File size: 4,462 Bytes
15245b5
686ef17
 
 
 
 
 
 
15d82cf
15245b5
 
 
15d82cf
 
15245b5
 
15d82cf
 
15245b5
 
686ef17
121a196
0318ee3
686ef17
15245b5
686ef17
 
15245b5
 
686ef17
15245b5
 
 
 
 
 
 
 
 
 
 
686ef17
 
 
15245b5
 
 
 
 
 
 
 
 
686ef17
15245b5
686ef17
15245b5
686ef17
15245b5
686ef17
 
15245b5
 
 
 
 
 
 
686ef17
15245b5
 
 
 
686ef17
15245b5
 
 
 
 
 
 
 
 
 
686ef17
15245b5
 
 
 
 
686ef17
15245b5
 
 
 
 
 
 
 
 
 
 
 
 
 
686ef17
15245b5
686ef17
 
15245b5
686ef17
 
 
 
15245b5
 
 
 
 
 
 
 
 
 
686ef17
15245b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
686ef17
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
from transformers import AutoTokenizer, TextStreamer
from PIL import Image
import torch
from threading import Thread
import gradio as gr
from gradio import FileData
import time
import spaces
from unsloth import FastVisionModel

# Load model and tokenizer
ckpt = "Daemontatox/DocumentLlama"
model, tokenizer = FastVisionModel.from_pretrained(
    ckpt,
    load_in_4bit=True,
    use_gradient_checkpointing="unsloth",
)

# Enable inference mode
FastVisionModel.for_inference(model)

@spaces.GPU()
def bot_streaming(message, history, max_new_tokens=2048):
    txt = message["text"]
    messages = []
    images = []
    
    # Process history
    for i, msg in enumerate(history):
        if isinstance(msg[0], tuple):
            messages.append({
                "role": "user",
                "content": [
                    {"type": "text", "text": history[i+1][0]},
                    {"type": "image"}
                ]
            })
            messages.append({
                "role": "assistant",
                "content": [{"type": "text", "text": history[i+1][1]}]
            })
            images.append(Image.open(msg[0][0]).convert("RGB"))
        elif isinstance(history[i-1], tuple) and isinstance(msg[0], str):
            pass
        elif isinstance(history[i-1][0], str) and isinstance(msg[0], str):
            messages.append({
                "role": "user",
                "content": [{"type": "text", "text": msg[0]}]
            })
            messages.append({
                "role": "assistant",
                "content": [{"type": "text", "text": msg[1]}]
            })

    # Handle current message
    if len(message["files"]) == 1:
        if isinstance(message["files"][0], str):  # examples
            image = Image.open(message["files"][0]).convert("RGB")
        else:  # regular input
            image = Image.open(message["files"][0]["path"]).convert("RGB")
        images.append(image)
        messages.append({
            "role": "user",
            "content": [
                {"type": "image"},
                {"type": "text", "text": txt}
            ]
        })
    else:
        messages.append({
            "role": "user",
            "content": [{"type": "text", "text": txt}]
        })

    # Prepare inputs
    input_text = tokenizer.apply_chat_template(messages, add_generation_prompt=True)
    
    if images:
        inputs = tokenizer(
            images[-1],  # Use the last image
            input_text,
            add_special_tokens=False,
            return_tensors="pt"
        ).to("cuda")
    else:
        inputs = tokenizer(
            input_text,
            add_special_tokens=False,
            return_tensors="pt"
        ).to("cuda")

    # Setup streaming
    text_streamer = TextStreamer(tokenizer, skip_prompt=True)
    buffer = ""

    def generate():
        nonlocal buffer
        output_ids = model.generate(
            **inputs,
            streamer=text_streamer,
            max_new_tokens=max_new_tokens,
            use_cache=True,
            temperature=1.5,
            min_p=0.1
        )
    
    thread = Thread(target=generate)
    thread.start()
    
    for new_text in text_streamer:
        buffer += new_text
        time.sleep(0.01)
        yield buffer

# Setup Gradio interface
demo = gr.ChatInterface(
    fn=bot_streaming,
    title="Document Analyzer",
    examples=[
        [{"text": "Which era does this piece belong to? Give details about the era.", "files":["./examples/rococo.jpg"]}, 200],
        [{"text": "Where do the droughts happen according to this diagram?", "files":["./examples/weather_events.png"]}, 250],
        [{"text": "What happens when you take out white cat from this chain?", "files":["./examples/ai2d_test.jpg"]}, 250],
        [{"text": "How long does it take from invoice date to due date? Be short and concise.", "files":["./examples/invoice.png"]}, 250],
        [{"text": "Where to find this monument? Can you give me other recommendations around the area?", "files":["./examples/wat_arun.jpg"]}, 250],
    ],
    textbox=gr.MultimodalTextbox(),
    additional_inputs=[
        gr.Slider(
            minimum=10,
            maximum=500,
            value=2048,
            step=10,
            label="Maximum number of new tokens to generate",
        )
    ],
    cache_examples=False,
    description="MllM",
    stop_btn="Stop Generation",
    fill_height=True,
    multimodal=True
)

demo.launch(debug=True)