nousr

aofe scaffolding spec — cleaned answers

1. domain characterization

1.1 — what constitutes a "project" or "domain" as input to the scaffolder?
the input should be a short, slug-style project name. it does not need to be smart or highly structured. its main job is to seed the rest of the project scaffold.

1.2 — is the scaffolder doing discovery or audit?
neither, really. the project description should simply act as a short extension of the slug: a few sentences describing the task being optimized. it is there to seed the scaffold, not to discover or verify deep domain properties.

nousr

""""
Proof of concept "DiM" - nousr

general structure was "transpiled" from DiT by meta
bi-direction idea comes from DifuSSM (https://arxiv.org/abs/2311.18257)
"""
import torch
import math

from timm.models.vision_transformer import PatchEmbed

nousr

from flask import Flask, request, jsonify
import torch
import base64
import io
from PIL import Image
import requests
from io import BytesIO

# Import the necessary modules from the llava directory
from transformers import AutoTokenizer

nousr

#!/bin/bash
clip-retrieval inference \
--input_dataset="<parent folder containing images>" \
--output_folder="<output s3 bucket or local folder>" \
--input_format="files" \
--enable_metadata=False \
--write_batch_size=500 \
--num_prepro_workers=2 \
--batch_size=64 \
--enable_wandb=True \

nousr

from langchain.llms.base import LLM
from typing import Optional, List
from revChatGPT.V1 import Chatbot

class revChatGPT(LLM):
    
    chatbot: Chatbot = Chatbot(config={
        "access_token": "<your token here>"
    })

nousr

import torch
import webdataset as wds
from typing import Union, Sequence
from time import perf_counter
from accelerate import Accelerator

def process(src):
    """
    Filter empty samples from the clap dataset.

nousr

@lru_cache(maxsize=None)
def load_safety_model(clip_model):
    """load the safety model"""
    import torch  # pylint: disable=import-outside-toplevel
    import autokeras as ak  # pylint: disable=import-outside-toplevel
    from tensorflow.keras.models import load_model  # pylint: disable=import-outside-toplevel

    class H14_NSFW_Detector(nn.Module):
            def __init__(self, input_size=1024):
                super().__init__()

nousr

name,prompt,negative_prompt
None,,
MDJ_style_base,"an extremely realistic digital painting of a (futuristic an alleyway) street in the style of cyberpunk, a photograph of a cyberpunk street alley at sunset, digital art trending on artstation, ((pixel art)), light rays shining","blurry, bad resolution, low quality, watermarked image, america, jpeg artifacts, background, words, font, text, boring, robot, human, full body, people in a street, purple"
MDJ_street_car,"an extremely realistic digital painting of a (cyberpunk volkswagen r32) on the street in the style of cyberpunk, a photograph of a cyberpunk street alley at night, digital art trending on artstation, ((pixel art)), light rays shining","bad resolution, low quality, watermarked image, human, ((simple)), [dark], blurry, bad resolution, low quality, ((watermarked image)), america, jpeg artifacts, words, font, text, boring, robot, human, full body, people in a street stock photo AND alamy, shutterstock, pixabay, flickr AND picture frame, bad perspective, 

nousr

#!/bin/bash

# step 1: get environment variables
# step 2: setup rank 0 to be the master and start the indexing python file (the last two operations happen in parallel)
# step 3: setup rank 1-N to be the workers and start the worker script (they then will listen for work from the master)

# get environment variables
GLOBAL_RANK=$SLURM_PROCID
CPUS=$SLURM_CPUS_PER_TASK
MEM=$SLURM_MEM_PER_NODE # seems to be in MB

nousr

import wandb
import torch
from dalle2_pytorch import (
    T5OpenClipAdapter,
    DiffusionPriorNetwork,
    DiffusionPrior,
)
from dalle2_pytorch.trainer import DiffusionPriorTrainer
from transformers import T5Tokenizer
from accelerate import Accelerator