Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
import streamlit as st
|
2 |
import torch
|
3 |
-
from transformers import
|
4 |
from PIL import Image
|
5 |
from byaldi import RAGMultiModalModel
|
6 |
from qwen_vl_utils import process_vision_info
|
@@ -8,24 +8,24 @@ from qwen_vl_utils import process_vision_info
|
|
8 |
# Model and processor names
|
9 |
RAG_MODEL = "vidore/colpali"
|
10 |
QWN_MODEL = "Qwen/Qwen2-VL-7B-Instruct"
|
11 |
-
QWN_PROCESSOR = "Qwen/Qwen2-VL-2B-Instruct"
|
12 |
|
13 |
@st.cache_resource
|
14 |
def load_models():
|
15 |
RAG = RAGMultiModalModel.from_pretrained(RAG_MODEL)
|
16 |
|
17 |
-
model =
|
18 |
QWN_MODEL,
|
19 |
torch_dtype=torch.bfloat16,
|
|
|
|
|
20 |
trust_remote_code=True
|
21 |
-
).
|
22 |
|
23 |
-
processor = AutoProcessor.from_pretrained(
|
24 |
-
tokenizer = AutoTokenizer.from_pretrained(QWN_PROCESSOR, trust_remote_code=True)
|
25 |
|
26 |
-
return RAG, model, processor
|
27 |
|
28 |
-
RAG, model, processor
|
29 |
|
30 |
def document_rag(text_query, image):
|
31 |
messages = [
|
@@ -40,7 +40,7 @@ def document_rag(text_query, image):
|
|
40 |
],
|
41 |
}
|
42 |
]
|
43 |
-
text =
|
44 |
messages, tokenize=False, add_generation_prompt=True
|
45 |
)
|
46 |
image_inputs, video_inputs = process_vision_info(messages)
|
@@ -51,12 +51,12 @@ def document_rag(text_query, image):
|
|
51 |
padding=True,
|
52 |
return_tensors="pt",
|
53 |
)
|
54 |
-
inputs = inputs.to(
|
55 |
generated_ids = model.generate(**inputs, max_new_tokens=50)
|
56 |
generated_ids_trimmed = [
|
57 |
out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
|
58 |
]
|
59 |
-
output_text =
|
60 |
generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
|
61 |
)
|
62 |
return output_text[0]
|
|
|
1 |
import streamlit as st
|
2 |
import torch
|
3 |
+
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
|
4 |
from PIL import Image
|
5 |
from byaldi import RAGMultiModalModel
|
6 |
from qwen_vl_utils import process_vision_info
|
|
|
8 |
# Model and processor names
|
9 |
RAG_MODEL = "vidore/colpali"
|
10 |
QWN_MODEL = "Qwen/Qwen2-VL-7B-Instruct"
|
|
|
11 |
|
12 |
@st.cache_resource
|
13 |
def load_models():
|
14 |
RAG = RAGMultiModalModel.from_pretrained(RAG_MODEL)
|
15 |
|
16 |
+
model = Qwen2VLForConditionalGeneration.from_pretrained(
|
17 |
QWN_MODEL,
|
18 |
torch_dtype=torch.bfloat16,
|
19 |
+
attn_implementation="flash_attention_2",
|
20 |
+
device_map="auto",
|
21 |
trust_remote_code=True
|
22 |
+
).eval()
|
23 |
|
24 |
+
processor = AutoProcessor.from_pretrained(QWN_MODEL, trust_remote_code=True)
|
|
|
25 |
|
26 |
+
return RAG, model, processor
|
27 |
|
28 |
+
RAG, model, processor = load_models()
|
29 |
|
30 |
def document_rag(text_query, image):
|
31 |
messages = [
|
|
|
40 |
],
|
41 |
}
|
42 |
]
|
43 |
+
text = processor.apply_chat_template(
|
44 |
messages, tokenize=False, add_generation_prompt=True
|
45 |
)
|
46 |
image_inputs, video_inputs = process_vision_info(messages)
|
|
|
51 |
padding=True,
|
52 |
return_tensors="pt",
|
53 |
)
|
54 |
+
inputs = inputs.to(model.device)
|
55 |
generated_ids = model.generate(**inputs, max_new_tokens=50)
|
56 |
generated_ids_trimmed = [
|
57 |
out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
|
58 |
]
|
59 |
+
output_text = processor.batch_decode(
|
60 |
generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
|
61 |
)
|
62 |
return output_text[0]
|