summaryrefslogtreecommitdiff
path: root/priv/python/pyerlai/genservers/clip_ask.py
blob: 0cb40dc96f3314e6f9ac01674110601586cdb56c (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
from term import Atom
from pyrlang.gen.server import GenServer
from pyrlang.gen.decorators import call, cast, info
from PIL import Image
import io
import sys

from transformers import CLIPProcessor, CLIPModel

PROMPTS=[
"photo",
"dog photo",
"cat photo",
"food photo",
"meme",
"painting",
"drawing",
"selfie",
"portrait photography",
"tv capture",
"screenshot",
"terminal/ssh/console screenshot",
"twitter screenshot",
"chat log",
"4chan screenshot",
"scanned document",
"book picture"]

class ClipAsk(GenServer):
    def __init__(self, node, logger) -> None:
        super().__init__()
        node.register_name(self, Atom('clip_ask'))
        self.logger = logger
        self.model = None
        self.processor = None
        self.ready = False
        print("clipask: starting")
        mypid = self.pid_
        node.send_nowait(mypid, mypid, "register")
        self.logger.info("initialized process: clip_ask.")

    @info(0, lambda msg: msg == 'register')
    def setup(self, msg):
        print("clipask: doing setup")
        self.logger.info("image_to_text_vit_gpt2: setup...")
        self.model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
        self.processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
        self.logger.info("clip_ask: setup finished.")
        self.ready = True
        print("clipask: ready")

    @call(1, lambda msg: type(msg) == tuple and msg[0] == Atom("run"))
    def run(self, msg):
        if self.ready:
            self.logger.info("clip_ask: inference")
            image = Image.open(io.BytesIO(msg[1]))
            inputs = self.processor(text=PROMPTS, images=image, return_tensors="pt", padding=True)
            outputs = self.model(**inputs)
            logits_per_image = outputs.logits_per_image
            probs = logits_per_image.softmax(dim=1)
            labels_with_probs = dict(zip(PROMPTS, probs.detach().numpy()[0]))
            results = dict(sorted(labels_with_probs.items(), key=lambda item: item[1], reverse=True))
            return (Atom('ok'), {k: v.item() for k, v in results.items()})
        else:
            return (Atom('error'), Atom('not_ready'))