec40c7b6b5d87bc0c51edee45e21f5f29b449d8"
|
27 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
28 |
+
"gokaygokay/moondream2-docci-with-instruction",
|
29 |
+
trust_remote_code=True,
|
30 |
+
revision=revision
|
31 |
+
)
|
32 |
+
moondream = AutoModelForCausalLM.from_pretrained(
|
33 |
+
"gokaygokay/moondream2-docci-with-instruction",
|
34 |
+
trust_remote_code=True,
|
35 |
+
torch_dtype=DTYPE,
|
36 |
+
device_map={"": DEVICE},
|
37 |
+
attn_implementation="flash_attention_2",
|
38 |
+
revision=revision
|
39 |
+
)
|
40 |
+
moondream.eval()
|
41 |
+
|
42 |
+
image_path = "<your_image_path>"
|
43 |
+
image = Image.open(image_path).convert("RGB")
|
44 |
+
md_answer = moondream.answer_question(
|
45 |
+
moondream.encode_image(image),
|
46 |
+
"what is this picture about",
|
47 |
+
tokenizer=tokenizer,
|
48 |
+
)
|
49 |
+
|
50 |
+
print(md_answer)
|
51 |
+
```
|