Spaces:
Running
Running
File size: 13,809 Bytes
8f38740 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 |
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"id": "cdad6b21-030a-40d3-9b31-a229e5b6196d",
"metadata": {},
"outputs": [],
"source": [
"import torch\n",
"from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, AutoTokenizer, AutoConfig, CLIPImageProcessor"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "1f832710-0e8c-42ec-b581-1b15fd2a6acc",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[2024-01-25 14:31:58,511] [INFO] [real_accelerator.py:110:get_accelerator] Setting ds_accelerator to cuda (auto detect)\n"
]
}
],
"source": [
"from model import LlavaPhiForCausalLM"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "9e68f1d4-1ae3-4d45-b818-4600218d2215",
"metadata": {},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "e5e13e666e3a43d4ad26cc70904abee8",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Loading checkpoint shards: 0%| | 0/3 [00:00<?, ?it/s]"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"model_name = \"RaviNaik/Llava-Phi2\"\n",
"model = LlavaPhiForCausalLM.from_pretrained(model_name)"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "49edfa0d-e08a-4d3c-a1d6-34068b122419",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.\n"
]
}
],
"source": [
"tokenizer = AutoTokenizer.from_pretrained(model_name)"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "dcec20cd-d946-42d7-8e10-c198cd49b486",
"metadata": {},
"outputs": [],
"source": [
"image_processor = CLIPImageProcessor.from_pretrained(model_name)\n",
"mm_use_im_start_end = getattr(model.config, \"mm_use_im_start_end\", False)\n",
"mm_use_im_patch_token = getattr(model.config, \"mm_use_im_patch_token\", True)"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "443c13c4-b7e6-4bc5-b6c7-c577bd4708f6",
"metadata": {},
"outputs": [],
"source": [
"if mm_use_im_patch_token:\n",
" tokenizer.add_tokens([DEFAULT_IMAGE_PATCH_TOKEN], special_tokens=True)\n",
"if mm_use_im_start_end:\n",
" tokenizer.add_tokens([DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN], special_tokens=True)\n",
" \n",
"if hasattr(model.config, \"max_sequence_length\"):\n",
" context_len = model.config.max_sequence_length\n",
"else:\n",
" context_len = 2048"
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "d8caee43-0d2a-46d4-bdbc-2cfc7dec9e52",
"metadata": {},
"outputs": [],
"source": [
"from transformers import WhisperProcessor, WhisperForConditionalGeneration"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "3acea526-d8ae-4eb6-8dfc-4ea72651b547",
"metadata": {},
"outputs": [],
"source": [
"class AudioLanguageConnector:\n",
" def __init__(self, projection_dim):\n",
" model_name = \"microsoft/phi-2\"\n",
" self.phi2_tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)\n",
" self.phi2_tokenizer.pad_token = self.phi2_tokenizer.eos_token\n",
" self.phi2_tokenizer.max_length = projection_dim\n",
"\n",
" def __call__(self, text):\n",
" text = f\"<audio_start> {text} <audio_end>\"\n",
" tokens = self.phi2_tokenizer(text, return_tensors=\"pt\", return_attention_mask=False)\n",
" return tokens\n",
" \n",
"\n",
"class WhisperWithProjection:\n",
" def __init__(self, projection_dim, device):\n",
" self.device = device\n",
" self.processor = WhisperProcessor.from_pretrained(\"openai/whisper-tiny\", device_map=device)\n",
" self.model = WhisperForConditionalGeneration.from_pretrained(\"openai/whisper-tiny\", device_map=device)\n",
" self.model.config.forced_decoder_ids = None\n",
" # self.audio_language_connector = AudioLanguageConnector(projection_dim)\n",
" \n",
" def __call__(self, audio):\n",
" input_features = self.processor(audio[\"array\"],\n",
" sampling_rate=audio[\"sampling_rate\"],\n",
" return_tensors=\"pt\").input_features\n",
" # generate token ids\n",
" predicted_ids = self.model.generate(input_features.to(self.device))\n",
" # decode token ids to text \n",
" transcription = self.processor.batch_decode(predicted_ids, skip_special_tokens=True)\n",
"\n",
" # audio_embeddings = self.audio_language_connector(transcription)\n",
" return transcription"
]
},
{
"cell_type": "code",
"execution_count": 10,
"id": "a2757c91-2ec1-4fe7-9216-03740bf80061",
"metadata": {},
"outputs": [],
"source": [
"IGNORE_INDEX = -100\n",
"IMAGE_TOKEN_INDEX = -200\n",
"DEFAULT_IMAGE_TOKEN = \"<image>\"\n",
"DEFAULT_IMAGE_PATCH_TOKEN = \"<im_patch>\"\n",
"DEFAULT_IM_START_TOKEN = \"<im_start>\"\n",
"DEFAULT_IM_END_TOKEN = \"<im_end>\"\n",
"\n",
"from conversation import conv_templates, SeparatorStyle\n",
"\n",
"class MultiModalPhi2:\n",
" def __init__(self, modelname_or_path=\"RaviNaik/Llava-Phi2\",\n",
" temperature=0.2,\n",
" max_new_tokens=1024,\n",
" device=\"cuda:0\"):\n",
" self.model_name = modelname_or_path\n",
" self.temperature = temperature\n",
" self.max_new_tokens = max_new_tokens\n",
" self.device = device\n",
" self.disable_torch_init()\n",
" self.whisper_w_proj = WhisperWithProjection(projection_dim=512, device=device)\n",
" self.load_pretrained_model()\n",
" \n",
" def disable_torch_init(self):\n",
" \"\"\"\n",
" Disable the redundant torch default initialization to accelerate model creation.\n",
" \"\"\"\n",
" setattr(torch.nn.Linear, \"reset_parameters\", lambda self: None)\n",
" setattr(torch.nn.LayerNorm, \"reset_parameters\", lambda self: None)\n",
" \n",
" def load_pretrained_model(self):\n",
" self.model = LlavaPhiForCausalLM.from_pretrained(self.model_name, device_map=self.device)\n",
" self.tokenizer = AutoTokenizer.from_pretrained(self.model_name)\n",
" self.image_processor = CLIPImageProcessor.from_pretrained(self.model_name)\n",
" mm_use_im_start_end = getattr(self.model.config, \"mm_use_im_start_end\", False)\n",
" mm_use_im_patch_token = getattr(self.model.config, \"mm_use_im_patch_token\", True)\n",
" if mm_use_im_patch_token:\n",
" self.tokenizer.add_tokens([DEFAULT_IMAGE_PATCH_TOKEN], special_tokens=True)\n",
" if mm_use_im_start_end:\n",
" self.tokenizer.add_tokens([DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN], special_tokens=True)\n",
" \n",
" def tokenizer_image_token(self, prompt, tokenizer, image_token_index=IMAGE_TOKEN_INDEX, return_tensors=None):\n",
" prompt_chunks = [tokenizer(chunk).input_ids for chunk in prompt.split('<image>')]\n",
" \n",
" def insert_separator(X, sep):\n",
" return [ele for sublist in zip(X, [sep]*len(X)) for ele in sublist][:-1]\n",
" \n",
" input_ids = []\n",
" offset = 0\n",
" if len(prompt_chunks) > 0 and len(prompt_chunks[0]) > 0 and prompt_chunks[0][0] == tokenizer.bos_token_id:\n",
" offset = 1\n",
" input_ids.append(prompt_chunks[0][0])\n",
" for x in insert_separator(prompt_chunks, [image_token_index] * (offset + 1)):\n",
" input_ids.extend(x[offset:])\n",
" \n",
" if return_tensors is not None:\n",
" if return_tensors == 'pt':\n",
" return torch.tensor(input_ids, dtype=torch.long)\n",
" raise ValueError(f'Unsupported tensor type: {return_tensors}')\n",
" return input_ids\n",
" \n",
" def __call__(self, text, audio, image):\n",
" qs = DEFAULT_IM_START_TOKEN + DEFAULT_IMAGE_TOKEN + DEFAULT_IM_END_TOKEN + '\\n' + text\n",
" conv = conv_templates[\"phi-2_v0\"].copy()\n",
" conv.append_message(conv.roles[0], qs)\n",
" conv.append_message(conv.roles[1], None)\n",
" prompt = conv.get_prompt()\n",
"\n",
" image_tensor = self.image_processor.preprocess(image, return_tensors='pt')['pixel_values'].to(self.device)\n",
" \n",
" input_ids = self.tokenizer_image_token(prompt, self.tokenizer, IMAGE_TOKEN_INDEX, return_tensors='pt').unsqueeze(0)\n",
" if audio is not None:\n",
" audio_transcript = self.whisper_w_proj(audio)\n",
" audio_embed = self.tokenizer(audio_transcript, return_tensors='pt')[\"input_ids\"]\n",
" input_ids = torch.concat([input_ids, audio_embed], dim=1)\n",
" input_ids = input_ids.to(self.device)\n",
" \n",
" stop_str = conv.sep if conv.sep_style != SeparatorStyle.TWO else conv.sep2\n",
"\n",
" with torch.inference_mode():\n",
" output_ids = self.model.generate(\n",
" input_ids,\n",
" images=image_tensor,\n",
" do_sample=True,\n",
" temperature=self.temperature,\n",
" max_new_tokens=self.max_new_tokens,\n",
" eos_token_id=self.tokenizer.eos_token_id, # End of sequence token\n",
" pad_token_id=self.tokenizer.eos_token_id, # Pad token\n",
" use_cache=True,\n",
" )\n",
"\n",
" input_token_len = input_ids.shape[1]\n",
" n_diff_input_output = (input_ids != output_ids[:, :input_token_len]).sum().item()\n",
" if n_diff_input_output > 0:\n",
" print(f'[Warning] {n_diff_input_output} output_ids are not the same as the input_ids')\n",
" outputs = self.tokenizer.batch_decode(output_ids[:, input_token_len:], skip_special_tokens=True)[0]\n",
" outputs = outputs.strip()\n",
" if outputs.endswith(stop_str):\n",
" outputs = outputs[:-len(stop_str)]\n",
" outputs = outputs.strip()\n",
" return outputs"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "cc47e6a0-3544-4a60-930f-ccae87ef945a",
"metadata": {},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "9ef56077307d4cef907e25b092061611",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Loading checkpoint shards: 0%| | 0/3 [00:00<?, ?it/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.\n"
]
}
],
"source": [
"multimodal_phi2 = MultiModalPhi2()"
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "cb8aca1b-7d75-45e7-b5a4-71d151f792e1",
"metadata": {},
"outputs": [],
"source": [
"from PIL import Image\n",
"import requests\n",
"\n",
"url = \"https://www.ilankelman.org/stopsigns/australia.jpg\"\n",
"image = Image.open(requests.get(url, stream=True).raw)\n",
"\n",
"from datasets import load_dataset\n",
"audio_ds = load_dataset(\"hf-internal-testing/librispeech_asr_dummy\", \"clean\", split=\"validation\")\n",
"audio = audio_ds[0][\"audio\"]\n",
"\n",
"text = \"tell me about the image\""
]
},
{
"cell_type": "code",
"execution_count": 14,
"id": "6767efc6-be4f-44d3-84ff-34db57d9f940",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"'In the image, there is a Chinese writing on a pole in a foreign language. This suggests that the image was taken in a foreign country, possibly in a foreign country. The sign is in a foreign language, which might be in a foreign language. The sign is written in Japanese, which is a common language in Japan. The sign is also written in two different languages, which suggests that it is written in a language that is not in the native language.'"
]
},
"execution_count": 14,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"multimodal_phi2(text, None, image)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0bdd0b8a-709b-4c82-ac1d-dc746d3a0748",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.10.12"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|