File size: 3,747 Bytes
357c750
 
 
 
 
be75a6c
 
 
 
 
841b7c4
357c750
be75a6c
 
 
357c750
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
be75a6c
 
 
 
 
 
 
 
 
 
 
357c750
be75a6c
357c750
 
 
 
 
 
 
 
 
 
 
 
 
 
be75a6c
357c750
 
 
 
be75a6c
357c750
 
 
 
 
be75a6c
357c750
 
 
be75a6c
 
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
import requests
import torch
from PIL import Image
from io import BytesIO

from transformers import (
    AutoProcessor,
    AutoModelForVision2Seq,
    AwqConfig,
)
import awq_ext

import time

MODE = "fused_quantized"
DEVICE = "cuda"
PROCESSOR = AutoProcessor.from_pretrained("HuggingFaceM4/idefics2-tfrm-compatible")
BAD_WORDS_IDS = PROCESSOR.tokenizer(
    ["<image>", "<fake_token_around_image>"], add_special_tokens=False
).input_ids
EOS_WORDS_IDS = PROCESSOR.tokenizer(
    "<end_of_utterance>", add_special_tokens=False
).input_ids + [PROCESSOR.tokenizer.eos_token_id]

# Load model
if MODE == "regular":
    model = AutoModelForVision2Seq.from_pretrained(
        "HuggingFaceM4/idefics2-tfrm-compatible",
        torch_dtype=torch.float16,
        trust_remote_code=True,
        _attn_implementation="flash_attention_2",
        revision="3dc93be345d64fb6b1c550a233fe87ddb36f183d",
    ).to(DEVICE)
elif MODE == "quantized":
    quant_path = "HuggingFaceM4/idefics2-tfrm-compatible-AWQ"
    model = AutoModelForVision2Seq.from_pretrained(
        quant_path, trust_remote_code=True
    ).to(DEVICE)
elif MODE == "fused_quantized":
    quant_path = "HuggingFaceM4/idefics2-tfrm-compatible-AWQ"
    quantization_config = AwqConfig(
        bits=4,
        fuse_max_seq_len=4096,
        modules_to_fuse={
            "attention": ["q_proj", "k_proj", "v_proj", "o_proj"],
            "mlp": ["gate_proj", "up_proj", "down_proj"],
            "layernorm": ["input_layernorm", "post_attention_layernorm", "norm"],
            "use_alibi": False,
            "num_attention_heads": 32,
            "num_key_value_heads": 8,
            "hidden_size": 4096,
        },
    )
    model = AutoModelForVision2Seq.from_pretrained(
        quant_path,
        quantization_config=quantization_config,
        trust_remote_code=True,
    ).to(DEVICE)
else:
    raise ValueError("Unknown mode")


def download_image(url):
    try:
        # Send a GET request to the URL to download the image
        response = requests.get(url)
        # Check if the request was successful (status code 200)
        if response.status_code == 200:
            # Open the image using PIL
            image = Image.open(BytesIO(response.content))
            # Return the PIL image object
            return image
        else:
            print(f"Failed to download image. Status code: {response.status_code}")
            return None
    except Exception as e:
        print(f"An error occurred: {e}")
        return None


# Create inputs
image1 = download_image(
    "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
)


def reset_awq_cache(model):
    """
    Simple method to reset the AWQ fused modules cache
    """
    from awq.modules.fused.attn import QuantAttentionFused

    for name, module in model.named_modules():
        if isinstance(module, QuantAttentionFused):
            module.start_pos = 0


def ask_vlm(image, instruction):
    global model
    prompts = [
        "User:",
        image,
        f"{instruction}.<end_of_utterance>\n",
        "Assistant:",
    ]
    inputs = PROCESSOR(prompts)
    inputs = {k: torch.tensor(v).to(DEVICE) for k, v in inputs.items()}
    generated_ids = model.generate(
        **inputs,
        bad_words_ids=BAD_WORDS_IDS,
        max_new_tokens=100,
    )
    generated_texts = PROCESSOR.batch_decode(generated_ids, skip_special_tokens=True)
    reset_awq_cache(model)
    return generated_texts


now = time.time()

print(ask_vlm(image1, "What is this?")[0].split("\nAssistant: ")[1])

print("resp:", time.time() - now)
import time


now = time.time()

print(ask_vlm(image1, "What is this?")[0].split("\nAssistant: ")[1])

print("resp:", time.time() - now)