Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,35 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
---
|
| 4 |
+
|
| 5 |
+
```
|
| 6 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 7 |
+
from PIL import Image
|
| 8 |
+
import requests
|
| 9 |
+
import torch
|
| 10 |
+
|
| 11 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 12 |
+
"anananan116/TinyVLM",
|
| 13 |
+
trust_remote_code = True,
|
| 14 |
+
torch_dtype=torch.float16,
|
| 15 |
+
).to('cuda').eval()
|
| 16 |
+
tokenizer = AutoTokenizer.from_pretrained("anananan116/TinyVLM")
|
| 17 |
+
|
| 18 |
+
# `<IMGPLH>` is the image placeholder which will be replaced by image embeddings.
|
| 19 |
+
# the number of `<IMGPLH>` should be equal to the number of input images
|
| 20 |
+
|
| 21 |
+
prompt = "Here's an image:<IMGPLH>Describe this image."
|
| 22 |
+
image = Image.open(requests.get('https://github.com/anananan116/TinyVLM/blob/main/test.png?raw=true',stream=True).raw)
|
| 23 |
+
inputs = model.prepare_input_ids_for_generation([prompt], [image], tokenizer)
|
| 24 |
+
|
| 25 |
+
with torch.no_grad():
|
| 26 |
+
outputs = model.generate(
|
| 27 |
+
input_ids=inputs['input_ids'].to("cuda"),
|
| 28 |
+
attention_mask=inputs['attention_mask'].to("cuda"),
|
| 29 |
+
encoded_image = inputs["encoded_image"],
|
| 30 |
+
max_new_tokens=128,
|
| 31 |
+
do_sample=True
|
| 32 |
+
)
|
| 33 |
+
|
| 34 |
+
output_text = tokenizer.batch_decode(outputs, skip_special_tokens=True)
|
| 35 |
+
```
|