test_vision.py 12.3 KB
Newer Older
1
2
# SPDX-License-Identifier: Apache-2.0

3
4
import openai
import pytest
5
import os
6
import pytest_asyncio
pansicheng's avatar
pansicheng committed
7
8
9
import requests
from PIL import Image
from transformers import AutoProcessor
10

11
from vllm.multimodal.utils import encode_image_base64, fetch_image
12

13
from ...utils import RemoteOpenAIServer, models_path_prefix, urls_port
14

15
MODEL_NAME = os.path.join(models_path_prefix, "microsoft/Phi-3.5-vision-instruct")
16
MAXIMUM_IMAGES = 2
17

zhuwenwen's avatar
zhuwenwen committed
18
19


20
# Test different image extensions (JPG/PNG) and formats (gray/RGB/RGBA)
zhuwenwen's avatar
zhuwenwen committed
21
22
23
24
25
26
# TEST_IMAGE_URLS = [
#     "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg",
#     "https://upload.wikimedia.org/wikipedia/commons/f/fa/Grayscale_8bits_palette_sample_image.png",
#     "https://upload.wikimedia.org/wikipedia/commons/thumb/9/91/Venn_diagram_rgb.svg/1280px-Venn_diagram_rgb.svg.png",
#     "https://upload.wikimedia.org/wikipedia/commons/0/0b/RGBA_comp.png",
# ]
27
TEST_IMAGE_URLS = [
28
29
30
31
    f"http://localhost:{urls_port}/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg",
    f"http://localhost:{urls_port}/Grayscale_8bits_palette_sample_image.png",
    f"http://localhost:{urls_port}/Venn_diagram_rgb.svg/1280px-Venn_diagram_rgb.svg.png",
    f"http://localhost:{urls_port}/RGBA_comp.png",
32
33
34
]


35
@pytest.fixture(scope="module")
36
def server():
37
    args = [
38
39
        "--task",
        "generate",
40
41
42
43
44
45
46
47
        "--max-model-len",
        "2048",
        "--max-num-seqs",
        "5",
        "--enforce-eager",
        "--trust-remote-code",
        "--limit-mm-per-prompt",
        f"image={MAXIMUM_IMAGES}",
48
49
50
    ]

    with RemoteOpenAIServer(MODEL_NAME, args) as remote_server:
51
        yield remote_server
52
53


54
55
56
57
@pytest_asyncio.fixture
async def client(server):
    async with server.get_async_client() as async_client:
        yield async_client
58
59


60
@pytest.fixture(scope="session")
61
def base64_encoded_image() -> dict[str, str]:
62
    return {
63
        image_url: encode_image_base64(fetch_image(image_url))
64
65
66
67
        for image_url in TEST_IMAGE_URLS
    }


pansicheng's avatar
pansicheng committed
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
def get_hf_prompt_tokens(model_name, content, image_url):
    processor = AutoProcessor.from_pretrained(model_name,
                                              trust_remote_code=True,
                                              num_crops=4)

    placeholder = "<|image_1|>\n"
    messages = [{
        "role": "user",
        "content": f"{placeholder}{content}",
    }]
    images = [Image.open(requests.get(image_url, stream=True).raw)]

    prompt = processor.tokenizer.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True)
    inputs = processor(prompt, images, return_tensors="pt")

    return inputs.input_ids.shape[1]


87
88
89
@pytest.mark.asyncio
@pytest.mark.parametrize("model_name", [MODEL_NAME])
@pytest.mark.parametrize("image_url", TEST_IMAGE_URLS)
90
async def test_single_chat_session_image(client: openai.AsyncOpenAI,
91
                                         model_name: str, image_url: str):
pansicheng's avatar
pansicheng committed
92
    content_text = "What's in this image?"
93
94
95
96
97
98
99
100
101
102
103
104
    messages = [{
        "role":
        "user",
        "content": [
            {
                "type": "image_url",
                "image_url": {
                    "url": image_url
                }
            },
            {
                "type": "text",
pansicheng's avatar
pansicheng committed
105
                "text": content_text
106
107
108
109
            },
        ],
    }]

pansicheng's avatar
pansicheng committed
110
    max_completion_tokens = 10
111
    # test single completion
112
113
114
    chat_completion = await client.chat.completions.create(
        model=model_name,
        messages=messages,
pansicheng's avatar
pansicheng committed
115
        max_completion_tokens=max_completion_tokens,
116
        logprobs=True,
117
        temperature=0.0,
118
        top_logprobs=5)
119
120
121
122
    assert len(chat_completion.choices) == 1

    choice = chat_completion.choices[0]
    assert choice.finish_reason == "length"
pansicheng's avatar
pansicheng committed
123
124
    hf_prompt_tokens = get_hf_prompt_tokens(model_name, content_text,
                                            image_url)
125
    assert chat_completion.usage == openai.types.CompletionUsage(
pansicheng's avatar
pansicheng committed
126
127
128
        completion_tokens=max_completion_tokens,
        prompt_tokens=hf_prompt_tokens,
        total_tokens=hf_prompt_tokens + max_completion_tokens)
129
130
131
132
133
134
135
136
137
138
139
140

    message = choice.message
    message = chat_completion.choices[0].message
    assert message.content is not None and len(message.content) >= 10
    assert message.role == "assistant"
    messages.append({"role": "assistant", "content": message.content})

    # test multi-turn dialogue
    messages.append({"role": "user", "content": "express your result in json"})
    chat_completion = await client.chat.completions.create(
        model=model_name,
        messages=messages,
141
        max_completion_tokens=10,
142
143
144
145
146
    )
    message = chat_completion.choices[0].message
    assert message.content is not None and len(message.content) >= 0


147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
@pytest.mark.asyncio
@pytest.mark.parametrize("model_name", [MODEL_NAME])
@pytest.mark.parametrize("image_url", TEST_IMAGE_URLS)
async def test_single_chat_session_image_beamsearch(client: openai.AsyncOpenAI,
                                                    model_name: str,
                                                    image_url: str):
    messages = [{
        "role":
        "user",
        "content": [
            {
                "type": "image_url",
                "image_url": {
                    "url": image_url
                }
            },
            {
                "type": "text",
                "text": "What's in this image?"
            },
        ],
    }]

    chat_completion = await client.chat.completions.create(
        model=model_name,
        messages=messages,
        n=2,
174
        max_completion_tokens=10,
175
176
177
178
179
180
181
182
        logprobs=True,
        top_logprobs=5,
        extra_body=dict(use_beam_search=True))
    assert len(chat_completion.choices) == 2
    assert chat_completion.choices[
        0].message.content != chat_completion.choices[1].message.content


183
184
185
186
@pytest.mark.asyncio
@pytest.mark.parametrize("model_name", [MODEL_NAME])
@pytest.mark.parametrize("image_url", TEST_IMAGE_URLS)
async def test_single_chat_session_image_base64encoded(
187
        client: openai.AsyncOpenAI, model_name: str, image_url: str,
188
        base64_encoded_image: dict[str, str]):
189

pansicheng's avatar
pansicheng committed
190
    content_text = "What's in this image?"
191
192
193
194
195
196
197
198
199
200
201
202
203
    messages = [{
        "role":
        "user",
        "content": [
            {
                "type": "image_url",
                "image_url": {
                    "url":
                    f"data:image/jpeg;base64,{base64_encoded_image[image_url]}"
                }
            },
            {
                "type": "text",
pansicheng's avatar
pansicheng committed
204
                "text": content_text
205
206
207
208
            },
        ],
    }]

pansicheng's avatar
pansicheng committed
209
    max_completion_tokens = 10
210
    # test single completion
211
212
213
    chat_completion = await client.chat.completions.create(
        model=model_name,
        messages=messages,
pansicheng's avatar
pansicheng committed
214
        max_completion_tokens=max_completion_tokens,
215
        logprobs=True,
216
        temperature=0.0,
217
        top_logprobs=5)
218
219
220
221
    assert len(chat_completion.choices) == 1

    choice = chat_completion.choices[0]
    assert choice.finish_reason == "length"
pansicheng's avatar
pansicheng committed
222
223
    hf_prompt_tokens = get_hf_prompt_tokens(model_name, content_text,
                                            image_url)
224
    assert chat_completion.usage == openai.types.CompletionUsage(
pansicheng's avatar
pansicheng committed
225
226
227
        completion_tokens=max_completion_tokens,
        prompt_tokens=hf_prompt_tokens,
        total_tokens=hf_prompt_tokens + max_completion_tokens)
228
229
230
231
232
233
234
235
236
237
238
239

    message = choice.message
    message = chat_completion.choices[0].message
    assert message.content is not None and len(message.content) >= 10
    assert message.role == "assistant"
    messages.append({"role": "assistant", "content": message.content})

    # test multi-turn dialogue
    messages.append({"role": "user", "content": "express your result in json"})
    chat_completion = await client.chat.completions.create(
        model=model_name,
        messages=messages,
240
        max_completion_tokens=10,
241
        temperature=0.0,
242
243
244
245
246
    )
    message = chat_completion.choices[0].message
    assert message.content is not None and len(message.content) >= 0


247
248
249
250
251
@pytest.mark.asyncio
@pytest.mark.parametrize("model_name", [MODEL_NAME])
@pytest.mark.parametrize("image_url", TEST_IMAGE_URLS)
async def test_single_chat_session_image_base64encoded_beamsearch(
        client: openai.AsyncOpenAI, model_name: str, image_url: str,
252
        base64_encoded_image: dict[str, str]):
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274

    messages = [{
        "role":
        "user",
        "content": [
            {
                "type": "image_url",
                "image_url": {
                    "url":
                    f"data:image/jpeg;base64,{base64_encoded_image[image_url]}"
                }
            },
            {
                "type": "text",
                "text": "What's in this image?"
            },
        ],
    }]
    chat_completion = await client.chat.completions.create(
        model=model_name,
        messages=messages,
        n=2,
275
        max_completion_tokens=10,
276
277
278
279
280
281
        extra_body=dict(use_beam_search=True))
    assert len(chat_completion.choices) == 2
    assert chat_completion.choices[
        0].message.content != chat_completion.choices[1].message.content


282
283
284
@pytest.mark.asyncio
@pytest.mark.parametrize("model_name", [MODEL_NAME])
@pytest.mark.parametrize("image_url", TEST_IMAGE_URLS)
285
async def test_chat_streaming_image(client: openai.AsyncOpenAI,
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
                                    model_name: str, image_url: str):
    messages = [{
        "role":
        "user",
        "content": [
            {
                "type": "image_url",
                "image_url": {
                    "url": image_url
                }
            },
            {
                "type": "text",
                "text": "What's in this image?"
            },
        ],
    }]

    # test single completion
    chat_completion = await client.chat.completions.create(
        model=model_name,
        messages=messages,
308
        max_completion_tokens=10,
309
310
311
312
313
314
315
316
317
        temperature=0.0,
    )
    output = chat_completion.choices[0].message.content
    stop_reason = chat_completion.choices[0].finish_reason

    # test streaming
    stream = await client.chat.completions.create(
        model=model_name,
        messages=messages,
318
        max_completion_tokens=10,
319
320
321
        temperature=0.0,
        stream=True,
    )
322
    chunks: list[str] = []
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
    finish_reason_count = 0
    async for chunk in stream:
        delta = chunk.choices[0].delta
        if delta.role:
            assert delta.role == "assistant"
        if delta.content:
            chunks.append(delta.content)
        if chunk.choices[0].finish_reason is not None:
            finish_reason_count += 1
    # finish reason should only return in last block
    assert finish_reason_count == 1
    assert chunk.choices[0].finish_reason == stop_reason
    assert delta.content
    assert "".join(chunks) == output


@pytest.mark.asyncio
@pytest.mark.parametrize("model_name", [MODEL_NAME])
341
342
343
@pytest.mark.parametrize(
    "image_urls",
    [TEST_IMAGE_URLS[:i] for i in range(2, len(TEST_IMAGE_URLS))])
344
async def test_multi_image_input(client: openai.AsyncOpenAI, model_name: str,
345
                                 image_urls: list[str]):
346
347
348
349
350

    messages = [{
        "role":
        "user",
        "content": [
351
            *({
352
353
354
355
                "type": "image_url",
                "image_url": {
                    "url": image_url
                }
356
            } for image_url in image_urls),
357
358
359
360
361
362
363
            {
                "type": "text",
                "text": "What's in this image?"
            },
        ],
    }]

364
365
366
367
368
    if len(image_urls) > MAXIMUM_IMAGES:
        with pytest.raises(openai.BadRequestError):  # test multi-image input
            await client.chat.completions.create(
                model=model_name,
                messages=messages,
369
                max_completion_tokens=10,
370
371
372
373
374
375
376
377
378
379
380
381
382
383
                temperature=0.0,
            )

        # the server should still work afterwards
        completion = await client.completions.create(
            model=model_name,
            prompt=[0, 0, 0, 0, 0],
            max_tokens=5,
            temperature=0.0,
        )
        completion = completion.choices[0].text
        assert completion is not None and len(completion) >= 0
    else:
        chat_completion = await client.chat.completions.create(
384
385
            model=model_name,
            messages=messages,
386
            max_completion_tokens=10,
387
388
            temperature=0.0,
        )
389
390
        message = chat_completion.choices[0].message
        assert message.content is not None and len(message.content) >= 0