osmarks/meme-search-engine
1
0
Fork 0
mirror of https://github.com/osmarks/meme-search-engine.git synced 2024-12-21 23:50:33 +00:00
Code Issues Releases Wiki Activity

initial commit

Browse Source
This commit is contained in:
osmarks 2023-09-28 17:30:20 +01:00
commit 9d89e6e4f5
19 changed files with 1318 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.gitignore vendored Normal file
View File

@ -0,0 +1,2 @@
*/node_modules
*/static/app.*

7
LICENSE Normal file
View File

@ -0,0 +1,7 @@
Copyright © 2023 osmarks
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

44
README.md Normal file
View File

@ -0,0 +1,44 @@
# Meme Search Engine
Do you have a large folder of memes you want to search semantically? Do you have a Linux server with an Nvidia GPU? You do; this is now mandatory.
## Features
They say a picture is worth a thousand words. Unfortunately, not all (most?) sets of words can be adequately described by pictures. Regardless, here is a picture. You can use a running instance [here](https://mse.osmarks.net/).
![Meme Search Engine's frontend.](/demo-image.png)
* Infinite-scroll masonry UI for dense meme viewing.
* Online reindexing (a good reason to use it over [clip-retrieval](https://github.com/rom1504/clip-retrieval)) - reload memes without a slow expensive rebuild step.
* Complex query support - query using text and images, including multiple terms at once, with weighting (including negative).
* Reasonably fast.
## Setup
* Serve your meme library from a static webserver.
* I use nginx. If you're in a hurry, you can use `python -m http.server`.
* Install Python dependencies with `pip` from `requirements.txt` (the versions probably shouldn't need to match exactly if you need to change them; I just put in what I currently have installed).
* Run `clip_server.py` (as a background service).
* It is configured with a JSON file given to it as its first argument. An example is in `clip_server_config.json`.
* `device` should probably be `cuda` or `cpu`. The model will run on here.
* `model` is the [OpenCLIP](https://github.com/mlfoundations/open_clip) model to use.
* `model_name` is the name of the model for metrics purposes.
* `max_batch_size` controls the maximum allowed batch size. Higher values generally result in somewhat better performance (the bottleneck in most cases is elsewhere right now though) at the cost of higher VRAM use.
* `port` is the port to run the HTTP server on.
* Run `mse.py` (also as a background service)
* This needs to be exposed somewhere the frontend can reach it. Configure your reverse proxy appropriately.
* It has a JSON config file as well.
* `clip_server` is the full URL for the backend server.
* `db_path` is the path for the SQLite database of images and embedding vectors.
* `files` is where meme files will be read from. Subdirectories are indexed.
* `port` is the port to serve HTTP on.
* Build clipfront2, host on your favourite static webserver.
* `npm install`, `node src/build.js`.
* You will need to rebuild it whenever you edit `frontend_config.json`.
* `image_path` is the base URL of your meme webserver (with trailing slash).
* `backend_url` is the URL `mse.py` is exposed on (trailing slash probably optional).
* If you want, configure Prometheus to monitor `mse.py` and `clip_server.py`.
## Scaling
Meme Search Engine uses an in-memory FAISS index to hold its embedding vectors, because I was lazy and it works fine (~100MB total RAM used for my 8000 memes). If you want to store significantly more than that you will have to switch to a more efficient/compact index (see [here](https://github.com/facebookresearch/faiss/wiki/Guidelines-to-choose-an-index)). As vector indices are held exclusively in memory, you will need to either persist them to disk or use ones which are fast to build/remove from/add to (presumably PCA/PQ indices). At some point if you increase total traffic the CLIP model may also become a bottleneck, as I also have no batching strategy. Indexing appears to actually be CPU-bound (specifically, it's limited by single-threaded image decoding and serialization) - improving that would require a lot of redesigns so I haven't. You may also want to scale down displayed memes to cut bandwidth needs.

142
clip_server.py Normal file
View File

@ -0,0 +1,142 @@
import torch
import time
import threading
from aiohttp import web
import aiohttp
import asyncio
import traceback
import umsgpack
import collections
import queue
import open_clip
from PIL import Image
from prometheus_client import Counter, Histogram, REGISTRY, generate_latest
import io
import json
import sys
with open(sys.argv[1], "r") as config_file:
CONFIG = json.load(config_file)
device = torch.device(CONFIG["device"])
model, _, preprocess = open_clip.create_model_and_transforms(CONFIG["model"], device=device, pretrained=dict(open_clip.list_pretrained())[CONFIG["model"]], precision="fp16")
model.eval()
tokenizer = open_clip.get_tokenizer(CONFIG["model"])
print("Model loaded")
BS = CONFIG["max_batch_size"]
MODELNAME = CONFIG["model_name"]
InferenceParameters = collections.namedtuple("InferenceParameters", ["text", "images", "callback"])
items_ctr = Counter("modelserver_total_items", "Items run through model server", ["model", "modality"])
inference_time_hist = Histogram("modelserver_inftime", "Time running inference", ["model", "batch_size"])
batch_count_ctr = Counter("modelserver_batchcount", "Inference batches run", ["model"])
torch.set_grad_enabled(False)
def do_inference(params: InferenceParameters):
with torch.no_grad():
try:
text, images, callback = params
if text is not None:
items_ctr.labels(MODELNAME, "text").inc(text.shape[0])
with inference_time_hist.labels(MODELNAME + "-text", text.shape[0]).time():
features = model.encode_text(text)
elif images is not None:
with inference_time_hist.labels(MODELNAME + "-image", images.shape[0]).time():
items_ctr.labels(MODELNAME, "image").inc(images.shape[0])
features = model.encode_image(images)
batch_count_ctr.labels(MODELNAME).inc()
features /= features.norm(dim=-1, keepdim=True)
callback(True, features.cpu().numpy())
except Exception as e:
traceback.print_exc()
callback(False, str(e))
finally:
torch.cuda.empty_cache()
iq = queue.Queue(10)
def infer_thread():
while True:
do_inference(iq.get())
pq = queue.Queue(10)
def preprocessing_thread():
while True:
text, images, callback = pq.get()
try:
if text:
assert len(text) <= BS, f"max batch size is {BS}"
text = tokenizer(text).to(device)
elif images:
assert len(images) <= BS, f"max batch size is {BS}"
images = torch.stack([ preprocess(Image.open(io.BytesIO(im))).half() for im in images ]).to(device)
else:
assert False, "images or text required"
iq.put(InferenceParameters(text, images, callback))
except Exception as e:
traceback.print_exc()
callback(False, str(e))
app = web.Application(client_max_size=2**26)
routes = web.RouteTableDef()
@routes.post("/")
async def run_inference(request):
loop = asyncio.get_event_loop()
data = umsgpack.loads(await request.read())
event = asyncio.Event()
results = None
def callback(*argv):
loop.call_soon_threadsafe(lambda: event.set())
nonlocal results
results = argv
pq.put_nowait(InferenceParameters(data.get("text"), data.get("images"), callback))
await event.wait()
body_data = results[1]
if results[0]:
status = 200
body_data = [x.astype("float16").tobytes() for x in body_data]
else:
status = 500
print(results[1])
return web.Response(body=umsgpack.dumps(body_data), status=status, content_type="application/msgpack")
@routes.get("/config")
async def config(request):
return web.Response(body=umsgpack.dumps({
"model": CONFIG["model"],
"batch": BS,
"image_size": model.visual.image_size,
"embedding_size": model.visual.output_dim
}), status=200, content_type="application/msgpack")
@routes.get("/")
async def health(request):
return web.Response(status=204)
@routes.get("/metrics")
async def metrics(request):
return web.Response(body=generate_latest(REGISTRY))
app.router.add_routes(routes)
async def run_webserver():
runner = web.AppRunner(app)
await runner.setup()
site = web.TCPSite(runner, "", CONFIG["port"])
print("Ready")
await site.start()
try:
th = threading.Thread(target=infer_thread)
th.start()
th = threading.Thread(target=preprocessing_thread)
th.start()
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
loop.run_until_complete(run_webserver())
loop.run_forever()
except KeyboardInterrupt:
import sys
sys.exit(0)

7
clip_server_config.json Normal file
View File

@ -0,0 +1,7 @@
{
"device": "cuda:0",
"model": "ViT-H-14",
"model_name": "openclip-ViT-H-14",
"max_batch_size": 128,
"port": 1708
}

463
clipfront2/package-lock.json generated Normal file
View File

@ -0,0 +1,463 @@
{
"name": "clipfront2",
"lockfileVersion": 2,
"requires": true,
"packages": {
"": {
"devDependencies": {
"esbuild": "^0.12.15",
"esbuild-svelte": "^0.5.3",
"sass": "^1.68.0",
"svelte-preprocess-sass": "^2.0.1"
}
},
"node_modules/anymatch": {
"version": "3.1.2",
"resolved": "https://registry.npmjs.org/anymatch/-/anymatch-3.1.2.tgz",
"integrity": "sha512-P43ePfOAIupkguHUycrc4qJ9kz8ZiuOUijaETwX7THt0Y/GNK7v0aa8rY816xWjZ7rJdA5XdMcpVFTKMq+RvWg==",
"dev": true,
"dependencies": {
"normalize-path": "^3.0.0",
"picomatch": "^2.0.4"
},
"engines": {
"node": ">= 8"
}
},
"node_modules/binary-extensions": {
"version": "2.2.0",
"resolved": "https://registry.npmjs.org/binary-extensions/-/binary-extensions-2.2.0.tgz",
"integrity": "sha512-jDctJ/IVQbZoJykoeHbhXpOlNBqGNcwXJKJog42E5HDPUwQTSdjCHdihjj0DlnheQ7blbT6dHOafNAiS8ooQKA==",
"dev": true,
"engines": {
"node": ">=8"
}
},
"node_modules/braces": {
"version": "3.0.2",
"resolved": "https://registry.npmjs.org/braces/-/braces-3.0.2.tgz",
"integrity": "sha512-b8um+L1RzM3WDSzvhm6gIz1yfTbBt6YTlcEKAvsmqCZZFw46z626lVj9j1yEPW33H5H+lBQpZMP1k8l+78Ha0A==",
"dev": true,
"dependencies": {
"fill-range": "^7.0.1"
},
"engines": {
"node": ">=8"
}
},
"node_modules/chokidar": {
"version": "3.5.2",
"resolved": "https://registry.npmjs.org/chokidar/-/chokidar-3.5.2.tgz",
"integrity": "sha512-ekGhOnNVPgT77r4K/U3GDhu+FQ2S8TnK/s2KbIGXi0SZWuwkZ2QNyfWdZW+TVfn84DpEP7rLeCt2UI6bJ8GwbQ==",
"dev": true,
"dependencies": {
"anymatch": "~3.1.2",
"braces": "~3.0.2",
"glob-parent": "~5.1.2",
"is-binary-path": "~2.1.0",
"is-glob": "~4.0.1",
"normalize-path": "~3.0.0",
"readdirp": "~3.6.0"
},
"engines": {
"node": ">= 8.10.0"
},
"optionalDependencies": {
"fsevents": "~2.3.2"
}
},
"node_modules/esbuild": {
"version": "0.12.15",
"resolved": "https://registry.npmjs.org/esbuild/-/esbuild-0.12.15.tgz",
"integrity": "sha512-72V4JNd2+48eOVCXx49xoSWHgC3/cCy96e7mbXKY+WOWghN00cCmlGnwVLRhRHorvv0dgCyuMYBZlM2xDM5OQw==",
"dev": true,
"hasInstallScript": true,
"bin": {
"esbuild": "bin/esbuild"
}
},
"node_modules/esbuild-svelte": {
"version": "0.5.3",
"resolved": "https://registry.npmjs.org/esbuild-svelte/-/esbuild-svelte-0.5.3.tgz",
"integrity": "sha512-KByKD/yt8QaqKjLu32MG3MXBExJYlDM0QwzW3pzKLJR4eev0923DrUKRHPBBjB+OVirUtZnEJE/qitjdW/WyAw==",
"dev": true,
"dependencies": {
"svelte": "^3.38.3"
},
"peerDependencies": {
"esbuild": ">=0.9.6"
}
},
"node_modules/fill-range": {
"version": "7.0.1",
"resolved": "https://registry.npmjs.org/fill-range/-/fill-range-7.0.1.tgz",
"integrity": "sha512-qOo9F+dMUmC2Lcb4BbVvnKJxTPjCm+RRpe4gDuGrzkL7mEVl/djYSu2OdQ2Pa302N4oqkSg9ir6jaLWJ2USVpQ==",
"dev": true,
"dependencies": {
"to-regex-range": "^5.0.1"
},
"engines": {
"node": ">=8"
}
},
"node_modules/fsevents": {
"version": "2.3.2",
"resolved": "https://registry.npmjs.org/fsevents/-/fsevents-2.3.2.tgz",
"integrity": "sha512-xiqMQR4xAeHTuB9uWm+fFRcIOgKBMiOBP+eXiyT7jsgVCq1bkVygt00oASowB7EdtpOHaaPgKt812P9ab+DDKA==",
"dev": true,
"hasInstallScript": true,
"optional": true,
"os": [
"darwin"
],
"engines": {
"node": "^8.16.0 || ^10.6.0 || >=11.0.0"
}
},
"node_modules/glob-parent": {
"version": "5.1.2",
"resolved": "https://registry.npmjs.org/glob-parent/-/glob-parent-5.1.2.tgz",
"integrity": "sha512-AOIgSQCepiJYwP3ARnGx+5VnTu2HBYdzbGP45eLw1vr3zB3vZLeyed1sC9hnbcOc9/SrMyM5RPQrkGz4aS9Zow==",
"dev": true,
"dependencies": {
"is-glob": "^4.0.1"
},
"engines": {
"node": ">= 6"
}
},
"node_modules/immutable": {
"version": "4.0.0",
"resolved": "https://registry.npmjs.org/immutable/-/immutable-4.0.0.tgz",
"integrity": "sha512-zIE9hX70qew5qTUjSS7wi1iwj/l7+m54KWU247nhM3v806UdGj1yDndXj+IOYxxtW9zyLI+xqFNZjTuDaLUqFw==",
"dev": true
},
"node_modules/is-binary-path": {
"version": "2.1.0",
"resolved": "https://registry.npmjs.org/is-binary-path/-/is-binary-path-2.1.0.tgz",
"integrity": "sha512-ZMERYes6pDydyuGidse7OsHxtbI7WVeUEozgR/g7rd0xUimYNlvZRE/K2MgZTjWy725IfelLeVcEM97mmtRGXw==",
"dev": true,
"dependencies": {
"binary-extensions": "^2.0.0"
},
"engines": {
"node": ">=8"
}
},
"node_modules/is-extglob": {
"version": "2.1.1",
"resolved": "https://registry.npmjs.org/is-extglob/-/is-extglob-2.1.1.tgz",
"integrity": "sha1-qIwCU1eR8C7TfHahueqXc8gz+MI=",
"dev": true,
"engines": {
"node": ">=0.10.0"
}
},
"node_modules/is-glob": {
"version": "4.0.1",
"resolved": "https://registry.npmjs.org/is-glob/-/is-glob-4.0.1.tgz",
"integrity": "sha512-5G0tKtBTFImOqDnLB2hG6Bp2qcKEFduo4tZu9MT/H6NQv/ghhy30o55ufafxJ/LdH79LLs2Kfrn85TLKyA7BUg==",
"dev": true,
"dependencies": {
"is-extglob": "^2.1.1"
},
"engines": {
"node": ">=0.10.0"
}
},
"node_modules/is-number": {
"version": "7.0.0",
"resolved": "https://registry.npmjs.org/is-number/-/is-number-7.0.0.tgz",
"integrity": "sha512-41Cifkg6e8TylSpdtTpeLVMqvSBEVzTttHvERD741+pnZ8ANv0004MRL43QKPDlK9cGvNp6NZWZUBlbGXYxxng==",
"dev": true,
"engines": {
"node": ">=0.12.0"
}
},
"node_modules/normalize-path": {
"version": "3.0.0",
"resolved": "https://registry.npmjs.org/normalize-path/-/normalize-path-3.0.0.tgz",
"integrity": "sha512-6eZs5Ls3WtCisHWp9S2GUy8dqkpGi4BVSz3GaqiE6ezub0512ESztXUwUB6C6IKbQkY2Pnb/mD4WYojCRwcwLA==",
"dev": true,
"engines": {
"node": ">=0.10.0"
}
},
"node_modules/picomatch": {
"version": "2.3.0",
"resolved": "https://registry.npmjs.org/picomatch/-/picomatch-2.3.0.tgz",
"integrity": "sha512-lY1Q/PiJGC2zOv/z391WOTD+Z02bCgsFfvxoXXf6h7kv9o+WmsmzYqrAwY63sNgOxE4xEdq0WyUnXfKeBrSvYw==",
"dev": true,
"engines": {
"node": ">=8.6"
},
"funding": {
"url": "https://github.com/sponsors/jonschlinkert"
}
},
"node_modules/readdirp": {
"version": "3.6.0",
"resolved": "https://registry.npmjs.org/readdirp/-/readdirp-3.6.0.tgz",
"integrity": "sha512-hOS089on8RduqdbhvQ5Z37A0ESjsqz6qnRcffsMU3495FuTdqSm+7bhJ29JvIOsBDEEnan5DPu9t3To9VRlMzA==",
"dev": true,
"dependencies": {
"picomatch": "^2.2.1"
},
"engines": {
"node": ">=8.10.0"
}
},
"node_modules/sass": {
"version": "1.68.0",
"resolved": "https://registry.npmjs.org/sass/-/sass-1.68.0.tgz",
"integrity": "sha512-Lmj9lM/fef0nQswm1J2HJcEsBUba4wgNx2fea6yJHODREoMFnwRpZydBnX/RjyXw2REIwdkbqE4hrTo4qfDBUA==",
"dev": true,
"dependencies": {
"chokidar": ">=3.0.0 <4.0.0",
"immutable": "^4.0.0",
"source-map-js": ">=0.6.2 <2.0.0"
},
"bin": {
"sass": "sass.js"
},
"engines": {
"node": ">=14.0.0"
}
},
"node_modules/source-map-js": {
"version": "1.0.2",
"resolved": "https://registry.npmjs.org/source-map-js/-/source-map-js-1.0.2.tgz",
"integrity": "sha512-R0XvVJ9WusLiqTCEiGCmICCMplcCkIwwR11mOSD9CR5u+IXYdiseeEuXCVAjS54zqwkLcPNnmU4OeJ6tUrWhDw==",
"dev": true,
"engines": {
"node": ">=0.10.0"
}
},
"node_modules/svelte": {
"version": "3.38.3",
"resolved": "https://registry.npmjs.org/svelte/-/svelte-3.38.3.tgz",
"integrity": "sha512-N7bBZJH0iF24wsalFZF+fVYMUOigaAUQMIcEKHO3jstK/iL8VmP9xE+P0/a76+FkNcWt+TDv2Gx1taUoUscrvw==",
"dev": true,
"engines": {
"node": ">= 8"
}
},
"node_modules/svelte-preprocess-filter": {
"version": "1.0.0",
"resolved": "https://registry.npmjs.org/svelte-preprocess-filter/-/svelte-preprocess-filter-1.0.0.tgz",
"integrity": "sha512-92innv59nyEx24xbfcSurB5ocwC8qFdDtGli/JVMHzJsxyvV2yjQKIcbUqU9VIV5mKUWO2PoY93nncS2yF4ULQ==",
"dev": true
},
"node_modules/svelte-preprocess-sass": {
"version": "2.0.1",
"resolved": "https://registry.npmjs.org/svelte-preprocess-sass/-/svelte-preprocess-sass-2.0.1.tgz",
"integrity": "sha512-0y4FjRsRWcN7rJeNJnSfZ7LVAz6S7/j9Dg24XFRelr/rjMMjXORdEvXy4r38fUYmyk9Y7yjwlHCiqyGxMHhEbg==",
"dev": true,
"dependencies": {
"svelte-preprocess-filter": "^1.0.0"
},
"peerDependencies": {
"sass": "^1.35.2"
}
},
"node_modules/to-regex-range": {
"version": "5.0.1",
"resolved": "https://registry.npmjs.org/to-regex-range/-/to-regex-range-5.0.1.tgz",
"integrity": "sha512-65P7iz6X5yEr1cwcgvQxbbIw7Uk3gOy5dIdtZ4rDveLqhrdJP+Li/Hx6tyK0NEb+2GCyneCMJiGqrADCSNk8sQ==",
"dev": true,
"dependencies": {
"is-number": "^7.0.0"
},
"engines": {
"node": ">=8.0"
}
}
},
"dependencies": {
"anymatch": {
"version": "3.1.2",
"resolved": "https://registry.npmjs.org/anymatch/-/anymatch-3.1.2.tgz",
"integrity": "sha512-P43ePfOAIupkguHUycrc4qJ9kz8ZiuOUijaETwX7THt0Y/GNK7v0aa8rY816xWjZ7rJdA5XdMcpVFTKMq+RvWg==",
"dev": true,
"requires": {
"normalize-path": "^3.0.0",
"picomatch": "^2.0.4"
}
},
"binary-extensions": {
"version": "2.2.0",
"resolved": "https://registry.npmjs.org/binary-extensions/-/binary-extensions-2.2.0.tgz",
"integrity": "sha512-jDctJ/IVQbZoJykoeHbhXpOlNBqGNcwXJKJog42E5HDPUwQTSdjCHdihjj0DlnheQ7blbT6dHOafNAiS8ooQKA==",
"dev": true
},
"braces": {
"version": "3.0.2",
"resolved": "https://registry.npmjs.org/braces/-/braces-3.0.2.tgz",
"integrity": "sha512-b8um+L1RzM3WDSzvhm6gIz1yfTbBt6YTlcEKAvsmqCZZFw46z626lVj9j1yEPW33H5H+lBQpZMP1k8l+78Ha0A==",
"dev": true,
"requires": {
"fill-range": "^7.0.1"
}
},
"chokidar": {
"version": "3.5.2",
"resolved": "https://registry.npmjs.org/chokidar/-/chokidar-3.5.2.tgz",
"integrity": "sha512-ekGhOnNVPgT77r4K/U3GDhu+FQ2S8TnK/s2KbIGXi0SZWuwkZ2QNyfWdZW+TVfn84DpEP7rLeCt2UI6bJ8GwbQ==",
"dev": true,
"requires": {
"anymatch": "~3.1.2",
"braces": "~3.0.2",
"fsevents": "~2.3.2",
"glob-parent": "~5.1.2",
"is-binary-path": "~2.1.0",
"is-glob": "~4.0.1",
"normalize-path": "~3.0.0",
"readdirp": "~3.6.0"
}
},
"esbuild": {
"version": "0.12.15",
"resolved": "https://registry.npmjs.org/esbuild/-/esbuild-0.12.15.tgz",
"integrity": "sha512-72V4JNd2+48eOVCXx49xoSWHgC3/cCy96e7mbXKY+WOWghN00cCmlGnwVLRhRHorvv0dgCyuMYBZlM2xDM5OQw==",
"dev": true
},
"esbuild-svelte": {
"version": "0.5.3",
"resolved": "https://registry.npmjs.org/esbuild-svelte/-/esbuild-svelte-0.5.3.tgz",
"integrity": "sha512-KByKD/yt8QaqKjLu32MG3MXBExJYlDM0QwzW3pzKLJR4eev0923DrUKRHPBBjB+OVirUtZnEJE/qitjdW/WyAw==",
"dev": true,
"requires": {
"svelte": "^3.38.3"
}
},
"fill-range": {
"version": "7.0.1",
"resolved": "https://registry.npmjs.org/fill-range/-/fill-range-7.0.1.tgz",
"integrity": "sha512-qOo9F+dMUmC2Lcb4BbVvnKJxTPjCm+RRpe4gDuGrzkL7mEVl/djYSu2OdQ2Pa302N4oqkSg9ir6jaLWJ2USVpQ==",
"dev": true,
"requires": {
"to-regex-range": "^5.0.1"
}
},
"fsevents": {
"version": "2.3.2",
"resolved": "https://registry.npmjs.org/fsevents/-/fsevents-2.3.2.tgz",
"integrity": "sha512-xiqMQR4xAeHTuB9uWm+fFRcIOgKBMiOBP+eXiyT7jsgVCq1bkVygt00oASowB7EdtpOHaaPgKt812P9ab+DDKA==",
"dev": true,
"optional": true
},
"glob-parent": {
"version": "5.1.2",
"resolved": "https://registry.npmjs.org/glob-parent/-/glob-parent-5.1.2.tgz",
"integrity": "sha512-AOIgSQCepiJYwP3ARnGx+5VnTu2HBYdzbGP45eLw1vr3zB3vZLeyed1sC9hnbcOc9/SrMyM5RPQrkGz4aS9Zow==",
"dev": true,
"requires": {
"is-glob": "^4.0.1"
}
},
"immutable": {
"version": "4.0.0",
"resolved": "https://registry.npmjs.org/immutable/-/immutable-4.0.0.tgz",
"integrity": "sha512-zIE9hX70qew5qTUjSS7wi1iwj/l7+m54KWU247nhM3v806UdGj1yDndXj+IOYxxtW9zyLI+xqFNZjTuDaLUqFw==",
"dev": true
},
"is-binary-path": {
"version": "2.1.0",
"resolved": "https://registry.npmjs.org/is-binary-path/-/is-binary-path-2.1.0.tgz",
"integrity": "sha512-ZMERYes6pDydyuGidse7OsHxtbI7WVeUEozgR/g7rd0xUimYNlvZRE/K2MgZTjWy725IfelLeVcEM97mmtRGXw==",
"dev": true,
"requires": {
"binary-extensions": "^2.0.0"
}
},
"is-extglob": {
"version": "2.1.1",
"resolved": "https://registry.npmjs.org/is-extglob/-/is-extglob-2.1.1.tgz",
"integrity": "sha1-qIwCU1eR8C7TfHahueqXc8gz+MI=",
"dev": true
},
"is-glob": {
"version": "4.0.1",
"resolved": "https://registry.npmjs.org/is-glob/-/is-glob-4.0.1.tgz",
"integrity": "sha512-5G0tKtBTFImOqDnLB2hG6Bp2qcKEFduo4tZu9MT/H6NQv/ghhy30o55ufafxJ/LdH79LLs2Kfrn85TLKyA7BUg==",
"dev": true,
"requires": {
"is-extglob": "^2.1.1"
}
},
"is-number": {
"version": "7.0.0",
"resolved": "https://registry.npmjs.org/is-number/-/is-number-7.0.0.tgz",
"integrity": "sha512-41Cifkg6e8TylSpdtTpeLVMqvSBEVzTttHvERD741+pnZ8ANv0004MRL43QKPDlK9cGvNp6NZWZUBlbGXYxxng==",
"dev": true
},
"normalize-path": {
"version": "3.0.0",
"resolved": "https://registry.npmjs.org/normalize-path/-/normalize-path-3.0.0.tgz",
"integrity": "sha512-6eZs5Ls3WtCisHWp9S2GUy8dqkpGi4BVSz3GaqiE6ezub0512ESztXUwUB6C6IKbQkY2Pnb/mD4WYojCRwcwLA==",
"dev": true
},
"picomatch": {
"version": "2.3.0",
"resolved": "https://registry.npmjs.org/picomatch/-/picomatch-2.3.0.tgz",
"integrity": "sha512-lY1Q/PiJGC2zOv/z391WOTD+Z02bCgsFfvxoXXf6h7kv9o+WmsmzYqrAwY63sNgOxE4xEdq0WyUnXfKeBrSvYw==",
"dev": true
},
"readdirp": {
"version": "3.6.0",
"resolved": "https://registry.npmjs.org/readdirp/-/readdirp-3.6.0.tgz",
"integrity": "sha512-hOS089on8RduqdbhvQ5Z37A0ESjsqz6qnRcffsMU3495FuTdqSm+7bhJ29JvIOsBDEEnan5DPu9t3To9VRlMzA==",
"dev": true,
"requires": {
"picomatch": "^2.2.1"
}
},
"sass": {
"version": "1.68.0",
"resolved": "https://registry.npmjs.org/sass/-/sass-1.68.0.tgz",
"integrity": "sha512-Lmj9lM/fef0nQswm1J2HJcEsBUba4wgNx2fea6yJHODREoMFnwRpZydBnX/RjyXw2REIwdkbqE4hrTo4qfDBUA==",
"dev": true,
"requires": {
"chokidar": ">=3.0.0 <4.0.0",
"immutable": "^4.0.0",
"source-map-js": ">=0.6.2 <2.0.0"
}
},
"source-map-js": {
"version": "1.0.2",
"resolved": "https://registry.npmjs.org/source-map-js/-/source-map-js-1.0.2.tgz",
"integrity": "sha512-R0XvVJ9WusLiqTCEiGCmICCMplcCkIwwR11mOSD9CR5u+IXYdiseeEuXCVAjS54zqwkLcPNnmU4OeJ6tUrWhDw==",
"dev": true
},
"svelte": {
"version": "3.38.3",
"resolved": "https://registry.npmjs.org/svelte/-/svelte-3.38.3.tgz",
"integrity": "sha512-N7bBZJH0iF24wsalFZF+fVYMUOigaAUQMIcEKHO3jstK/iL8VmP9xE+P0/a76+FkNcWt+TDv2Gx1taUoUscrvw==",
"dev": true
},
"svelte-preprocess-filter": {
"version": "1.0.0",
"resolved": "https://registry.npmjs.org/svelte-preprocess-filter/-/svelte-preprocess-filter-1.0.0.tgz",
"integrity": "sha512-92innv59nyEx24xbfcSurB5ocwC8qFdDtGli/JVMHzJsxyvV2yjQKIcbUqU9VIV5mKUWO2PoY93nncS2yF4ULQ==",
"dev": true
},
"svelte-preprocess-sass": {
"version": "2.0.1",
"resolved": "https://registry.npmjs.org/svelte-preprocess-sass/-/svelte-preprocess-sass-2.0.1.tgz",
"integrity": "sha512-0y4FjRsRWcN7rJeNJnSfZ7LVAz6S7/j9Dg24XFRelr/rjMMjXORdEvXy4r38fUYmyk9Y7yjwlHCiqyGxMHhEbg==",
"dev": true,
"requires": {
"svelte-preprocess-filter": "^1.0.0"
}
},
"to-regex-range": {
"version": "5.0.1",
"resolved": "https://registry.npmjs.org/to-regex-range/-/to-regex-range-5.0.1.tgz",
"integrity": "sha512-65P7iz6X5yEr1cwcgvQxbbIw7Uk3gOy5dIdtZ4rDveLqhrdJP+Li/Hx6tyK0NEb+2GCyneCMJiGqrADCSNk8sQ==",
"dev": true,
"requires": {
"is-number": "^7.0.0"
}
}
}
}

8
clipfront2/package.json Normal file
View File

@ -0,0 +1,8 @@
{
"devDependencies": {
"esbuild": "^0.12.15",
"esbuild-svelte": "^0.5.3",
"sass": "^1.68.0",
"svelte-preprocess-sass": "^2.0.1"
}
}

217
clipfront2/src/App.svelte Normal file
View File

@ -0,0 +1,217 @@
<style lang="sass">
\:global(*)
box-sizing: border-box
\:global(html)
scrollbar-color: black lightgray
\:global(body)
font-family: "Fira Sans", "Noto Sans", "Segoe UI", Verdana, sans-serif
font-weight: 300
//margin: 0
//min-height: 100vh
\:global(strong)
font-weight: bold
@mixin header
border-bottom: 1px solid gray
margin: 0
margin-bottom: 0.5em
font-weight: 500
//a
//color: inherit
\:global(h1)
@include header
\:global(h2)
@include header
\:global(h3)
@include header
\:global(h4)
@include header
\:global(h5)
@include header
\:global(h6)
@include header
\:global(ul)
list-style-type: square
padding: 0
padding-left: 1em
input, button, select
border-radius: 0
border: 1px solid gray
padding: 0.5em
.controls
input[type=search]
width: 80%
.ctrlbar
> *
margin: 0 -1px
width: 100%
ul
list-style-type: none
padding: 0
li
display: flex
align-items: center
> *
margin: 0 2px
.result
border: 1px solid gray
.result img
width: 100%
</style>
<h1>Meme Search Engine</h1>
<div class="controls">
<ul>
{#each queryTerms as term}
<li>
<button on:click={removeTerm(term)}>Remove</button>
<select bind:value={term.sign}>
<option>+</option>
<option>-</option>
</select>
<input type="range" min="0" max="2" bind:value={term.weight} step="0.01">
{#if term.type === "image"}
<span>{term.file.name}</span>
{:else if term.type === "text"}
<input type="search" use:focusEl on:keydown={handleKey} bind:value={term.text} />
{/if}
</li>
{/each}
</ul>
<div class="ctrlbar">
<input type="search" placeholder="Text Query" bind:value={query} on:keydown={handleKey} on:focus={newTextQuery}>
<button on:click={pickFile}>Image Query</button>
<button on:click={runSearch} style="margin-left: auto">Search</button>
</div>
</div>
{#if error}
<div>{error}</div>
{/if}
{#if resultPromise}
<Loading />
{/if}
{#if results}
<Masonry bind:refreshLayout={refreshLayout} colWidth="minmax(Min(20em, 100%), 1fr)" items={displayedResults}>
{#each displayedResults as result}
{#key result.file}
<div class="result"><a href={util.getURL(result.file)}><img src={util.getURL(result.file)} on:load={updateCounter} on:error={updateCounter} alt={result.caption || result.file}></a></div>
{/key}
{/each}
</Masonry>
{/if}
<svelte:window on:resize={redrawGrid} on:scroll={handleScroll}></svelte:window>
<script>
import * as util from "./util"
import Loading from "./Loading.svelte"
import Masonry from "./Masonry.svelte"
const chunkSize = 40
let queryTerms = []
const focusEl = el => el.focus()
const newTextQuery = () => {
queryTerms.push({ type: "text", weight: 1, sign: "+", text: "" })
queryTerms = queryTerms
}
const removeTerm = term => {
queryTerms = queryTerms.filter(x => x !== term)
}
let refreshLayout
let heightThreshold
let error
let pendingImageLoads
const recomputeScroll = () => {
const maxOffsets = new Map()
for (const el of document.querySelectorAll(".result")) {
if (el.getAttribute("data-h")) { // layouted
const rect = el.getBoundingClientRect()
maxOffsets.set(rect.left, Math.max(maxOffsets.get(rect.left) || 0, rect.top))
}
}
heightThreshold = Math.min(...maxOffsets.values())
console.log(heightThreshold, pendingImageLoads)
}
const redrawGrid = () => {
if (refreshLayout) refreshLayout().then(recomputeScroll)
}
let resultPromise
let results
let displayedResults = []
let query = ""
const runSearch = async () => {
if (!resultPromise) {
let args = {}
args.text = queryTerms.filter(x => x.type === "text" && x.text).map(({ text, weight, sign }) => [ text, weight * { "+": 1, "-": -1 }[sign] ])
args.images = queryTerms.filter(x => x.type === "image").map(({ imageData, weight, sign }) => [ imageData, weight * { "+": 1, "-": -1 }[sign] ])
resultPromise = util.doQuery(args).then(res => {
error = null
results = res
resultPromise = null
displayedResults = []
pendingImageLoads = 0
for (let i = 0; i < chunkSize; i++) {
displayedResults.push(results[i])
pendingImageLoads += 1
}
redrawGrid()
}).catch(e => { error = e; resultPromise = null })
}
}
const handleScroll = () => {
if (window.scrollY + window.innerHeight >= heightThreshold && pendingImageLoads === 0) {
recomputeScroll()
if (window.scrollY + window.innerHeight < heightThreshold) return;
let init = displayedResults.length
for (let i = 0; i < chunkSize; i++) {
displayedResults.push(results[init + i])
pendingImageLoads += 1
}
displayedResults = displayedResults
}
}
const handleKey = ev => {
if (ev.key === "Enter") {
runSearch()
}
}
const input = document.createElement("input")
input.type = "file"
const pickFile = () => {
input.oninput = ev => {
currentFile = ev.target.files[0]
console.log(currentFile)
if (currentFile) {
let reader = new FileReader()
reader.readAsDataURL(currentFile)
let term = { type: "image", file: currentFile, weight: 1, sign: "+" }
queryTerms.push(term)
queryTerms = queryTerms
reader.onload = () => {
term.imageData = reader.result.split(',')[1]
}
}
}
input.click()
}
const updateCounter = () => {
console.log("redraw")
pendingImageLoads -= 1
redrawGrid()
}
</script>

36
clipfront2/src/Loading.svelte Normal file
View File

@ -0,0 +1,36 @@
<style lang="sass">
.spinner
color: black
.spinner:before
animation: textSpinner 0.8s linear infinite
content: "⠋"
margin-right: 0.5em
padding-top: 0.5em
@keyframes textSpinner
10%
content: "⠙"
20%
content: "⠹"
30%
content: "⠸"
40%
content: "⠼"
50%
content: "⠴"
60%
content: "⠦"
70%
content: "⠧"
80%
content: "⠇"
90%
content: "⠏"
</style>
<script>
export let operation = "Loading"
</script>
<span class="spinner">{operation}</span>

119
clipfront2/src/Masonry.svelte Normal file
View File

@ -0,0 +1,119 @@
<!-- https://github.com/janzheng/svelte-masonry -->
<!--
An almost direct copy and paste of: https://css-tricks.com/a-lightweight-masonry-solution
Usage:
- stretchFirst stretches the first item across the top
<Masonry stretchFirst={true} >
{#each data as o}
<div class="_card _padding">
Here's some stuff {o.name}
<header>
<h3>{o.name}</h3>
</header>
<section>
<p>{o.text}</p>
</section>
</div>
{/each}
</Masonry>
-->
<div bind:this={masonryElement}
class={`__grid--masonry ${stretchFirst ? '__stretch-first' : ''}`}
style={`--grid-gap: ${gridGap}; --col-width: ${colWidth};`}
>
<slot></slot>
</div>
<script>
import { tick } from 'svelte'
export let stretchFirst = false,
gridGap = '0.5em',
colWidth = 'minmax(Min(20em, 100%), 1fr)',
items = [] // pass in data if it's dynamically updated
let grids = [], masonryElement
export const refreshLayout = async () => {
grids.forEach(async grid => {
/* get the post relayout number of columns */
let ncol = getComputedStyle(grid._el).gridTemplateColumns.split(' ').length
grid.items.forEach(c => {
let new_h = c.getBoundingClientRect().height;
if(new_h !== +c.dataset.h) {
c.dataset.h = new_h
grid.mod++
}
});
/* if the number of columns has changed */
if(grid.ncol !== ncol || grid.mod) {
/* update number of columns */
grid.ncol = ncol;
/* revert to initial positioning, no margin */
grid.items.forEach(c => c.style.removeProperty('margin-top'))
/* if we have more than one column */
if(grid.ncol > 1) {
grid.items.slice(ncol).forEach((c, i) => {
let prev_fin = grid.items[i].getBoundingClientRect().bottom /* bottom edge of item above */,
curr_ini = c.getBoundingClientRect().top /* top edge of current item */;
c.style.marginTop = `${prev_fin + grid.gap - curr_ini}px`
})
}
grid.mod = 0
}
})
}
const calcGrid = async (_masonryArr) => {
await tick()
if(_masonryArr.length && getComputedStyle(_masonryArr[0]).gridTemplateRows !== 'masonry') {
grids = _masonryArr.map(grid => {
return {
_el: grid,
gap: parseFloat(getComputedStyle(grid).gridRowGap),
items: [...grid.childNodes].filter(c => c.nodeType === 1 && +getComputedStyle(c).gridColumnEnd !== -1),
ncol: 0,
mod: 0
}
})
refreshLayout() /* initial load */
}
}
$: if(masonryElement) {
calcGrid([masonryElement])
}
$: if(items) { // update if items are changed
masonryElement = masonryElement // refresh masonryElement
}
</script>
<!--
$w: var(--col-width); // minmax(Min(20em, 100%), 1fr);
$s: var(--grid-gap); // .5em;
-->
<style>
:global(.__grid--masonry) {
display: grid;
grid-template-columns: repeat(auto-fit, var(--col-width));
grid-template-rows: masonry;
justify-content: center;
grid-gap: var(--grid-gap);
padding: var(--grid-gap);
}
:global(.__grid--masonry > *) {
align-self: start
}
:global(.__grid--masonry.__stretch-first > *:first-child) {
grid-column: 1/ -1;
}
</style>

5
clipfront2/src/app.js Normal file
View File

@ -0,0 +1,5 @@
import App from "./App.svelte"
new App({
target: document.body,
})

25
clipfront2/src/build.js Normal file
View File

@ -0,0 +1,25 @@
const esbuild = require("esbuild")
const sveltePlugin = require("esbuild-svelte")
const path = require("path")
const { sass } = require("svelte-preprocess-sass")
esbuild
.build({
entryPoints: [path.join(__dirname, "app.js")],
bundle: true,
minify: true,
outfile: path.join(__dirname, "../static/app.js"),
plugins: [sveltePlugin({
preprocess: {
style: sass()
}
})],
loader: {
".woff": "file",
".woff2": "file",
".ttf": "file"
},
logLevel: "info",
watch: process.argv.join(" ").includes("watch")
})
.catch(() => process.exit(1))

11
clipfront2/src/util.js Normal file
View File

@ -0,0 +1,11 @@
import * as config from "../../frontend_config.json"
export const getURL = x => config.image_path + x
export const doQuery = args => fetch(config.backend_url, {
method: "POST",
headers: {
"Content-Type": "application/json"
},
body: JSON.stringify(args)
}).then(x => x.json())

15
clipfront2/static/index.html Normal file
View File

@ -0,0 +1,15 @@
<!doctype html>
<html lang="en" height="100vh">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, minimum-scale=1, initial-scale=1, user-scalable=yes">
<meta name="description" content="osmarks.net meme library semantic search via CLIP; enhanced query UI edition">
<title>Meme Search Engine</title>
</style>
<link rel="stylesheet" href="app.css">
</head>
<body>
<script src="app.js"></script>
</body>
</html>

BIN
demo-image.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.8 MiB

4
frontend_config.json Normal file
View File

@ -0,0 +1,4 @@
{
"backend_url": "https://mse.osmarks.net/backend",
"image_path": "https://i2.osmarks.net/memes-or-something/"
}

199
mse.py Normal file
View File

@ -0,0 +1,199 @@
from aiohttp import web
import aiohttp
import asyncio
import traceback
import umsgpack
from PIL import Image
import base64
import aiosqlite
import faiss
import numpy
import os
import aiohttp_cors
import json
import sys
with open(sys.argv[1], "r") as config_file:
CONFIG = json.load(config_file)
app = web.Application(client_max_size=32*1024**2)
routes = web.RouteTableDef()
async def clip_server(query, unpack_buffer=True):
async with aiohttp.ClientSession() as sess:
async with sess.post(CONFIG["clip_server"], data=umsgpack.dumps(query)) as res:
response = umsgpack.loads(await res.read())
if res.status == 200:
if unpack_buffer:
response = [ numpy.frombuffer(x, dtype="float16") for x in response ]
return response
else:
raise Exception(response if res.headers.get("content-type") == "application/msgpack" else (await res.text()))
@routes.post("/")
async def run_query(request):
data = await request.json()
embeddings = []
if images := data.get("images", []):
embeddings.extend(await clip_server({ "images": [ base64.b64decode(x) for x, w in images ] }))
if text := data.get("text", []):
embeddings.extend(await clip_server({ "text": [ x for x, w in text ] }))
weights = [ w for x, w in images ] + [ w for x, w in text ]
embeddings = [ e * w for e, w in zip(embeddings, weights) ]
if not embeddings:
return web.json_response([])
return web.json_response(app["index"].search(sum(embeddings)))
@routes.get("/")
async def health_check(request):
return web.Response(text="OK")
@routes.post("/reload_index")
async def reload_index_route(request):
await request.app["index"].reload()
return web.json_response(True)
class Index:
def __init__(self, inference_server_config):
self.faiss_index = faiss.IndexFlatIP(inference_server_config["embedding_size"])
self.associated_filenames = []
self.inference_server_config = inference_server_config
self.lock = asyncio.Lock()
def search(self, query):
distances, indices = self.faiss_index.search(numpy.array([query]), 4000)
distances = distances[0]
indices = indices[0]
try:
indices = indices[:numpy.where(indices==-1)[0][0]]
except IndexError: pass
return [ { "score": float(distance), "file": self.associated_filenames[index] } for index, distance in zip(indices, distances) ]
async def reload(self):
async with self.lock:
print("Indexing")
conn = await aiosqlite.connect(CONFIG["db_path"], parent_loop=asyncio.get_running_loop())
conn.row_factory = aiosqlite.Row
await conn.executescript("""
CREATE TABLE IF NOT EXISTS files (
filename TEXT PRIMARY KEY,
modtime REAL NOT NULL,
embedding_vector BLOB NOT NULL
);
""")
try:
async with asyncio.TaskGroup() as tg:
batch_sem = asyncio.Semaphore(3)
modified = set()
async def do_batch(batch):
try:
query = { "images": [ arg[2] for arg in batch ] }
embeddings = await clip_server(query, False)
await conn.executemany("INSERT OR REPLACE INTO files VALUES (?, ?, ?)", [
(filename, modtime, embedding) for (filename, modtime, _), embedding in zip(batch, embeddings)
])
await conn.commit()
for filename, _, _ in batch:
modified.add(filename)
sys.stdout.write(".")
finally:
batch_sem.release()
async def dispatch_batch(batch):
await batch_sem.acquire()
tg.create_task(do_batch(batch))
files = {}
for filename, modtime in await conn.execute_fetchall("SELECT filename, modtime FROM files"):
files[filename] = modtime
await conn.commit()
batch = []
for dirpath, _, filenames in os.walk(CONFIG["files"]):
for file in filenames:
path = os.path.join(dirpath, file)
file = os.path.relpath(path, CONFIG["files"])
st = os.stat(path)
if st.st_mtime != files.get(file):
try:
im = Image.open(path).resize(self.inference_server_config["image_size"]).convert("RGB")
buf = io.BytesIO()
im.save(buf, format="BMP")
b = buf.getvalue()
except Exception as e:
print(file, "failed", e)
continue
batch.append((file, st.st_mtime, b))
if len(batch) % self.inference_server_config["batch"] == self.inference_server_config["batch"] - 1:
await dispatch_batch(batch)
batch = []
if batch:
await dispatch_batch(batch)
remove_indices = []
for index, filename in enumerate(self.associated_filenames):
if filename not in files or filename in modified:
remove_indices.append(index)
self.associated_filenames[index] = None
if filename not in files:
await conn.execute("DELETE FROM files WHERE filename = ?", (filename,))
await conn.commit()
# TODO concurrency
# TODO understand what that comment meant
if remove_indices:
self.faiss_index.remove_ids(numpy.array(remove_indices))
self.associated_filenames = [ x for x in self.associated_filenames if x is not None ]
filenames_set = set(self.associated_filenames)
new_data = []
new_filenames = []
async with conn.execute("SELECT * FROM files") as csr:
while row := await csr.fetchone():
filename, modtime, embedding_vector = row
if filename not in filenames_set:
new_data.append(numpy.frombuffer(embedding_vector, dtype="float16"))
new_filenames.append(filename)
new_data = numpy.array(new_data)
self.associated_filenames.extend(new_filenames)
self.faiss_index.add(new_data)
finally:
await conn.close()
app.router.add_routes(routes)
cors = aiohttp_cors.setup(app, defaults={
"*": aiohttp_cors.ResourceOptions(
allow_credentials=False,
expose_headers="*",
allow_headers="*",
)
})
for route in list(app.router.routes()):
cors.add(route)
async def main():
while True:
async with aiohttp.ClientSession() as sess:
try:
async with await sess.get(CONFIG["clip_server"] + "config") as res:
inference_server_config = umsgpack.unpackb(await res.read())
print("Backend config:", inference_server_config)
break
except:
traceback.print_exc()
await asyncio.sleep(1)
index = Index(inference_server_config)
app["index"] = index
await index.reload()
print("Ready")
runner = web.AppRunner(app)
await runner.setup()
site = web.TCPSite(runner, "", CONFIG["port"])
await site.start()
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
loop.run_until_complete(main())
loop.run_forever()

6
mse_config.json Normal file
View File

@ -0,0 +1,6 @@
{
"clip_server": "http://localhost:1708/",
"db_path": "/srv/mse/data.sqlite3",
"port": 1707,
"files": "/data/public/memes-or-something/"
}

8
requirements.txt Normal file
View File

@ -0,0 +1,8 @@
open_clip_torch==2.20.0
Pillow==10.0.1
prometheus-client==0.17.1
u-msgpack-python==2.8.0
aiohttp==3.8.5
aiohttp-cors==0.7.0
faiss-cpu==1.7.4
aiosqlite==0.19.0