stable-diffusion-webui-forge/backend/text_processing/anima_engine.py

import torch

from backend import memory_management
from backend.text_processing import emphasis, parsing
from modules.shared import opts


class PromptChunk:
    def __init__(self):
        self.qwen_tokens = []
        self.qwen_multipliers = []
        self.t5_tokens = []
        self.t5_multipliers = []


class AnimaTextProcessingEngine:
    def __init__(self, text_encoder, qwen_tokenizer, t5_tokenizer):
        super().__init__()

        self.text_encoder = text_encoder
        self.qwen_tokenizer = qwen_tokenizer
        self.t5_tokenizer = t5_tokenizer

        self.id_pad = 151643
        self.id_end = 1

    def tokenize(self, texts):
        return (
            self.qwen_tokenizer(texts, truncation=False, add_special_tokens=False)["input_ids"],
            self.t5_tokenizer(texts, truncation=False, add_special_tokens=False)["input_ids"],
        )

    def tokenize_line(self, line):
        parsed = parsing.parse_prompt_attention(line, self.emphasis.name)
        qwen_tokenized, t5_tokenized = self.tokenize([text for text, _ in parsed])

        chunks = []
        chunk = PromptChunk()

        def next_chunk():
            nonlocal chunk

            if not chunk.qwen_tokens:
                chunk.qwen_tokens.append(self.id_pad)
                chunk.qwen_multipliers.append(1.0)

            chunk.t5_tokens.append(self.id_end)
            chunk.t5_multipliers.append(1.0)

            chunks.append(chunk)
            chunk = PromptChunk()

        for tokens in qwen_tokenized:
            position = 0
            while position < len(tokens):
                token = tokens[position]
                chunk.qwen_tokens.append(token)
                chunk.qwen_multipliers.append(1.0)
                position += 1

        for tokens, (text, weight) in zip(t5_tokenized, parsed):
            position = 0
            while position < len(tokens):
                token = tokens[position]
                chunk.t5_tokens.append(token)
                chunk.t5_multipliers.append(weight)
                position += 1

        if not chunks:
            next_chunk()

        return chunks

    def __call__(self, texts):
        zs, ti, tw = [], [], []
        cache = {}

        self.emphasis = emphasis.get_current_option(opts.emphasis)()

        for line in texts:
            if line in cache:
                z = cache[line]
            else:
                chunks: list[PromptChunk] = self.tokenize_line(line)
                assert len(chunks) == 1

                for chunk in chunks:
                    tokens = chunk.qwen_tokens
                    multipliers = chunk.qwen_multipliers

                    z: torch.Tensor = self.process_tokens([tokens], [multipliers])[0]

                cache[line] = z

            zs.append(z)
            ti.append(torch.tensor(chunk.t5_tokens, dtype=torch.int))
            tw.append(torch.tensor(chunk.t5_multipliers))

        z = {
            "qwen_cond": zs,
            "t5_ids": ti,
            "t5_weights": tw,
        }

        return z

    def process_embeds(self, batch_tokens):
        device = memory_management.text_encoder_device()

        embeds_out = []
        attention_masks = []
        num_tokens = []

        for tokens in batch_tokens:
            attention_mask = []
            tokens_temp = []
            other_embeds = []
            eos = False
            index = 0

            for t in tokens:
                try:
                    token = int(t)
                    attention_mask.append(0 if eos else 1)
                    tokens_temp += [token]
                    if not eos and token == self.id_pad:
                        eos = True
                except TypeError:
                    other_embeds.append((index, t))
                index += 1

            tokens_embed = torch.tensor([tokens_temp], device=device, dtype=torch.long)
            tokens_embed = self.text_encoder.get_input_embeddings()(tokens_embed)

            index = 0
            embeds_info = []

            for o in other_embeds:
                emb, extra = self.text_encoder.preprocess_embed(o[1], device=device)
                if emb is None:
                    index += -1
                    continue

                ind = index + o[0]
                emb = emb.view(1, -1, emb.shape[-1]).to(device=device, dtype=torch.float32)
                emb_shape = emb.shape[1]

                assert emb.shape[-1] == tokens_embed.shape[-1]
                tokens_embed = torch.cat([tokens_embed[:, :ind], emb, tokens_embed[:, ind:]], dim=1)
                attention_mask = attention_mask[:ind] + [1] * emb_shape + attention_mask[ind:]
                index += emb_shape - 1
                emb_type = o[1].get("type", None)
                embeds_info.append({"type": emb_type, "index": ind, "size": emb_shape, "extra": extra})

            embeds_out.append(tokens_embed)
            attention_masks.append(attention_mask)
            num_tokens.append(sum(attention_mask))

        return torch.cat(embeds_out), torch.tensor(attention_masks, device=device, dtype=torch.long), num_tokens, embeds_info

    def process_tokens(self, batch_tokens, batch_multipliers):
        embeds, mask, count, info = self.process_embeds(batch_tokens)
        z, _ = self.text_encoder(input_ids=None, embeds=embeds, attention_mask=mask, num_tokens=count, embeds_info=info)
        return z