Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,63 @@
|
|
1 |
-
---
|
2 |
-
license: apache-2.0
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
library_name: transformers
|
4 |
+
---
|
5 |
+
|
6 |
+
|
7 |
+
#### Quickstart
|
8 |
+
|
9 |
+
```python
|
10 |
+
from PIL import Image
|
11 |
+
from transformers import AutoTokenizer, AutoModel, AutoImageProcessor, AutoModelForCausalLM
|
12 |
+
from transformers.generation.configuration_utils import GenerationConfig
|
13 |
+
import torch
|
14 |
+
|
15 |
+
import sys
|
16 |
+
sys.path.append(PATH_TO_BAAI_Emu3-Chat_MODEL)
|
17 |
+
from processing_emu3 import Emu3Processor
|
18 |
+
|
19 |
+
|
20 |
+
# model path
|
21 |
+
EMU_HUB = "BAAI/Emu3-Chat"
|
22 |
+
VQ_HUB = "BAAI/Emu3-VisionTokenizer"
|
23 |
+
|
24 |
+
# prepare model and processor
|
25 |
+
model = AutoModelForCausalLM.from_pretrained(
|
26 |
+
EMU_HUB,
|
27 |
+
device_map="cuda:0",
|
28 |
+
torch_dtype=torch.bfloat16,
|
29 |
+
attn_implementation="flash_attention_2",
|
30 |
+
trust_remote_code=True,
|
31 |
+
)
|
32 |
+
|
33 |
+
tokenizer = AutoTokenizer.from_pretrained(EMU_HUB, trust_remote_code=True)
|
34 |
+
image_processor = AutoImageProcessor.from_pretrained(VQ_HUB, trust_remote_code=True)
|
35 |
+
image_tokenizer = AutoModel.from_pretrained(VQ_HUB, device_map="cuda:0", trust_remote_code=True).eval()
|
36 |
+
processor = Emu3Processor(image_processor, image_tokenizer, tokenizer)
|
37 |
+
|
38 |
+
# prepare input
|
39 |
+
text = "Please describe the image"
|
40 |
+
image = Image.open("assets/demo.png")
|
41 |
+
|
42 |
+
inputs = processor(
|
43 |
+
text=text,
|
44 |
+
image=image,
|
45 |
+
mode='U',
|
46 |
+
padding_side="left",
|
47 |
+
padding="longest",
|
48 |
+
return_tensors="pt",
|
49 |
+
)
|
50 |
+
|
51 |
+
# prepare hyper parameters
|
52 |
+
GENERATION_CONFIG = GenerationConfig(pad_token_id=tokenizer.pad_token_id, bos_token_id=tokenizer.bos_token_id, eos_token_id=tokenizer.eos_token_id)
|
53 |
+
|
54 |
+
# generate
|
55 |
+
outputs = model.generate(
|
56 |
+
inputs.input_ids.to("cuda:0"),
|
57 |
+
GENERATION_CONFIG,
|
58 |
+
max_new_tokens=320,
|
59 |
+
)
|
60 |
+
|
61 |
+
outputs = outputs[:, inputs.input_ids.shape[-1]:]
|
62 |
+
print(processor.batch_decode(outputs, skip_special_tokens=True)[0])
|
63 |
+
```
|