{
"cells": [
{
"cell_type": "code",
"execution_count": 3,
"id": "d89263a2-a275-4c06-83fd-7e7eee0cb8b8",
"metadata": {},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "08a7dd60da6f4a2c957dfbd1795d7596",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Loading pipeline components...: 0%| | 0/7 [00:00"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"prompts = [\n",
"\"an astronaut riding a horse\"\n",
",\"A young woman with striking blue eyes and pointed ears, adorned with a floral kimono and a tattoo. Her hair is styled in a braid, and she wears a pair of ears\"\n",
",\"A muscular, topless male with tiger-like ears and a tail stands in a forest, holding a sword and wearing a blue outfit, gazing directly at the viewer\"\n",
",\"A young woman with striking features, a mix of black and gold hair floating around her head, vibrant yellow eyes half-closed, and a blue sweater contrasting the black background, framed in a simple yet striking composition.\"\n",
",'A fluffy domestic cat with piercing green eyes sits attentively in a sunlit room filled with natural light, its soft fur reflecting warm hues of orange through golden windows.'\n",
",\"A fierce woman in black-and-white, wearing a spiked iron mask with sharp metallic spikes, sporting a high ponytail, her skin marked with battle-dirt, and eyes that reflect determination and strength. The mask's menacing aura contrasts with her intense expression, capturing her mysterious and intimidating nature.\"\n",
",\"A close-up of an astronaut's helmet with frosted, opaque visor, reflecting space's cold, frozen texture, resting on the visor a butterfly with vibrant, intricately patterned wings, and distant stars' faint glow.\"\n",
"#,\"A watercolor painting of a knight in a hazy blue field, his armor blending soft greys and silvers, holding a large red rose, with a serene yet commanding posture against distant mountains and wildflowers.\"\n",
"#,\"A warm glow of lanterns casts soft light over a snowy forest trail lined with tall, snow-covered trees, their branches casting a gentle glow. Soft falling snowflakes create a serene atmosphere, while a dark, mysterious background adds to the solitude and magic of the scene.\"\n",
"#,\"A hauntingly ethereal figure with decaying flesh and glowing metallic enhancements, blending Art Nouveau and cybernetic futurism, stands in a twilight forest bathed in warm sunset hues, casting golden light on surreal, ghostly trees with autumn leaves. The atmosphere is melancholic, evoking the styles of Rockwell and Parrish.\"\n",
",\"A radiant voluptuous woman in Arizona's Grand Canyon at twilight, her fiery ginger hair cascading over a halter top with intricate lace, plunging neckline, black thongs, silver rings, and stacked bangles, surrounded by deep purples and burnt oranges in the sky, her light-blue eyes glowing with wonder as she gazes at the luminous beauty of the night.\"\n",
",\"A man with hair, a white suit and black scarf, anime-style elements, glowing pink horns, and cyberpunk manga art style, standing in a shadowy smoke-filled background, with glowing pink eyes and a side view portrait of his face, white skin, and ear piercings.\"\n",
",\"There is a young male character standing against a vibrant, colorful graffiti wall. he is wearing a straw hat, a black jacket adorned with gold accents, and black shorts.\"\n",
",\"A black BMW M3 sports car with black and yellow rims\"\n",
",\"A young girl in a flowing, vibrant dress, her glowing eyes capturing the warmth of the day, sits on a grassy field, surrounded by anime-style elements.\"\n",
"]\n",
"prompts2 = [ \"cat\", \"dog\", \"girl\", \"man\", \"A black BMW M3 sports car with black and yellow rims\", \"an astronaut riding a horse\",\"A muscular, topless male with tiger-like ears and a tail stands in a forest, holding a sword and wearing a blue outfit, gazing directly at the viewer\",\"white cyborg knight riding cyber horse with wings, long white gown, holding scythe, skeleton horse, zombies cyberpunk armor, feathers\",\"A striking character with red eyes and black uniform wields a sword in a defensive stance, poised for battle amidst a stark white background with vibrant red accents.\"]\n",
"\n",
"generator = torch.Generator(device=\"cuda\").manual_seed(42)\n",
"negative_prompt=\"low quality\"\n",
"\n",
"all_images = []\n",
"# Обработка батчей с прогресс-баром\n",
"for i, prompt in enumerate(prompts):\n",
" image = pipeline(\n",
" prompt = prompt,\n",
" negative_prompt = negative_prompt,\n",
" guidance_scale = 4,\n",
" width = 512,\n",
" height = 640,\n",
" seed = 42,\n",
" batch_size = 1,\n",
" )[0]\n",
" all_images.extend(image)\n",
"\n",
"import matplotlib.pyplot as plt\n",
"import math\n",
"\n",
"def display_image_grid(images, prompts, cols=3, save_path=None):\n",
" \"\"\"\n",
" Отображает грид изображений с сохранением соотношения сторон и подписями.\n",
" \"\"\"\n",
" n = len(images)\n",
" rows = math.ceil(n / cols)\n",
"\n",
" # Создаем фигуру с учетом реального соотношения сторон (640/576 ≈ 1.11)\n",
" fig_width = cols * 4\n",
" fig_height = rows * 4.5\n",
" fig, axes = plt.subplots(rows, cols, figsize=(fig_width, fig_height))\n",
"\n",
" # Если только один ряд или один столбец, делаем axes списком\n",
" if rows == 1:\n",
" axes = [axes]\n",
" if cols == 1:\n",
" axes = [[ax] for ax in axes]\n",
"\n",
" axes = axes.flatten()\n",
"\n",
" for i, (img, prompt) in enumerate(zip(images, prompts)):\n",
" ax = axes[i]\n",
" ax.imshow(img)\n",
" ax.axis(\"off\")\n",
" ax.set_aspect(\"equal\") # сохраняем соотношение сторон\n",
"\n",
" # Урезаем и разбиваем подпись\n",
" truncated_prompt = prompt[:80] + \"…\" if len(prompt) > 80 else prompt\n",
" words = truncated_prompt.split(\" \")\n",
" half = len(words) // 2\n",
" line1 = \" \".join(words[:half])\n",
" line2 = \" \".join(words[half:])\n",
" ax.set_title(f\"{line1}\\n{line2}\", fontsize=10, wrap=True)\n",
"\n",
" # Прячем лишние subplot'ы, если их больше, чем изображений\n",
" for j in range(len(images), len(axes)):\n",
" axes[j].axis(\"off\")\n",
"\n",
" plt.tight_layout(pad=1.5)\n",
"\n",
" if save_path:\n",
" plt.savefig(save_path, bbox_inches=\"tight\", dpi=400, format=\"jpeg\")\n",
"\n",
" plt.show()\n",
"\n",
"\n",
"display_image_grid(all_images, prompts, save_path=\"media/result_grid.jpg\")"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "044a8b5e-f786-48ea-86a3-cb581b3ca1c9",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Keyword arguments {'trust_remote_code': True} are not expected by SdxsPipeline and will be ignored.\n"
]
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "e65023bc60654e50bef56c3f84e0c76a",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Loading pipeline components...: 0%| | 0/7 [00:00"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"✨ refined_prompt_text:\n",
"girl, smiling, red eyes, blue hair, white shirt\n"
]
}
],
"source": [
"import torch\n",
"from diffusers import DiffusionPipeline\n",
"\n",
"device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
"dtype = torch.float16 if torch.cuda.is_available() else torch.float32\n",
"\n",
"pipe_id = \"/workspace/sdxs-1b\"\n",
"pipe = DiffusionPipeline.from_pretrained(\n",
" pipe_id,\n",
" torch_dtype=dtype,\n",
" trust_remote_code=True\n",
").to(device)\n",
"\n",
"print(\"Начинаем генерацию...\")\n",
"output = pipe(\n",
" width=512,\n",
" height=640,\n",
" prompt=\"girl, smiling, red eyes, blue hair, white shirt\",\n",
" negative_prompt=\"low quality, bad quality\",\n",
" seed=43,\n",
" refine_prompt=False\n",
")\n",
"\n",
"# Берем нулевую картинку и сам промпт (он будет строкой, так как мы подавали строку)\n",
"image = output.images[0]\n",
"refined_prompt_text = output.prompt \n",
"\n",
"image.show()\n",
"image.save(\"girl.jpg\")\n",
"\n",
"print(f\"✨ refined_prompt_text:\\n{refined_prompt_text}\")"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "76c7e40e-0326-42bc-b717-e46bcf790dfa",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"UNet2DConditionModel(\n",
" (conv_in): Conv2d(32, 320, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_proj): Timesteps()\n",
" (time_embedding): TimestepEmbedding(\n",
" (linear_1): Linear(in_features=320, out_features=1280, bias=True)\n",
" (act): SiLU()\n",
" (linear_2): Linear(in_features=1280, out_features=1280, bias=True)\n",
" )\n",
" (add_time_proj): Timesteps()\n",
" (add_embedding): TimestepEmbedding(\n",
" (linear_1): Linear(in_features=3584, out_features=1280, bias=True)\n",
" (act): SiLU()\n",
" (linear_2): Linear(in_features=1280, out_features=1280, bias=True)\n",
" )\n",
" (down_blocks): ModuleList(\n",
" (0): CrossAttnDownBlock2D(\n",
" (attentions): ModuleList(\n",
" (0-1): 2 x Transformer2DModel(\n",
" (norm): GroupNorm(32, 320, eps=1e-06, affine=True)\n",
" (proj_in): Conv2d(320, 320, kernel_size=(1, 1), stride=(1, 1))\n",
" (transformer_blocks): ModuleList(\n",
" (0): BasicTransformerBlock(\n",
" (norm1): LayerNorm((320,), eps=1e-05, elementwise_affine=True)\n",
" (attn1): Attention(\n",
" (to_q): Linear(in_features=320, out_features=320, bias=False)\n",
" (to_k): Linear(in_features=320, out_features=320, bias=False)\n",
" (to_v): Linear(in_features=320, out_features=320, bias=False)\n",
" (to_out): ModuleList(\n",
" (0): Linear(in_features=320, out_features=320, bias=True)\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" (norm2): LayerNorm((320,), eps=1e-05, elementwise_affine=True)\n",
" (attn2): Attention(\n",
" (to_q): Linear(in_features=320, out_features=320, bias=False)\n",
" (to_k): Linear(in_features=768, out_features=320, bias=False)\n",
" (to_v): Linear(in_features=768, out_features=320, bias=False)\n",
" (to_out): ModuleList(\n",
" (0): Linear(in_features=320, out_features=320, bias=True)\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" (norm3): LayerNorm((320,), eps=1e-05, elementwise_affine=True)\n",
" (ff): FeedForward(\n",
" (net): ModuleList(\n",
" (0): GEGLU(\n",
" (proj): Linear(in_features=320, out_features=2560, bias=True)\n",
" )\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" (2): Linear(in_features=1280, out_features=320, bias=True)\n",
" )\n",
" )\n",
" )\n",
" )\n",
" (proj_out): Conv2d(320, 320, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" )\n",
" (resnets): ModuleList(\n",
" (0-1): 2 x ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 320, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(320, 320, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=320, bias=True)\n",
" (norm2): GroupNorm(32, 320, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(320, 320, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" )\n",
" )\n",
" (downsamplers): ModuleList(\n",
" (0): Downsample2D(\n",
" (conv): Conv2d(320, 320, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))\n",
" )\n",
" )\n",
" )\n",
" (1): CrossAttnDownBlock2D(\n",
" (attentions): ModuleList(\n",
" (0-1): 2 x Transformer2DModel(\n",
" (norm): GroupNorm(32, 640, eps=1e-06, affine=True)\n",
" (proj_in): Conv2d(640, 640, kernel_size=(1, 1), stride=(1, 1))\n",
" (transformer_blocks): ModuleList(\n",
" (0-2): 3 x BasicTransformerBlock(\n",
" (norm1): LayerNorm((640,), eps=1e-05, elementwise_affine=True)\n",
" (attn1): Attention(\n",
" (to_q): Linear(in_features=640, out_features=640, bias=False)\n",
" (to_k): Linear(in_features=640, out_features=640, bias=False)\n",
" (to_v): Linear(in_features=640, out_features=640, bias=False)\n",
" (to_out): ModuleList(\n",
" (0): Linear(in_features=640, out_features=640, bias=True)\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" (norm2): LayerNorm((640,), eps=1e-05, elementwise_affine=True)\n",
" (attn2): Attention(\n",
" (to_q): Linear(in_features=640, out_features=640, bias=False)\n",
" (to_k): Linear(in_features=768, out_features=640, bias=False)\n",
" (to_v): Linear(in_features=768, out_features=640, bias=False)\n",
" (to_out): ModuleList(\n",
" (0): Linear(in_features=640, out_features=640, bias=True)\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" (norm3): LayerNorm((640,), eps=1e-05, elementwise_affine=True)\n",
" (ff): FeedForward(\n",
" (net): ModuleList(\n",
" (0): GEGLU(\n",
" (proj): Linear(in_features=640, out_features=5120, bias=True)\n",
" )\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" (2): Linear(in_features=2560, out_features=640, bias=True)\n",
" )\n",
" )\n",
" )\n",
" )\n",
" (proj_out): Conv2d(640, 640, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" )\n",
" (resnets): ModuleList(\n",
" (0): ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 320, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(320, 640, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=640, bias=True)\n",
" (norm2): GroupNorm(32, 640, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(640, 640, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" (conv_shortcut): Conv2d(320, 640, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" (1): ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 640, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(640, 640, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=640, bias=True)\n",
" (norm2): GroupNorm(32, 640, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(640, 640, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" )\n",
" )\n",
" (downsamplers): ModuleList(\n",
" (0): Downsample2D(\n",
" (conv): Conv2d(640, 640, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))\n",
" )\n",
" )\n",
" )\n",
" (2): CrossAttnDownBlock2D(\n",
" (attentions): ModuleList(\n",
" (0-1): 2 x Transformer2DModel(\n",
" (norm): GroupNorm(32, 1280, eps=1e-06, affine=True)\n",
" (proj_in): Conv2d(1280, 1280, kernel_size=(1, 1), stride=(1, 1))\n",
" (transformer_blocks): ModuleList(\n",
" (0-2): 3 x BasicTransformerBlock(\n",
" (norm1): LayerNorm((1280,), eps=1e-05, elementwise_affine=True)\n",
" (attn1): Attention(\n",
" (to_q): Linear(in_features=1280, out_features=1280, bias=False)\n",
" (to_k): Linear(in_features=1280, out_features=1280, bias=False)\n",
" (to_v): Linear(in_features=1280, out_features=1280, bias=False)\n",
" (to_out): ModuleList(\n",
" (0): Linear(in_features=1280, out_features=1280, bias=True)\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" (norm2): LayerNorm((1280,), eps=1e-05, elementwise_affine=True)\n",
" (attn2): Attention(\n",
" (to_q): Linear(in_features=1280, out_features=1280, bias=False)\n",
" (to_k): Linear(in_features=768, out_features=1280, bias=False)\n",
" (to_v): Linear(in_features=768, out_features=1280, bias=False)\n",
" (to_out): ModuleList(\n",
" (0): Linear(in_features=1280, out_features=1280, bias=True)\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" (norm3): LayerNorm((1280,), eps=1e-05, elementwise_affine=True)\n",
" (ff): FeedForward(\n",
" (net): ModuleList(\n",
" (0): GEGLU(\n",
" (proj): Linear(in_features=1280, out_features=10240, bias=True)\n",
" )\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" (2): Linear(in_features=5120, out_features=1280, bias=True)\n",
" )\n",
" )\n",
" )\n",
" )\n",
" (proj_out): Conv2d(1280, 1280, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" )\n",
" (resnets): ModuleList(\n",
" (0): ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 640, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(640, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=1280, bias=True)\n",
" (norm2): GroupNorm(32, 1280, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" (conv_shortcut): Conv2d(640, 1280, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" (1): ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 1280, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=1280, bias=True)\n",
" (norm2): GroupNorm(32, 1280, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" )\n",
" )\n",
" (downsamplers): ModuleList(\n",
" (0): Downsample2D(\n",
" (conv): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))\n",
" )\n",
" )\n",
" )\n",
" (3): DownBlock2D(\n",
" (resnets): ModuleList(\n",
" (0): ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 1280, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(1280, 1536, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=1536, bias=True)\n",
" (norm2): GroupNorm(32, 1536, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(1536, 1536, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" (conv_shortcut): Conv2d(1280, 1536, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" (1): ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 1536, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(1536, 1536, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=1536, bias=True)\n",
" (norm2): GroupNorm(32, 1536, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(1536, 1536, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" )\n",
" )\n",
" )\n",
" )\n",
" (up_blocks): ModuleList(\n",
" (0): UpBlock2D(\n",
" (resnets): ModuleList(\n",
" (0-1): 2 x ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 3072, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(3072, 1536, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=1536, bias=True)\n",
" (norm2): GroupNorm(32, 1536, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(1536, 1536, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" (conv_shortcut): Conv2d(3072, 1536, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" (2): ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 2816, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(2816, 1536, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=1536, bias=True)\n",
" (norm2): GroupNorm(32, 1536, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(1536, 1536, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" (conv_shortcut): Conv2d(2816, 1536, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" )\n",
" (upsamplers): ModuleList(\n",
" (0): Upsample2D(\n",
" (conv): Conv2d(1536, 1536, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" )\n",
" )\n",
" )\n",
" (1): CrossAttnUpBlock2D(\n",
" (attentions): ModuleList(\n",
" (0-2): 3 x Transformer2DModel(\n",
" (norm): GroupNorm(32, 1280, eps=1e-06, affine=True)\n",
" (proj_in): Conv2d(1280, 1280, kernel_size=(1, 1), stride=(1, 1))\n",
" (transformer_blocks): ModuleList(\n",
" (0-2): 3 x BasicTransformerBlock(\n",
" (norm1): LayerNorm((1280,), eps=1e-05, elementwise_affine=True)\n",
" (attn1): Attention(\n",
" (to_q): Linear(in_features=1280, out_features=1280, bias=False)\n",
" (to_k): Linear(in_features=1280, out_features=1280, bias=False)\n",
" (to_v): Linear(in_features=1280, out_features=1280, bias=False)\n",
" (to_out): ModuleList(\n",
" (0): Linear(in_features=1280, out_features=1280, bias=True)\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" (norm2): LayerNorm((1280,), eps=1e-05, elementwise_affine=True)\n",
" (attn2): Attention(\n",
" (to_q): Linear(in_features=1280, out_features=1280, bias=False)\n",
" (to_k): Linear(in_features=768, out_features=1280, bias=False)\n",
" (to_v): Linear(in_features=768, out_features=1280, bias=False)\n",
" (to_out): ModuleList(\n",
" (0): Linear(in_features=1280, out_features=1280, bias=True)\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" (norm3): LayerNorm((1280,), eps=1e-05, elementwise_affine=True)\n",
" (ff): FeedForward(\n",
" (net): ModuleList(\n",
" (0): GEGLU(\n",
" (proj): Linear(in_features=1280, out_features=10240, bias=True)\n",
" )\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" (2): Linear(in_features=5120, out_features=1280, bias=True)\n",
" )\n",
" )\n",
" )\n",
" )\n",
" (proj_out): Conv2d(1280, 1280, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" )\n",
" (resnets): ModuleList(\n",
" (0): ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 2816, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(2816, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=1280, bias=True)\n",
" (norm2): GroupNorm(32, 1280, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" (conv_shortcut): Conv2d(2816, 1280, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" (1): ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 2560, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(2560, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=1280, bias=True)\n",
" (norm2): GroupNorm(32, 1280, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" (conv_shortcut): Conv2d(2560, 1280, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" (2): ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 1920, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(1920, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=1280, bias=True)\n",
" (norm2): GroupNorm(32, 1280, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" (conv_shortcut): Conv2d(1920, 1280, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" )\n",
" (upsamplers): ModuleList(\n",
" (0): Upsample2D(\n",
" (conv): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" )\n",
" )\n",
" )\n",
" (2): CrossAttnUpBlock2D(\n",
" (attentions): ModuleList(\n",
" (0-2): 3 x Transformer2DModel(\n",
" (norm): GroupNorm(32, 640, eps=1e-06, affine=True)\n",
" (proj_in): Conv2d(640, 640, kernel_size=(1, 1), stride=(1, 1))\n",
" (transformer_blocks): ModuleList(\n",
" (0-2): 3 x BasicTransformerBlock(\n",
" (norm1): LayerNorm((640,), eps=1e-05, elementwise_affine=True)\n",
" (attn1): Attention(\n",
" (to_q): Linear(in_features=640, out_features=640, bias=False)\n",
" (to_k): Linear(in_features=640, out_features=640, bias=False)\n",
" (to_v): Linear(in_features=640, out_features=640, bias=False)\n",
" (to_out): ModuleList(\n",
" (0): Linear(in_features=640, out_features=640, bias=True)\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" (norm2): LayerNorm((640,), eps=1e-05, elementwise_affine=True)\n",
" (attn2): Attention(\n",
" (to_q): Linear(in_features=640, out_features=640, bias=False)\n",
" (to_k): Linear(in_features=768, out_features=640, bias=False)\n",
" (to_v): Linear(in_features=768, out_features=640, bias=False)\n",
" (to_out): ModuleList(\n",
" (0): Linear(in_features=640, out_features=640, bias=True)\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" (norm3): LayerNorm((640,), eps=1e-05, elementwise_affine=True)\n",
" (ff): FeedForward(\n",
" (net): ModuleList(\n",
" (0): GEGLU(\n",
" (proj): Linear(in_features=640, out_features=5120, bias=True)\n",
" )\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" (2): Linear(in_features=2560, out_features=640, bias=True)\n",
" )\n",
" )\n",
" )\n",
" )\n",
" (proj_out): Conv2d(640, 640, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" )\n",
" (resnets): ModuleList(\n",
" (0): ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 1920, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(1920, 640, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=640, bias=True)\n",
" (norm2): GroupNorm(32, 640, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(640, 640, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" (conv_shortcut): Conv2d(1920, 640, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" (1): ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 1280, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(1280, 640, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=640, bias=True)\n",
" (norm2): GroupNorm(32, 640, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(640, 640, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" (conv_shortcut): Conv2d(1280, 640, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" (2): ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 960, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(960, 640, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=640, bias=True)\n",
" (norm2): GroupNorm(32, 640, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(640, 640, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" (conv_shortcut): Conv2d(960, 640, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" )\n",
" (upsamplers): ModuleList(\n",
" (0): Upsample2D(\n",
" (conv): Conv2d(640, 640, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" )\n",
" )\n",
" )\n",
" (3): CrossAttnUpBlock2D(\n",
" (attentions): ModuleList(\n",
" (0-2): 3 x Transformer2DModel(\n",
" (norm): GroupNorm(32, 320, eps=1e-06, affine=True)\n",
" (proj_in): Conv2d(320, 320, kernel_size=(1, 1), stride=(1, 1))\n",
" (transformer_blocks): ModuleList(\n",
" (0): BasicTransformerBlock(\n",
" (norm1): LayerNorm((320,), eps=1e-05, elementwise_affine=True)\n",
" (attn1): Attention(\n",
" (to_q): Linear(in_features=320, out_features=320, bias=False)\n",
" (to_k): Linear(in_features=320, out_features=320, bias=False)\n",
" (to_v): Linear(in_features=320, out_features=320, bias=False)\n",
" (to_out): ModuleList(\n",
" (0): Linear(in_features=320, out_features=320, bias=True)\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" (norm2): LayerNorm((320,), eps=1e-05, elementwise_affine=True)\n",
" (attn2): Attention(\n",
" (to_q): Linear(in_features=320, out_features=320, bias=False)\n",
" (to_k): Linear(in_features=768, out_features=320, bias=False)\n",
" (to_v): Linear(in_features=768, out_features=320, bias=False)\n",
" (to_out): ModuleList(\n",
" (0): Linear(in_features=320, out_features=320, bias=True)\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" (norm3): LayerNorm((320,), eps=1e-05, elementwise_affine=True)\n",
" (ff): FeedForward(\n",
" (net): ModuleList(\n",
" (0): GEGLU(\n",
" (proj): Linear(in_features=320, out_features=2560, bias=True)\n",
" )\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" (2): Linear(in_features=1280, out_features=320, bias=True)\n",
" )\n",
" )\n",
" )\n",
" )\n",
" (proj_out): Conv2d(320, 320, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" )\n",
" (resnets): ModuleList(\n",
" (0): ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 960, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(960, 320, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=320, bias=True)\n",
" (norm2): GroupNorm(32, 320, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(320, 320, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" (conv_shortcut): Conv2d(960, 320, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" (1-2): 2 x ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 640, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(640, 320, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=320, bias=True)\n",
" (norm2): GroupNorm(32, 320, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(320, 320, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" (conv_shortcut): Conv2d(640, 320, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" )\n",
" )\n",
" )\n",
" (mid_block): UNetMidBlock2DCrossAttn(\n",
" (attentions): ModuleList(\n",
" (0): Transformer2DModel(\n",
" (norm): GroupNorm(32, 1536, eps=1e-06, affine=True)\n",
" (proj_in): Conv2d(1536, 1536, kernel_size=(1, 1), stride=(1, 1))\n",
" (transformer_blocks): ModuleList(\n",
" (0-2): 3 x BasicTransformerBlock(\n",
" (norm1): LayerNorm((1536,), eps=1e-05, elementwise_affine=True)\n",
" (attn1): Attention(\n",
" (to_q): Linear(in_features=1536, out_features=1536, bias=False)\n",
" (to_k): Linear(in_features=1536, out_features=1536, bias=False)\n",
" (to_v): Linear(in_features=1536, out_features=1536, bias=False)\n",
" (to_out): ModuleList(\n",
" (0): Linear(in_features=1536, out_features=1536, bias=True)\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" (norm2): LayerNorm((1536,), eps=1e-05, elementwise_affine=True)\n",
" (attn2): Attention(\n",
" (to_q): Linear(in_features=1536, out_features=1536, bias=False)\n",
" (to_k): Linear(in_features=768, out_features=1536, bias=False)\n",
" (to_v): Linear(in_features=768, out_features=1536, bias=False)\n",
" (to_out): ModuleList(\n",
" (0): Linear(in_features=1536, out_features=1536, bias=True)\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" )\n",
" )\n",
" (norm3): LayerNorm((1536,), eps=1e-05, elementwise_affine=True)\n",
" (ff): FeedForward(\n",
" (net): ModuleList(\n",
" (0): GEGLU(\n",
" (proj): Linear(in_features=1536, out_features=12288, bias=True)\n",
" )\n",
" (1): Dropout(p=0.0, inplace=False)\n",
" (2): Linear(in_features=6144, out_features=1536, bias=True)\n",
" )\n",
" )\n",
" )\n",
" )\n",
" (proj_out): Conv2d(1536, 1536, kernel_size=(1, 1), stride=(1, 1))\n",
" )\n",
" )\n",
" (resnets): ModuleList(\n",
" (0-1): 2 x ResnetBlock2D(\n",
" (norm1): GroupNorm(32, 1536, eps=1e-05, affine=True)\n",
" (conv1): Conv2d(1536, 1536, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (time_emb_proj): Linear(in_features=1280, out_features=1536, bias=True)\n",
" (norm2): GroupNorm(32, 1536, eps=1e-05, affine=True)\n",
" (dropout): Dropout(p=0.0, inplace=False)\n",
" (conv2): Conv2d(1536, 1536, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
" (nonlinearity): SiLU()\n",
" )\n",
" )\n",
" )\n",
" (conv_norm_out): GroupNorm(32, 320, eps=1e-05, affine=True)\n",
" (conv_act): SiLU()\n",
" (conv_out): Conv2d(320, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
")\n"
]
}
],
"source": [
"import torch\n",
"from diffusers import UNet2DConditionModel\n",
"\n",
"device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
"dtype = torch.float16 if torch.cuda.is_available() else torch.float32\n",
"\n",
"pipe_id = \"unet\"\n",
"unet = UNet2DConditionModel.from_pretrained(\n",
" pipe_id,\n",
" torch_dtype=dtype,\n",
").to(device)\n",
"\n",
"unet.save_pretrained(pipe_id)\n",
"print(unet)"
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "188279e8-f420-425a-9a29-01d5b60f8383",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"None\n",
"None\n",
"Модель успешно проапгрейжена! Можно дообучать.\n"
]
}
],
"source": [
"import torch\n",
"from diffusers import UNet2DConditionModel\n",
"\n",
"device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
"dtype = torch.float16 if torch.cuda.is_available() else torch.float32\n",
"\n",
"pipe_id = \"unet_simple\"\n",
"old_unet = UNet2DConditionModel.from_pretrained(\n",
" pipe_id,\n",
" torch_dtype=dtype,\n",
").to(device)\n",
"print(old_unet.config.addition_embed_type)\n",
"\n",
"# 2. Создаем новую конфигурацию (добавляем text_time)\n",
"new_config = old_unet.config\n",
"new_config.addition_embed_type = \"text_time\"\n",
"new_config.addition_time_embed_dim = 1024 # Размер вашего пулинга\n",
"\n",
"# 3. Инициализируем новую (пустую) модель\n",
"new_unet = UNet2DConditionModel.from_config(new_config)\n",
"\n",
"# 4. Переносим веса\n",
"new_state_dict = new_unet.state_dict()\n",
"old_state_dict = old_unet.state_dict()\n",
"\n",
"for name, param in old_state_dict.items():\n",
" if \"time_embedding.linear_1.weight\" in name:\n",
" # МАГИЯ ХИРУРГИИ\n",
" # param (старый вес) имеет форму [Out, 256]\n",
" # new_unet...weight имеет форму [Out, 256 + 1024]\n",
" \n",
" # 1. Берем веса новой модели (там сейчас мусор/рандом)\n",
" new_w = new_state_dict[name]\n",
" \n",
" # 2. Зануляем ту часть, которая отвечает за новый пулинг (это важно!)\n",
" # Предположим, что время идет первым, а пулинг вторым (зависит от реализации concat)\n",
" # Обычно concat([time, pool]), значит time занимает первые индексы\n",
" time_dim = param.shape[1] \n",
" \n",
" new_w[:, :time_dim] = param # Копируем старое знание\n",
" new_w[:, time_dim:] = 0 # Новые связи делаем \"неактивными\" на старте\n",
" \n",
" new_state_dict[name] = new_w\n",
" \n",
" elif \"time_embedding.linear_1.bias\" in name:\n",
" # Биас просто копируем, так как выходной размер слоя не меняется\n",
" new_state_dict[name] = param\n",
" \n",
" elif name in new_state_dict:\n",
" # Все остальные слои просто копируем\n",
" if new_state_dict[name].shape == param.shape:\n",
" new_state_dict[name] = param\n",
" else:\n",
" print(f\"Внимание: несовпадение форм для {name}, пропускаем\")\n",
"\n",
"# 5. Загружаем собранный словарь\n",
"new_unet.load_state_dict(new_state_dict)\n",
"print(new_unet.config.addition_embed_type)\n",
"new_unet.save_pretrained(\"unet\")\n",
"print(\"Модель успешно проапгрейжена! Можно дообучать.\")"
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "7c3461ac-8544-4df4-8ab6-e54b239ca4e5",
"metadata": {},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "239e51a54c2343a4ba2f73a6686f725c",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"tokenizer_config.json: 0.00B [00:00, ?B/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "0bcbbdde09184b5c8b7b0b31d5899565",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"vocab.json: 0.00B [00:00, ?B/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "3d4ef94499174412b8a1e1a6b0884a03",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"merges.txt: 0.00B [00:00, ?B/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "c98c9a3b005443209f2839682001028b",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"tokenizer.json: 0%| | 0.00/11.4M [00:00', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={\n",
"\t151643: AddedToken(\"<|endoftext|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n",
"\t151644: AddedToken(\"<|im_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n",
"\t151645: AddedToken(\"<|im_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n",
"\t151646: AddedToken(\"<|object_ref_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n",
"\t151647: AddedToken(\"<|object_ref_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n",
"\t151648: AddedToken(\"<|box_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n",
"\t151649: AddedToken(\"<|box_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n",
"\t151650: AddedToken(\"<|quad_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n",
"\t151651: AddedToken(\"<|quad_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n",
"\t151652: AddedToken(\"<|vision_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n",
"\t151653: AddedToken(\"<|vision_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n",
"\t151654: AddedToken(\"<|vision_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n",
"\t151655: AddedToken(\"<|image_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n",
"\t151656: AddedToken(\"<|video_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n",
"\t151657: AddedToken(\"\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n",
"\t151658: AddedToken(\"\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n",
"\t151659: AddedToken(\"<|fim_prefix|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n",
"\t151660: AddedToken(\"<|fim_middle|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n",
"\t151661: AddedToken(\"<|fim_suffix|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n",
"\t151662: AddedToken(\"<|fim_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n",
"\t151663: AddedToken(\"<|repo_name|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n",
"\t151664: AddedToken(\"<|file_sep|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n",
"\t151665: AddedToken(\"\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n",
"\t151666: AddedToken(\"\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n",
"\t151667: AddedToken(\"\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n",
"\t151668: AddedToken(\"\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n",
"}\n",
")\n",
"saved\n"
]
}
],
"source": [
"from transformers import AutoTokenizer, AutoModel\n",
"import torch\n",
"\n",
"device=\"cuda\"\n",
"dtype = torch.bfloat16 if torch.cuda.is_available() else torch.float32\n",
"\n",
"\n",
"model=\"Qwen/Qwen3-VL-2B-Instruct\"\n",
"tokenizer = AutoTokenizer.from_pretrained(model)\n",
"text_model = AutoModel.from_pretrained(model,torch_dtype=dtype).to(device).eval()\n",
"\n",
"print(text_model)\n",
"print(tokenizer)\n",
"tokenizer.save_pretrained(\"tokenizer\")\n",
"text_model.save_pretrained(\"text_encoder\")\n",
"print('saved')\n"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "638c946a-fd68-4bde-ae87-263ef5ea8679",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"1. Загружаем токенизатор и базовый конфиг...\n"
]
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "85ccbdedb64049719784bdb1c35a5ab7",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"tokenizer_config.json: 0.00B [00:00, ?B/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "cc43a6865576408eb99676f101cc754f",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"vocab.json: 0.00B [00:00, ?B/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "d83333a8404c4c82af5b55ed9500d232",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"merges.txt: 0.00B [00:00, ?B/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "f744e32d288f41bbb243ee90fef66000",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"tokenizer.json: 0.00B [00:00, ?B/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "2a4cf7d322ec45ada41bf2984081ea95",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"config.json: 0.00B [00:00, ?B/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"2. Собираем текстовый конфиг Qwen3...\n",
"3. Загружаем веса VL модели (через базовый AutoModel)...\n"
]
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "a848fd2acf984f1da854b3833ea1429e",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"model.safetensors: 0%| | 0.00/4.26G [00:00