Commit
•
8b56eb5
1
Parent(s):
0ddead0
Update README.md
Browse files
README.md
CHANGED
@@ -3,55 +3,61 @@ tags:
|
|
3 |
- generated_from_trainer
|
4 |
base_model: llava-hf/llava-1.5-7b-hf
|
5 |
model-index:
|
6 |
-
- name:
|
7 |
results: []
|
8 |
---
|
9 |
|
10 |
-
|
11 |
-
should probably proofread and complete it, then remove this comment. -->
|
12 |
|
13 |
-
|
14 |
|
15 |
-
|
16 |
|
17 |
-
|
18 |
|
19 |
-
|
20 |
|
21 |
-
##
|
22 |
|
23 |
-
|
|
|
|
|
|
|
|
|
24 |
|
25 |
-
|
26 |
|
27 |
-
More information needed
|
28 |
|
29 |
-
|
|
|
|
|
|
|
30 |
|
31 |
-
### Training hyperparameters
|
32 |
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
- eval_batch_size: 1
|
37 |
-
- seed: 42
|
38 |
-
- distributed_type: multi-GPU
|
39 |
-
- num_devices: 8
|
40 |
-
- gradient_accumulation_steps: 16
|
41 |
-
- total_train_batch_size: 128
|
42 |
-
- total_eval_batch_size: 8
|
43 |
-
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
44 |
-
- lr_scheduler_type: cosine
|
45 |
-
- lr_scheduler_warmup_ratio: 0.03
|
46 |
-
- num_epochs: 1.0
|
47 |
|
48 |
-
|
|
|
|
|
|
|
49 |
|
|
|
|
|
|
|
50 |
|
|
|
|
|
|
|
51 |
|
52 |
-
|
|
|
53 |
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
|
|
|
|
|
|
|
3 |
- generated_from_trainer
|
4 |
base_model: llava-hf/llava-1.5-7b-hf
|
5 |
model-index:
|
6 |
+
- name: Mantis-llava-7b
|
7 |
results: []
|
8 |
---
|
9 |
|
10 |
+
# Mantis: Interleaved Multi-Image Instruction Tuning
|
|
|
11 |
|
12 |
+
**Mantis** is a multimodal conversational AI model that can chat with users about images and text. It's optimized for multi-image reasoning, where interleaved text and images can be used to generate responses.
|
13 |
|
14 |
+
Mantis is trained on the newly curated dataset **Mantis-Instruct**, a large-scale multi-image QA dataset that covers various multi-image reasoning tasks.
|
15 |
|
16 |
+
|[Demo](https://huggingface.co/spaces/TIGER-Lab/Mantis) | [Blog](https://tiger-ai-lab.github.io/Blog/mantis) | [Github](https://github.com/TIGER-AI-Lab/Mantis) | [Models](https://huggingface.co/collections/TIGER-Lab/mantis-6619b0834594c878cdb1d6e4) |
|
17 |
|
18 |
+
![Mantis](./overall_barchart.jpeg)
|
19 |
|
20 |
+
## Inference
|
21 |
|
22 |
+
You can install Mantis's GitHub codes as a Python package
|
23 |
+
```bash
|
24 |
+
pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
|
25 |
+
```
|
26 |
+
then run inference with codes here: [examples/run_mantis.py](https://github.com/TIGER-AI-Lab/Mantis/blob/main/examples/run_mantis_hf.py)
|
27 |
|
28 |
+
Or, you can run the model without relying on the mantis codes, using pure hugging face transformers. See [examples/run_mantis_hf.py](https://github.com/TIGER-AI-Lab/Mantis/blob/main/examples/run_mantis_hf.py) for details.
|
29 |
|
|
|
30 |
|
31 |
+
```python
|
32 |
+
from mantis.models.mllava import chat_mllava
|
33 |
+
from PIL import Image
|
34 |
+
import torch
|
35 |
|
|
|
36 |
|
37 |
+
image1 = "image1.jpg"
|
38 |
+
image2 = "image2.jpg"
|
39 |
+
images = [Image.open(image1), Image.open(image2)]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
|
41 |
+
# load processor and model
|
42 |
+
from mantis.models.mllava import MLlavaProcessor, LlavaForConditionalGeneration
|
43 |
+
processor = MLlavaProcessor.from_pretrained("TIGER-Lab/Mantis-bakllava-7b")
|
44 |
+
model = LlavaForConditionalGeneration.from_pretrained("TIGER-Lab/Mantis-bakllava-7b", device_map="auto", torch_dtype=torch.bfloat16, attn_implementation="flash_attention_2")
|
45 |
|
46 |
+
# chat
|
47 |
+
text = "<image> <image> What's the difference between these two images? Please describe as much as you can."
|
48 |
+
response, history = chat_mllava(text, images, model, processor)
|
49 |
|
50 |
+
print("USER: ", text)
|
51 |
+
print("ASSISTANT: ", response)
|
52 |
+
# The image on the right has a larger number of wallets displayed compared to the image on the left. The wallets in the right image are arranged in a grid pattern, while the wallets in the left image are displayed in a more scattered manner. The wallets in the right image have various colors, including red, purple, and brown, while the wallets in the left image are primarily brown.
|
53 |
|
54 |
+
text = "How many items are there in image 1 and image 2 respectively?"
|
55 |
+
response, history = chat_mllava(text, images, model, processor, history=history)
|
56 |
|
57 |
+
print("USER: ", text)
|
58 |
+
print("ASSISTANT: ", response)
|
59 |
+
# There are two items in image 1 and four items in image 2.
|
60 |
+
```
|
61 |
+
|
62 |
+
## Training
|
63 |
+
Training codes will be released soon.
|