initial app test

app.ipynb

@@ -0,0 +1,263 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#| default_exp app"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#| export\n",
+    "import gradio as gr\n",
+    "import cf_guidance\n",
+    "import min_diffusion\n",
+    "import torch\n",
+    "import nbdev"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#| export\n",
+    "\n",
+    "## MODEL SETUP\n",
+    "######################################\n",
+    "######################################\n",
+    "model_name = 'stabilityai/stable-diffusion-2'\n",
+    "revision = 'fp16'\n",
+    "dtype = torch.float16\n",
+    "device = ('cpu','cuda')[torch.cuda.is_available()]\n",
+    "\n",
+    "# model parameters\n",
+    "better_vae = ''\n",
+    "unet_attn_slice = True\n",
+    "sampler_kls = 'dpm_multi'\n",
+    "hf_sampler = 'dpm_multi'\n",
+    "\n",
+    "model_kwargs = {\n",
+    "    'better_vae': better_vae,\n",
+    "    'unet_attn_slice': unet_attn_slice,\n",
+    "    'sampler_kls': hf_sampler,\n",
+    "}\n",
+    "\n",
+    "def load_model():\n",
+    "    pipeline = min_diffusion.core.MinimalDiffusion(\n",
+    "        model_name,\n",
+    "        device,\n",
+    "        dtype,\n",
+    "        revision,\n",
+    "        **model_kwargs,\n",
+    "    )\n",
+    "    pipeline.load()\n",
+    "    return pipeline\n",
+    "######################################\n",
+    "######################################"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#| export \n",
+    "\n",
+    "## GENERATION PARAMETERS\n",
+    "######################################\n",
+    "######################################\n",
+    "num_steps = 18\n",
+    "height, width = 768, 768\n",
+    "k_sampler = 'k_dpmpp_2m' #'k_dpmpp_sde'\n",
+    "use_karras_sigmas = True\n",
+    "\n",
+    "# a good negative prompt\n",
+    "NEG_PROMPT = \"ugly, stock photo, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, mutation, mutated, extra limbs, extra legs, extra arms, disfigured, deformed, cross-eye, body out of frame, blurry, bad art, bad anatomy, blurred, text, watermark, grainy\"\n",
+    "\n",
+    "generation_kwargs = {\n",
+    "    'num_steps': num_steps,\n",
+    "    'height': height,\n",
+    "    'width': width,\n",
+    "    'k_sampler':  k_sampler,\n",
+    "    'negative_prompt': NEG_PROMPT,\n",
+    "    'use_karras_sigmas': use_karras_sigmas,\n",
+    "}\n",
+    "######################################\n",
+    "######################################"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#| export \n",
+    "\n",
+    "## dynamicCFG SETUP\n",
+    "######################################\n",
+    "######################################\n",
+    "\n",
+    "# default cosine schedule parameters\n",
+    "baseline_g        = 9    # default, static guidance value\n",
+    "max_val           = 9    # the max scheduled guidance scaling value\n",
+    "min_val           = 6    # the minimum scheduled guidance value\n",
+    "num_warmup_steps  = 0    # number of warmup steps\n",
+    "warmup_init_val   = 0    # the intial warmup value\n",
+    "num_cycles        = 0.5  # number of cosine cycles\n",
+    "k_decay           = 1    # k-decay for cosine curve scaling \n",
+    "\n",
+    "# group the default schedule parameters\n",
+    "DEFAULT_COS_PARAMS = {\n",
+    "    'max_val':           max_val,\n",
+    "    'num_steps':         num_steps,\n",
+    "    'min_val':           min_val,\n",
+    "    'num_cycles':        num_cycles,\n",
+    "    'k_decay':           k_decay,\n",
+    "    'num_warmup_steps':  num_warmup_steps,\n",
+    "    'warmup_init_val':   warmup_init_val,\n",
+    "}\n",
+    "\n",
+    "def cos_harness(new_params: dict) -> dict:\n",
+    "    '''Creates cosine schedules with updated parameters in `new_params`\n",
+    "    '''\n",
+    "    # start from the given baseline `default_params`\n",
+    "    cos_params = dict(DEFAULT_COS_PARAMS)\n",
+    "    # update the with the new, given parameters\n",
+    "    cos_params.update(new_params)\n",
+    "    \n",
+    "    # return the new cosine schedule\n",
+    "    sched = cf_guidance.schedules.get_cos_sched(**cos_params)\n",
+    "    return sched\n",
+    "\n",
+    "\n",
+    "# build the static schedule\n",
+    "static_sched = [baseline_g] * num_steps\n",
+    "\n",
+    "# build the inverted kdecay schedule\n",
+    "k_sched = cos_harness({'k_decay': 0.2})\n",
+    "inv_k_sched = [max_val - g + min_val for g in k_sched]\n",
+    "\n",
+    "# group the schedules \n",
+    "schedules = {\n",
+    "    'cosine': {'g': inv_k_sched},\n",
+    "    'static': {'g': static_sched},\n",
+    "}\n",
+    "######################################\n",
+    "######################################"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#| export \n",
+    "\n",
+    "def compare_dynamic_guidance(prompt):\n",
+    "    '''\n",
+    "    Compares the default, static Classifier-free Guidance to a dynamic schedule.  \n",
+    "\n",
+    "    Model and sampling paramters:\n",
+    "        Stable Diffusion 2 v-model\n",
+    "        Half-precision\n",
+    "        DPM++ 2M sampler, with Karras sigma schedule\n",
+    "        18 sampling steps\n",
+    "        (768 x 768) image\n",
+    "        Using a generic negative prompt\n",
+    "\n",
+    "    Schedules:\n",
+    "        Static guidance with scale of 9\n",
+    "        Inverse kDecay (cosine variant) scheduled guidance\n",
+    "    '''\n",
+    "    # load the model\n",
+    "    pipeline = load_model()\n",
+    "\n",
+    "    # stores the output images\n",
+    "    res = []\n",
+    "\n",
+    "    # generate images with static and dynamic schedules\n",
+    "    for (name,sched) in schedules.items():\n",
+    "        # make the guidance norm\n",
+    "        gtfm = cf_guidance.transforms.GuidanceTfm(sched)\n",
+    "        # generate the image\n",
+    "        with torch.autocast(device), torch.no_grad():\n",
+    "            img = pipeline.generate(prompt, gtfm, **generation_kwargs)\n",
+    "        # add the generated image\n",
+    "        res.append(name)\n",
+    "\n",
+    "    # return the generated images\n",
+    "    return {\n",
+    "        'values': res,\n",
+    "        'label': 'Cosine vs. Static CFG'\n",
+    "    }"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "\n",
+    "#| export\n",
+    "\n",
+    "iface = gr.Interface(\n",
+    "    compare_dynamic_guidance,\n",
+    "    inputs=\"text\",\n",
+    "    outputs=gr.Gallery(),\n",
+    "    title=\"Comparison with dynamic Classifier-free Guidance Comparison\",\n",
+    ")\n",
+    "iface.launch()\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import nbdev\n",
+    "nbdev.export.nb_export('app.ipynb', '')"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "sdiffkernel",
+   "language": "python",
+   "name": "sdiffkernel"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.8"
+  },
+  "orig_nbformat": 4,
+  "vscode": {
+   "interpreter": {
+    "hash": "7aa72ffd68a1153f913726b8656445c52d825f656451987cb25ebe84c64ea44d"
+   }
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

app.py

@@ -1,7 +1,174 @@
+# AUTOGENERATED! DO NOT EDIT! File to edit: app.ipynb.
+
+# %% auto 0
+__all__ = ['model_name', 'revision', 'dtype', 'device', 'better_vae', 'unet_attn_slice', 'sampler_kls', 'hf_sampler',
+           'model_kwargs', 'num_steps', 'height', 'width', 'k_sampler', 'use_karras_sigmas', 'NEG_PROMPT',
+           'generation_kwargs', 'baseline_g', 'max_val', 'min_val', 'num_warmup_steps', 'warmup_init_val', 'num_cycles',
+           'k_decay', 'DEFAULT_COS_PARAMS', 'static_sched', 'k_sched', 'inv_k_sched', 'schedules', 'iface',
+           'load_model', 'cos_harness', 'compare_dynamic_guidance']
+
+# %% app.ipynb 1
 import gradio as gr
+import cf_guidance
+import min_diffusion
+import torch
+import nbdev
+
+# %% app.ipynb 2
+## MODEL SETUP
+######################################
+######################################
+model_name = 'stabilityai/stable-diffusion-2'
+revision = 'fp16'
+dtype = torch.float16
+device = ('cpu','cuda')[torch.cuda.is_available()]
+
+# model parameters
+better_vae = ''
+unet_attn_slice = True
+sampler_kls = 'dpm_multi'
+hf_sampler = 'dpm_multi'
+
+model_kwargs = {
+    'better_vae': better_vae,
+    'unet_attn_slice': unet_attn_slice,
+    'sampler_kls': hf_sampler,
+}
+
+def load_model():
+    pipeline = min_diffusion.core.MinimalDiffusion(
+        model_name,
+        device,
+        dtype,
+        revision,
+        **model_kwargs,
+    )
+    pipeline.load()
+    return pipeline
+######################################
+######################################
+
+# %% app.ipynb 3
+## GENERATION PARAMETERS
+######################################
+######################################
+num_steps = 18
+height, width = 768, 768
+k_sampler = 'k_dpmpp_2m' #'k_dpmpp_sde'
+use_karras_sigmas = True
+
+# a good negative prompt
+NEG_PROMPT = "ugly, stock photo, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, mutation, mutated, extra limbs, extra legs, extra arms, disfigured, deformed, cross-eye, body out of frame, blurry, bad art, bad anatomy, blurred, text, watermark, grainy"
+
+generation_kwargs = {
+    'num_steps': num_steps,
+    'height': height,
+    'width': width,
+    'k_sampler':  k_sampler,
+    'negative_prompt': NEG_PROMPT,
+    'use_karras_sigmas': use_karras_sigmas,
+}
+######################################
+######################################
+
+# %% app.ipynb 4
+## dynamicCFG SETUP
+######################################
+######################################
+
+# default cosine schedule parameters
+baseline_g        = 9    # default, static guidance value
+max_val           = 9    # the max scheduled guidance scaling value
+min_val           = 6    # the minimum scheduled guidance value
+num_warmup_steps  = 0    # number of warmup steps
+warmup_init_val   = 0    # the intial warmup value
+num_cycles        = 0.5  # number of cosine cycles
+k_decay           = 1    # k-decay for cosine curve scaling 
+
+# group the default schedule parameters
+DEFAULT_COS_PARAMS = {
+    'max_val':           max_val,
+    'num_steps':         num_steps,
+    'min_val':           min_val,
+    'num_cycles':        num_cycles,
+    'k_decay':           k_decay,
+    'num_warmup_steps':  num_warmup_steps,
+    'warmup_init_val':   warmup_init_val,
+}
 
-def greet(name):
-    return "Hello " + name + "!!"
+def cos_harness(new_params: dict) -> dict:
+    '''Creates cosine schedules with updated parameters in `new_params`
+    '''
+    # start from the given baseline `default_params`
+    cos_params = dict(DEFAULT_COS_PARAMS)
+    # update the with the new, given parameters
+    cos_params.update(new_params)
+    
+    # return the new cosine schedule
+    sched = cf_guidance.schedules.get_cos_sched(**cos_params)
+    return sched
 
-iface = gr.Interface(fn=greet, inputs="text", outputs="text")
+
+# build the static schedule
+static_sched = [baseline_g] * num_steps
+
+# build the inverted kdecay schedule
+k_sched = cos_harness({'k_decay': 0.2})
+inv_k_sched = [max_val - g + min_val for g in k_sched]
+
+# group the schedules 
+schedules = {
+    'cosine': {'g': inv_k_sched},
+    'static': {'g': static_sched},
+}
+######################################
+######################################
+
+# %% app.ipynb 5
+def compare_dynamic_guidance(prompt):
+    '''
+    Compares the default, static Classifier-free Guidance to a dynamic schedule.  
+
+    Model and sampling paramters:
+        Stable Diffusion 2 v-model
+        Half-precision
+        DPM++ 2M sampler, with Karras sigma schedule
+        18 sampling steps
+        (768 x 768) image
+        Using a generic negative prompt
+
+    Schedules:
+        Static guidance with scale of 9
+        Inverse kDecay (cosine variant) scheduled guidance
+    '''
+    # load the model
+    pipeline = load_model()
+
+    # stores the output images
+    res = []
+
+    # generate images with static and dynamic schedules
+    for (name,sched) in schedules.items():
+        # make the guidance norm
+        gtfm = cf_guidance.transforms.GuidanceTfm(sched)
+        # generate the image
+        with torch.autocast(device), torch.no_grad():
+            img = pipeline.generate(prompt, gtfm, **generation_kwargs)
+        # add the generated image
+        res.append(name)
+
+    # return the generated images
+    return {
+        'values': res,
+        'label': 'Cosine vs. Static CFG'
+    }
+
+# %% app.ipynb 6
+iface = gr.Interface(
+    compare_dynamic_guidance,
+    inputs="text",
+    outputs=gr.Gallery(),
+    title="Comparison with dynamic Classifier-free Guidance Comparison",
+)
 iface.launch()
+

requirements.txt

@@ -0,0 +1,6 @@
+min_diffusion
+cf_guidance
+torch
+transformers
+diffusers
+nbdev