Skip to content

Minor improvements to avoid out of memory issues in stable_diffusion.ipynb notebook (attempt 2) #5

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jph00 merged 2 commits into from
Oct 13, 2022
Merged

Minor improvements to avoid out of memory issues in stable_diffusion.ipynb notebook (attempt 2) #5

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f43862b
Memory efficiency improvements
jantic Oct 13, 2022
2cb2544
Replacing autocast calls with conversions to fp16 on model inputs
jantic Oct 13, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
50 changes: 30 additions & 20 deletions stable_diffusion.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -363,8 +363,8 @@
"name": "stdout",
"output_type": "stream",
"text": [
"models--CompVis--stable-diffusion-v1-4\tmodels--google--ddpm-church-256\r\n",
"models--google--ddpm-celebahq-256\r\n"
"models--CompVis--stable-diffusion-v1-4\tmodels--google--ddpm-church-256\n",
"models--google--ddpm-celebahq-256\n"
]
}
],
Expand Down Expand Up @@ -1323,9 +1323,7 @@
{
"cell_type": "code",
"execution_count": 24,
"metadata": {
"scrolled": false
},
"metadata": {},
"outputs": [
{
"data": {
Expand Down Expand Up @@ -1384,7 +1382,11 @@
"cell_type": "code",
"execution_count": 4,
"metadata": {
"collapsed": true
"collapsed": true,
"jupyter": {
"outputs_hidden": true
},
"tags": []
},
"outputs": [
{
Expand Down Expand Up @@ -1613,8 +1615,8 @@
}
],
"source": [
"db_pipe = StableDiffusionPipeline.from_pretrained(\"pcuenq/jh_dreambooth_1000\", torch_dtype=torch.float16)\n",
"db_pipe = db_pipe.to(\"cuda\")"
"pipe = StableDiffusionPipeline.from_pretrained(\"pcuenq/jh_dreambooth_1000\", torch_dtype=torch.float16)\n",
"pipe = pipe.to(\"cuda\")"
]
},
{
Expand Down Expand Up @@ -1652,7 +1654,7 @@
"torch.manual_seed(1000)\n",
"\n",
"prompt = \"Painting of sks person in the style of Paul Signac\"\n",
"images = db_pipe(prompt, num_images_per_prompt=4).images\n",
"images = pipe(prompt, num_images_per_prompt=4).images\n",
"image_grid(images, 1, 4)"
]
},
Expand Down Expand Up @@ -1806,6 +1808,15 @@
"First, we need the text encoder and the tokenizer. These come from the text portion of a standard CLIP model, so we'll use the weights released by Open AI."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"del pipe"
]
},
{
"cell_type": "code",
"execution_count": 25,
Expand All @@ -1814,8 +1825,8 @@
"source": [
"from transformers import CLIPTextModel, CLIPTokenizer\n",
"\n",
"tokenizer = CLIPTokenizer.from_pretrained(\"openai/clip-vit-large-patch14\")\n",
"text_encoder = CLIPTextModel.from_pretrained(\"openai/clip-vit-large-patch14\")"
"tokenizer = CLIPTokenizer.from_pretrained(\"openai/clip-vit-large-patch14\", torch_dtype=torch.float16)\n",
"text_encoder = CLIPTextModel.from_pretrained(\"openai/clip-vit-large-patch14\", torch_dtype=torch.float16)"
]
},
{
Expand All @@ -1833,8 +1844,8 @@
"source": [
"from diffusers import AutoencoderKL, UNet2DConditionModel\n",
"\n",
"vae = AutoencoderKL.from_pretrained(\"CompVis/stable-diffusion-v1-4\", subfolder=\"vae\")\n",
"unet = UNet2DConditionModel.from_pretrained(\"CompVis/stable-diffusion-v1-4\", subfolder=\"unet\")"
"vae = AutoencoderKL.from_pretrained(\"CompVis/stable-diffusion-v1-4\", subfolder=\"vae\", torch_dtype=torch.float16)\n",
"unet = UNet2DConditionModel.from_pretrained(\"CompVis/stable-diffusion-v1-4\", subfolder=\"unet\", torch_dtype=torch.float16)"
]
},
{
Expand Down Expand Up @@ -2059,7 +2070,7 @@
}
],
"source": [
"text_embeddings = text_encoder(text_input.input_ids.to(\"cuda\"))[0]\n",
"text_embeddings = text_encoder(text_input.input_ids.to(\"cuda\"))[0].half()\n",
"text_embeddings.shape"
]
},
Expand Down Expand Up @@ -2093,7 +2104,7 @@
"uncond_input = tokenizer(\n",
" [\"\"] * batch_size, padding=\"max_length\", max_length=max_length, return_tensors=\"pt\"\n",
")\n",
"uncond_embeddings = text_encoder(uncond_input.input_ids.to(\"cuda\"))[0]\n",
"uncond_embeddings = text_encoder(uncond_input.input_ids.to(\"cuda\"))[0].half()\n",
"uncond_embeddings.shape"
]
},
Expand Down Expand Up @@ -2147,7 +2158,7 @@
"source": [
"torch.manual_seed(100)\n",
"latents = torch.randn((batch_size, unet.in_channels, height // 8, width // 8))\n",
"latents = latents.to(\"cuda\")\n",
"latents = latents.to(\"cuda\").half()\n",
"latents.shape"
]
},
Expand Down Expand Up @@ -2356,8 +2367,7 @@
"height": 529
},
"id": "AAVZStIokTVv",
"outputId": "7af6a1ea-f20a-4445-d756-8bb0dd6a0747",
"scrolled": false
"outputId": "7af6a1ea-f20a-4445-d756-8bb0dd6a0747"
},
"outputs": [
{
Expand Down Expand Up @@ -2405,7 +2415,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.10"
"version": "3.10.6"
},
"toc": {
"base_numbering": 1,
Expand Down Expand Up @@ -16198,5 +16208,5 @@
}
},
"nbformat": 4,
"nbformat_minor": 1
"nbformat_minor": 4
}