generate-allid-embeddings.py

#!/bin/env python

"""
    Take a CLIPTextModel compatible text encoder.
    Go through the official range of tokens IDs (0-49405)
    Generate the official "embedding" tensor for each one.
    Save the result set to "temp.allids.safetensors"

    Defaults to loading openai/clip-vit-large-patch14 from huggingface hub.
    However, can take optional pair of arguments to a .safetensor model, and config file
    RULES of the loader:
    1. the model file must appear to be either in current directory or one down. So,
      badpath1=some/directory/tree/file.here
      badpath2=/absolutepath
    2. yes, you MUST have a matching config.json file
    3. if you have no alternative, you can get away with using pytorch_model.bin

    Sample location for such things that you can download:
        https://huggingface.co/stablediffusionapi/edge-of-realism/tree/main/text_encoder/
    If there is a .safetensors AND a .bin file, ignore the .bin file

    You can also convert a singlefile model, such as is downloaded from civitai,
    by using the utility at
        https://github.com/huggingface/diffusers/blob/main/scripts/convert_original_stable_diffusion_to_diffusers.py
    Args should look like
        convert_original_stable_diffusion_to_diffusers.py --checkpoint_file somemodel.safetensors \
            --dump_path extractdir --to_safetensors --from_safetensors
    
"""


import sys
import json
import torch
from safetensors.torch import save_file
from transformers import CLIPProcessor,CLIPModel,CLIPTextModel

clipsrc="openai/clip-vit-large-patch14"
processor=None
model=None
encfile=None
configfile=None

if len(sys.argv) == 3:
    encfile=sys.argv[1]
    configfile=sys.argv[2]

device=torch.device("cuda")


def init():
    global processor
    global model
    global encfile
    global configfile

    # Load the processor and model
    print("loading processor from "+clipsrc,file=sys.stderr)
    processor = CLIPProcessor.from_pretrained(clipsrc)
    print("done",file=sys.stderr)

    # originally done this way. But its not the right one to use
    #print("loading model from "+clipsrc,file=sys.stderr)
    #model = CLIPModel.from_pretrained(clipsrc)
    #print("done",file=sys.stderr)
    if encfile != None:
        print("loading model from "+encfile,file=sys.stderr)
        model = CLIPTextModel.from_pretrained(
                encfile,config=configfile,local_files_only=True,use_safetensors=True
                )
    else:
        print("loading model from "+clipsrc,file=sys.stderr)
        model = CLIPTextModel.from_pretrained(clipsrc)

    print("done",file=sys.stderr)

    model = model.to(device)


# "inputs" == magic pre-embedding format
def embed_from_inputs(inputs):
    with torch.no_grad():
        # This way is for CLIPModel
        #text_features = model.get_text_features(**inputs)
        #embedding = text_features[0]

        outputs = model(**inputs)
        embeddings = outputs.pooler_output
        embedding = embeddings

    return embedding


init()
inputs = processor(text="dummy", return_tensors="pt")
inputs.to(device)

all_embeddings = []

for id in range(49405):
    inputs.input_ids[0][1]=id

    emb=embed_from_inputs(inputs)
    all_embeddings.append(emb)
    if (id %100) ==0:
        print(id)

embs = torch.cat(all_embeddings,dim=0)
print("Shape of result = ",embs.shape)

outputfile="cliptextmodel.temp.allids.safetensors"
print(f"Saving the calculatiuons to {outputfile}...")
save_file({"embeddings": embs}, outputfile)