\n
docker-compose.yml\u3067Docker\u30b3\u30f3\u30c6\u30ca\u306e\u8a2d\u5b9a\u3092\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
docker-compose.yml\u306eYAML\u30d5\u30a1\u30a4\u30eb\u3092\u4f5c\u6210\u3057\u3066\u958b\u304d\u307e\u3059\u3002<\/p>\n\n\n\n
nano docker-compose.yml<\/code><\/pre>\n\n\n\n\u4ee5\u4e0b\u306e\u30b3\u30fc\u30c9\u3092\u30b3\u30d4\u30fc\u3057\u3066\u3001YAML\u30d5\u30a1\u30a4\u30eb\u306b\u8cbc\u308a\u4ed8\u3051\u307e\u3059\u3002<\/p>\n\n\n\n
services:\n phi3_vision:\n build:\n context: .\n dockerfile: Dockerfile\n image: phi3_vision\n runtime: nvidia\n container_name: phi3_vision\n ports:\n - \"8888:8888\"\n volumes:\n - .:\/app\/phi3_vision\n deploy:\n resources:\n reservations:\n devices:\n - driver: nvidia\n count: 1\n capabilities: [gpu]\n command: >\n bash -c '\/app\/.venv\/bin\/jupyter lab --ip=\"*\" --port=8888 --NotebookApp.token=\"\" --NotebookApp.password=\"\" --no-browser --allow-root'<\/code><\/pre>\n\n\n\n[Ctrl + S]\u30ad\u30fc\u3067\u5909\u66f4\u5185\u5bb9\u3092\u4fdd\u5b58\u3057\u3001[Ctrl + X]\u30ad\u30fc\u3067\u7de8\u96c6\u30e2\u30fc\u30c9\u304b\u3089\u629c\u3051\u307e\u3059\u3002<\/p>\n\n\n\n\n<\/div><\/div>\n\n\n\n
STEP<\/span><\/div>Docker\u30b3\u30f3\u30c6\u30ca\u3092\u8d77\u52d5<\/div>
\n
Dockerfile\u304b\u3089\u30d3\u30eb\u30c9\u3057\u3066\u30b3\u30f3\u30c6\u30ca\u3092\u8d77\u52d5\u3057\u307e\u3059\u3002 <\/p>\n\n\n\n
docker compose up<\/code><\/pre>\n\n\n\n <\/p>\n\n\n\n
Docker\u306e\u8d77\u52d5\u5f8c\u306b\u30d6\u30e9\u30a6\u30b6\u306e\u691c\u7d22\u7a93\u306b”localhost:8888″\u3092\u5165\u529b\u3059\u308b\u3068\u3001Jupyter Lab\u3092\u30d6\u30e9\u30a6\u30b6\u3067\u8868\u793a\u3067\u304d\u307e\u3059\u3002<\/p>\n\n\n\n
localhost:8888<\/code><\/pre>\n<\/div><\/div>\n<\/div>\n\n\n\nPhi-3 Vision\u306e\u5b9f\u88c5<\/h3>\n\n\n\n![\"\u898b\u51fa\u3057\u753b\u50cf\"](\"https:\/\/highreso.jp\/edgehub\/wp-content\/uploads\/2024\/06\/shilvercurve_midashi.jpg\")
<\/figure>\n\n\n\nDocker\u30b3\u30f3\u30c6\u30ca\u3067\u8d77\u52d5\u3057\u305fJupyter Lab\u4e0a\u3067Phi-3 Vision\u306e\u5b9f\u88c5\u3092\u3057\u307e\u3059\u3002<\/span><\/p>\n\n\n\n\n
STEP<\/span><\/div>\u30e9\u30a4\u30d6\u30e9\u30ea\u306e\u30a4\u30f3\u30dd\u30fc\u30c8<\/div>
\n
Jupyter Lab\u306e\u30b3\u30fc\u30c9\u30bb\u30eb\u306b\u6b21\u306e\u30b3\u30de\u30f3\u30c9\u3092\u5b9f\u884c\u3057\u3066\u3001\u30e9\u30a4\u30d6\u30e9\u30ea\u3092\u30a4\u30f3\u30dd\u30fc\u30c8\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
from PIL import Image \nimport requests \nfrom transformers import AutoModelForCausalLM \nfrom transformers import AutoProcessor<\/code><\/pre>\n<\/div><\/div>\n\n\n\nSTEP<\/span><\/div>\u30e2\u30c7\u30eb\u3068\u30d7\u30ed\u30bb\u30c3\u30b5\u30fc\u306e\u8a2d\u5b9a<\/div>
\n
Phi-3 Vision\u306e\u30e2\u30c7\u30eb\u3068\u30c8\u30fc\u30af\u30ca\u30a4\u30b6\u30fc\u3092\u8aad\u307f\u8fbc\u307f\u307e\u3059\u3002<\/p>\n\n\n\n
model_id = \"microsoft\/Phi-3-vision-128k-instruct\" \nmodel = AutoModelForCausalLM.from_pretrained(\n model_id,\n device_map=\"cuda\",\n trust_remote_code=True,\n torch_dtype=\"auto\",\n _attn_implementation='flash_attention_2'\n)\n\nprocessor = AutoProcessor.from_pretrained(\nmodel_id,\ntrust_remote_code=True\n)<\/code><\/pre>\n\n\n\n\u30b3\u30fc\u30c9\u306e\u8aac\u660e<\/span><\/div>\n
microsoft\/Phi-3-vision-128k-instruct<\/span> : Phi-3 Vision\u306e\u30e2\u30c7\u30eb\u30bf\u30a4\u30d7\u3092\u6307\u5b9a\u3057\u307e\u3059\u3002<\/p>\n\n\n\nAutoModelForCausalLM.from_pretrained<\/span>(<\/span>)<\/span>\uff1a\u30e2\u30c7\u30eb\u3092\u8aad\u307f\u8fbc\u307f\u307e\u3059\u3002<\/p>\n\n\n\nflash_attention_2<\/span>\uff1a\u63a8\u8ad6\u3092\u9ad8\u901f\u5316\u3059\u308bFlash Attention\u3092\u6307\u5b9a\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\npipeline<\/span>(<\/span>)<\/span> : \u30c6\u30ad\u30b9\u30c8\u751f\u6210\u30bf\u30b9\u30af\u306e\u305f\u3081\u306eTransformer\u306e\u30d1\u30a4\u30d7\u30e9\u30a4\u30f3\u3092\u8a2d\u5b9a\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\nAutoProcessor.from_pretrained<\/span>(<\/span>)<\/span>\uff1a\u753b\u50cf\u51e6\u7406\u306e\u30d7\u30ed\u30bb\u30c3\u30b5\u30fc\u3068\u30c8\u30fc\u30af\u30ca\u30a4\u30b6\u30fc\u3092\u8aad\u307f\u8fbc\u307f\u307e\u3059\u3002<\/p>\n<\/div><\/div>\n\n\n\n| \u30e2\u30c7\u30ebID<\/td> | \u30d1\u30e9\u30e1\u30fc\u30bf\u30b5\u30a4\u30ba<\/td> | \u30b3\u30f3\u30c6\u30ad\u30b9\u30c8\u9577<\/td> | GPU\u30e1\u30e2\u30ea\u4f7f\u7528\u91cf<\/td><\/tr> |
| microsoft\/Phi-3-vision-128k-instruct<\/td> | 42\u5104\u30d1\u30e9\u30e1\u30fc\u30bf<\/td> | 128k\u30c8\u30fc\u30af\u30f3<\/td> | 8GB<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n \u30e2\u30c7\u30eb\u3092\u8aad\u307f\u8fbc\u3080\u969b\u306bGPU\u30e1\u30e2\u30ea\u3092\u6d88\u8cbb\u3057\u307e\u3059\u306e\u3067\u3001\u4f59\u88d5\u3092\u6301\u3063\u305fGPU\u30e1\u30e2\u30ea\u3092\u3054\u7528\u610f\u304f\u3060\u3055\u3044\u3002<\/p>\n<\/div><\/div>\n\n\n\n STEP<\/span><\/div>\u753b\u50cf\u304b\u3089\u30c6\u30ad\u30b9\u30c8\u3092\u751f\u6210\u3059\u308b\u95a2\u6570\u3092\u5b9a\u7fa9<\/div> \n \u753b\u50cf\u3068\u30c6\u30ad\u30b9\u30c8\u3092\u5165\u529b\u3057\u3066\u3001\u30c6\u30ad\u30b9\u30c8\u3092\u751f\u6210\u3059\u308b\u95a2\u6570\u3092\u5b9a\u7fa9\u3057\u307e\u3059\u3002<\/span><\/p>\n\n\n\ndef textvision(messages, path_image):\n \n image = Image.open(path_image)\n\n prompt = processor.tokenizer.apply_chat_template(\n messages,\n tokenize=False,\n add_generation_prompt=True\n )\n\n inputs = processor(prompt, [image], return_tensors=\"pt\").to(\"cuda:0\")\n\n generate_ids = model.generate(\n **inputs,\n eos_token_id=processor.tokenizer.eos_token_id,\n max_new_tokens=256,\n do_sample=False,\n )\n\n generate_ids = generate_ids[:, inputs[\"input_ids\"].shape[1] :]\n\n response = processor.batch_decode(\n generate_ids,\n skip_special_tokens=True,\n clean_up_tokenization_spaces=False\n )[0]\n \n resized_image = image.resize((int(image.width * 0.5), int(image.height * 0.5)))\n resized_image.show()\n print(response)<\/code><\/pre>\n\n\n\n\u30b3\u30fc\u30c9\u306e\u8aac\u660e<\/span><\/div>\n Image.open()<\/span><\/p>\n\n\n\n\u6307\u5b9a\u3055\u308c\u305f\u30d1\u30b9\u3092\u3082\u3068\u306b\u753b\u50cf\u3092\u958b\u304d\u307e\u3059\u3002<\/p>\n\n\n\n processor.tokenizer.apply_chat_template()<\/span><\/p>\n\n\n\n\u30e1\u30c3\u30bb\u30fc\u30b8\u306b\u30c1\u30e3\u30c3\u30c8\u30c6\u30f3\u30d7\u30ec\u30fc\u30c8\u3092\u9069\u7528\u3057\u307e\u3059\u3002<\/p>\n\n\n\n processor(prompt, [image], return_tensors=”pt”).to(“cuda:0”)<\/span><\/p>\n\n\n\n\u30c6\u30ad\u30b9\u30c8\u3068\u753b\u50cf\u3092\u30c6\u30f3\u30bd\u30eb\u5909\u63db\u3057\u3066Processor\u306b\u6e21\u3057\u3001GPU\u306b\u8ee2\u9001\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n model.generate() <\/span><\/p>\n\n\n\n\u5165\u529b\u30c7\u30fc\u30bf\u3092\u3082\u3068\u306b\u30c8\u30fc\u30af\u30f3ID\u3092\u751f\u6210\u3057\u307e\u3059\u3002<\/p>\n\n\n\n processor.batch_decode() <\/span><\/p>\n\n\n\n\u30e2\u30c7\u30eb\u304c\u751f\u6210\u3057\u305f\u30c8\u30fc\u30af\u30f3ID\u3092\u30c6\u30ad\u30b9\u30c8\u306b\u5909\u63db\u3057\u307e\u3059\u3002<\/p>\n<\/div><\/div>\n<\/div><\/div>\n<\/div>\n\n\n\n Phi-3 Vision\u3067\u753b\u50cf\u304b\u3089\u30c6\u30ad\u30b9\u30c8\u3092\u751f\u6210<\/h2>\n\n\n\n![\"\u898b\u51fa\u3057\u753b\u50cf\"](\"https:\/\/highreso.jp\/edgehub\/wp-content\/uploads\/2024\/06\/lotofbuildings_midashi.jpg\") <\/figure>\n\n\n\nPhi-3 Vision\u3092\u4f7f\u3063\u3066\u3001\u753b\u50cf\u306e\u5165\u529b\u304b\u3089\u30c6\u30ad\u30b9\u30c8\u3092\u751f\u6210\u3057\u3066\u307f\u307e\u3059\u3002<\/span><\/p>\n\n\n\n\u82f1\u8a9e\u3067\u753b\u50cf\u8aac\u660e<\/h3>\n\n\n\n\u300c\u6307\u5b9a\u3057\u305f\u753b\u50cf\u306b\u3064\u3044\u3066\u8aac\u660e\u3057\u3066\u304f\u3060\u3055\u3044\u3002\u300d\u3068\u3044\u3046\u30d7\u30ed\u30f3\u30d7\u30c8\u3092\u82f1\u8a9e\u3067\u5b9f\u884c\u3057\u3066\u307f\u307e\u3059\u3002<\/span><\/p>\n\n\n\nprompt_input = [\n {\"role\": \"user\", \"content\": \"<|image_1|>\\n Please describe this image.\"}\n]\n\npath_image = \"testimage1.jpg\"\n\ntextvision(prompt_input, path_image)<\/code><\/pre>\n\n\n\n\u30b3\u30fc\u30c9\u306e\u8aac\u660e<\/span><\/div>\n [{“role”: “user”, “content”: “<|image_1|>\\n Please describe this image.”}]<\/span><\/p>\n\n\n\n\u300cPlease describe this image\u300d\u3068\u3044\u3046\u30d7\u30ed\u30f3\u30d7\u30c8\u3092\u5165\u529b\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n path_image = “testimage1.jpg”<\/span><\/p>\n\n\n\n\u753b\u50cf\u3092\u6307\u5b9a\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n<\/div><\/div>\n\n\n \n ![\"\u732b\u306e\u306c\u3044\u3050\u308b\u307f\"](\"https:\/\/highreso.jp\/edgehub\/wp-content\/uploads\/2024\/06\/phi3vision_1-300x300.jpg\") <\/figure>\n<\/div>\n\n\n Please describe this image.
————————————–
(\u7ffb\u8a33)
\u3053\u306e\u753b\u50cf\u306b\u3064\u3044\u3066\u8aac\u660e\u3057\u3066\u304f\u3060\u3055\u3044\u3002<\/p>\n<\/span><\/span><\/span><\/div><\/div><\/div><\/div>\n\n\n The image features a cozy indoor setting with a wooden table at the center. On the table, there is a white cup of coffee with a heart-shaped foam on top, accompanied by a small potted plant. To the right of the cup, there is a plush toy of a cat with orange fur and white paws. The background shows a window with a view of a street scene, including a building with a red roof and a car parked on the street. The room is softly lit with warm lighting, creating a welcoming atmosphere.
————————————————————————–
\uff08\u7ffb\u8a33\uff09
\u753b\u50cf\u306b\u306f\u3001\u4e2d\u592e\u306b\u6728\u88fd\u306e\u30c6\u30fc\u30d6\u30eb\u304c\u3042\u308b\u5c45\u5fc3\u5730\u306e\u826f\u3044\u5ba4\u5185\u306e\u69d8\u5b50\u304c\u5199\u3063\u3066\u3044\u307e\u3059\u3002\u30c6\u30fc\u30d6\u30eb\u306e\u4e0a\u306b\u306f\u3001\u30cf\u30fc\u30c8\u578b\u306e\u6ce1\u304c\u4e57\u3063\u305f\u767d\u3044\u30b3\u30fc\u30d2\u30fc\u30ab\u30c3\u30d7\u3068\u3001\u5c0f\u3055\u306a\u9262\u690d\u3048\u306e\u690d\u7269\u304c\u3042\u308a\u307e\u3059\u3002\u30ab\u30c3\u30d7\u306e\u53f3\u5074\u306b\u306f\u3001\u30aa\u30ec\u30f3\u30b8\u8272\u306e\u6bdb\u4e26\u307f\u3068\u767d\u3044\u8db3\u3092\u6301\u3064\u732b\u306e\u306c\u3044\u3050\u308b\u307f\u304c\u7f6e\u304b\u308c\u3066\u3044\u307e\u3059\u3002\u80cc\u666f\u306b\u306f\u3001\u8d64\u3044\u5c4b\u6839\u306e\u5efa\u7269\u3084\u901a\u308a\u306b\u99d0\u8eca\u3055\u308c\u305f\u8eca\u306a\u3069\u304c\u898b\u3048\u308b\u7a93\u304c\u3042\u308a\u307e\u3059\u3002\u90e8\u5c4b\u306f\u6696\u304b\u3044\u7167\u660e\u3067\u3084\u308f\u3089\u304b\u304f\u7167\u3089\u3055\u308c\u3001\u6b53\u8fce\u306e\u96f0\u56f2\u6c17\u3092\u91b8\u3057\u51fa\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n<\/span><\/span><\/span><\/div><\/div><\/div><\/div>\n\n\n\u7d30\u90e8\u306b\u308f\u305f\u3063\u3066\u753b\u50cf\u306e\u8aac\u660e\u304c\u3067\u304d\u3066\u3044\u307e\u3059\uff01<\/p>\n\n\n\n \u65e5\u672c\u8a9e\u3067\u753b\u50cf\u8aac\u660e<\/h3>\n\n\n\n\u300c\u3053\u306e\u753b\u50cf\u306b\u3064\u3044\u3066\u8aac\u660e\u3057\u3066\u304f\u3060\u3055\u3044<\/span>\u3002\u300d\u3068\u3044\u3046\u30d7\u30ed\u30f3\u30d7\u30c8\u3092\u65e5\u672c\u8a9e\u3067\u5b9f\u884c\u3057\u3066\u307f\u307e\u3059\u3002<\/span><\/p>\n\n\n\nprompt_input = [\n {\"role\": \"user\", \"content\": \"<|image_1|>\\n \u3053\u306e\u753b\u50cf\u306b\u3064\u3044\u3066\u8aac\u660e\u3057\u3066\u304f\u3060\u3055\u3044\u3002\"},\n {\"role\": \"assistant\", \"content\": \"\u3042\u306a\u305f\u306f\u753b\u50cf\u3092\u65e5\u672c\u8a9e\u3067\u56de\u7b54\u3059\u308b\u30a2\u30b7\u30b9\u30bf\u30f3\u30c8\u3067\u3059\u3002\"}\n]\n\npath_image = \"testimage3.jpg\"\n\ntextvision(prompt_input, path_image)<\/code><\/pre>\n\n\n\n\u30b3\u30fc\u30c9\u306e\u8aac\u660e<\/span><\/div>\n {“role”: “assistant”, “content”: “\u3042\u306a\u305f\u306f\u65e5\u672c\u8a9e\u3067\u56de\u7b54\u3059\u308b\u30a2\u30b7\u30b9\u30bf\u30f3\u30c8\u3067\u3059”}<\/span><\/p>\n\n\n\n\u56de\u7b54\u3059\u308b\u30e2\u30c7\u30eb\u306e\u5f79\u5272\uff08\u30ed\u30fc\u30eb\uff09\u3092\u6307\u5b9a\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n<\/div><\/div>\n\n\n \n ![\"\u732b\u30b7\u30a7\u30d5\"](\"https:\/\/highreso.jp\/edgehub\/wp-content\/uploads\/2024\/06\/phi3vision_2-300x300.jpg\") <\/figure>\n<\/div>\n\n\n Please describe this image.
————————————–
(\u7ffb\u8a33)
\u3053\u306e\u753b\u50cf\u306b\u3064\u3044\u3066\u8aac\u660e\u3057\u3066\u304f\u3060\u3055\u3044\u3002<\/p>\n<\/span><\/span><\/span><\/div><\/div><\/div><\/div>\n\n\n \u3053\u306e\u753b\u50cf\u3067\u306f\u3001\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306b\u6e96\u5099\u3055\u308c\u305f\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u30b7\u30fc\u30f3\u304c\u63cf\u304b\u308c\u3066\u3044\u307e\u3059\u3002\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u4e2d\u3067\u3001\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u4e2d\u3067\u3001\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u4e2d\u3067\u3001\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u4e2d\u3067\u3001\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u4e2d\u3067\u3001\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u4e2d\u3067\u3001\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u4e2d\u3067\u3001\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u4e2d\u3067\u3001\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf\u30fc\u306e\u4e2d\u3067\u3001\u30ad\u30e3\u30c3\u30b7\u30e5\u30ec\u30b9\u30bf<\/p>\n<\/span><\/span><\/span><\/div><\/div><\/div><\/div>\n\n\n\u65e5\u672c\u8a9e\u3067\u306e\u56de\u7b54\u306f\u3001\u5b8c\u5168\u306b\u7834\u7dbb\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n \u82f1\u8a9e\u3067\u540c\u3058\u5185\u5bb9\u3092\u751f\u6210\u3057\u3066\u307f\u307e\u3059\u3002<\/span><\/p>\n\n\n\n <\/figure>\n<\/div>\n\n\n Please describe this image.
————————————–
(\u7ffb\u8a33)
\u3053\u306e\u753b\u50cf\u306b\u3064\u3044\u3066\u8aac\u660e\u3057\u3066\u304f\u3060\u3055\u3044\u3002<\/p>\n<\/span><\/span><\/span><\/div><\/div><\/div><\/div>\n\n\n The image features a cat dressed in a white chef’s hat and a white jacket with black buttons, sitting on a wooden cutting board in a kitchen setting. The cat has a playful expression and is looking directly at the camera. The kitchen is equipped with various cooking utensils such as pans, a rolling pin, and a pot, as well as a range hood and a stove. There are also shelves with jars and spices, and a tomato is visible on the counter.
————————————————————————–
\uff08\u7ffb\u8a33\uff09
\u753b\u50cf\u306b\u306f\u3001\u767d\u3044\u30b7\u30a7\u30d5\u5e3d\u3068\u9ed2\u3044\u30dc\u30bf\u30f3\u306e\u4ed8\u3044\u305f\u767d\u3044\u30b8\u30e3\u30b1\u30c3\u30c8\u3092\u7740\u305f\u732b\u304c\u3001\u30ad\u30c3\u30c1\u30f3\u306e\u6728\u88fd\u306e\u307e\u306a\u677f\u306e\u4e0a\u306b\u5ea7\u3063\u3066\u3044\u308b\u69d8\u5b50\u304c\u63cf\u304b\u308c\u3066\u3044\u307e\u3059\u3002\u732b\u306f\u904a\u3073\u5fc3\u306e\u3042\u308b\u8868\u60c5\u3067\u30ab\u30e1\u30e9\u3092\u898b\u3066\u3044\u307e\u3059\u3002\u30ad\u30c3\u30c1\u30f3\u306b\u306f\u3001\u30d5\u30e9\u30a4\u30d1\u30f3\u3001\u3081\u3093\u68d2\u3001\u934b\u306a\u3069\u306e\u3055\u307e\u3056\u307e\u306a\u8abf\u7406\u5668\u5177\u3084\u3001\u30ec\u30f3\u30b8\u30d5\u30fc\u30c9\u3068\u30b9\u30c8\u30fc\u30d6\u304c\u5099\u308f\u3063\u3066\u3044\u307e\u3059\u3002\u307e\u305f\u3001\u68da\u306b\u306f\u74f6\u3084\u30b9\u30d1\u30a4\u30b9\u304c\u7f6e\u304b\u308c\u3066\u304a\u308a\u3001\u30ab\u30a6\u30f3\u30bf\u30fc\u306b\u306f\u30c8\u30de\u30c8\u304c\u898b\u3048\u307e\u3059\u3002<\/p>\n<\/span><\/span><\/span><\/div><\/div><\/div><\/div>\n\n\n\u82f1\u8a9e\u5165\u529b\u306e\u5834\u5408\u306f\u3001\u9ad8\u3044\u7cbe\u5ea6\u306e\u5fdc\u7b54\u306b\u306a\u308a\u307e\u3057\u305f\uff01<\/p>\n\n\n\n \u753b\u50cf\u304b\u3089\u30c6\u30ad\u30b9\u30c8\u306e\u8aad\u307f\u53d6\u308a\uff08OCR\uff09<\/h3>\n\n\n\n\u300c\u753b\u50cf\u306e\u30c6\u30ad\u30b9\u30c8\u3092OCR\u3067\u62bd\u51fa\u3057\u3066\u304f\u3060\u3055\u3044\u300d\u3068\u3044\u3046\u30d7\u30ed\u30f3\u30d7\u30c8\u3092\u6295\u3052\u3066\u307f\u307e\u3059\u3002<\/span><\/p>\n\n\n\nprompt_input = [\n {\"role\": \"user\", \"content\": \"<|image_1|>\\n OCR the text of the image. Extract the text of the following fields and put it in a JSON format: \\\n 'Company', 'Name','Title','Phone', 'E-mail','URL','Adress'\"}]\n\npath_image = \"testocr.jpg\"\n\ntextvision(prompt_input, path_image)<\/code><\/pre>\n\n\n\n ![\"\u898b\u51fa\u3057\u753b\u50cf\"](\"https:\/\/highreso.jp\/edgehub\/wp-content\/uploads\/2024\/06\/catocr_midashi.jpg\") <\/figure>\n<\/div>\n\n
|
![\"\u898b\u51fa\u3057\u753b\u50cf\"](\"https:\/\/highreso.jp\/edgehub\/wp-content\/uploads\/2024\/06\/colorfulart_midashi-1.jpg\")
<\/figure>\n\n\n\n![\"\"](\"https:\/\/highreso.jp\/edgehub\/wp-content\/uploads\/2024\/06\/phi3toha-eyecatch-300x158.jpg\")
<\/figure><\/div>\t\t\t\t\t\n\t\t\t\t\t\tPhi-3\u3068\u306f\uff1f\u4f7f\u3044\u65b9\u304b\u3089\u6027\u80fd\u3001\u5546\u7528\u5229\u7528\u307e\u3067\u5206\u304b\u308a\u3084\u3059\u304f\u89e3\u8aac\uff01<\/a>\n\t\t\t\t\t\tPhi-3\u306f\u3001Microsoft\u304c\u958b\u767a\u3057\u305f\u7121\u6599\u3067\u4f7f\u3048\u308b\u9ad8\u6027\u80fd\u306aSLM(\u5c0f\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb)\u3067\u3059\u3002 \u8efd\u91cf\u30e2\u30c7\u30eb\u3067\u3042\u308b\u305f\u3081\u3001\u5c0f\u3055\u306aGPU\u30e1\u30e2\u30ea\u306e\u30ed\u30fc\u30ab\u30eb\u74b0\u5883\u3067\u3082\u52d5\u304b\u3059\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002 \u3053…<\/span>\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t<\/div>\n\t\t<\/div>\n\n\n1\u6642\u959350\u5186\u304b\u3089\u4f7f\u3048\u308bGPU\u30af\u30e9\u30a6\u30c9\u300cGPUSOROBAN\u300d<\/div>![\"GPUSOROBAN\"](\"https:\/\/highreso.jp\/edgehub\/wp-content\/uploads\/2024\/06\/gpusorobanlllm_banner.jpg\")
<\/a><\/div><\/div><\/i><\/i><\/i><\/i><\/i><\/div>GPU\u306e\u30b3\u30b9\u30c8\u3067\u304a\u60a9\u307f\u306e\u65b9\u306b\uff01<\/div>\u751f\u6210AI\u30fbLLM\u5411\u3051\u306e\u9ad8\u901fGPU\u3092\u696d\u754c\u6700\u5b89\u7d1a\u306e1\u6642\u959350\u5186\u304b\u3089\u5229\u7528\u3067\u304d\u307e\u3059\u3002\r\n\u30a4\u30f3\u30bf\u30fc\u30cd\u30c3\u30c8\u74b0\u5883\u3055\u3048\u3042\u308c\u3070\u3001GPU\u30b5\u30fc\u30d0\u30fc\u3092\u3059\u3050\u306b\u4f7f\u3048\u307e\u3059\u3002\u30a4\u30f3\u30b9\u30bf\u30f3\u30b9\u3092\u8d77\u52d5\u3057\u3066\u3044\u308b\u3068\u304d\u306e\u307f\u306e\u8ab2\u91d1\u3067\u3001\u30b3\u30b9\u30c8\u3092\u6700\u5c0f\u5316\u3067\u304d\u307e\u3059\u3002\r\n<\/div><\/div><\/div>![\"\u898b\u51fa\u3057\u753b\u50cf\"](\"https:\/\/highreso.jp\/edgehub\/wp-content\/uploads\/2024\/06\/colorfulart3_midashi.jpg\")
<\/figure>\n\n\n\n![\"\"](\"https:\/\/highreso.jp\/edgehub\/wp-content\/uploads\/2024\/06\/ubuntudocker-eyecatch-300x158.jpg\")
<\/figure><\/div>\t\t\t\t\t