src/flux/modules/conditioner.py

from torch import Tensor, nn
from transformers import (CLIPTextModel, CLIPTokenizer, T5EncoderModel,
                          T5Tokenizer)
import os

class HFEmbedder(nn.Module):
    def __init__(self, version: str, max_length: int, is_clip, **hf_kwargs):
        super().__init__()
        self.is_clip = is_clip
        self.max_length = max_length
        self.output_key = "pooler_output" if self.is_clip else "last_hidden_state"

        if self.is_clip:
            self.tokenizer: CLIPTokenizer = CLIPTokenizer.from_pretrained(version, max_length=max_length)
            self.hf_module: CLIPTextModel = CLIPTextModel.from_pretrained(version, **hf_kwargs)
        else:
            self.tokenizer: T5Tokenizer = T5Tokenizer.from_pretrained(version, max_length=max_length)
            self.hf_module: T5EncoderModel = T5EncoderModel.from_pretrained(version, **hf_kwargs)

        self.hf_module = self.hf_module.eval().requires_grad_(False)


    def forward(self, text: list[str]) -> Tensor:
        batch_encoding = self.tokenizer(
            text,
            truncation=True,
            max_length=self.max_length,
            return_length=False,
            return_overflowing_tokens=False,
            padding="max_length",
            return_tensors="pt",
        )

        if self.is_clip:
            flag = 'clip'
        else:
            flag = 't5'
        print(f'foward {flag}')
        input_ids = batch_encoding["input_ids"]
        print(f"input_ids shape: {input_ids.shape}, max_length: {self.max_length}")  # Debug
        assert input_ids.shape[1] == self.max_length, f"Sequence length {input_ids.shape[1]} does not match max_length {self.max_length}"
        print(input_ids)

        print(f"self.tokenizer.vocab_size: {self.tokenizer.vocab_size}")  # Debug
        print(f"self.hf_module.config.vocab_size: {self.hf_module.config.vocab_size}")  # Debug
        print(f"self.tokenizer.vocab_size: {self.tokenizer.vocab_size}")  # Debug
        print(f"self.hf_module.config.vocab_size: {self.hf_module.config.vocab_size}")  # Debug

        outputs = self.hf_module(
            input_ids=input_ids.to(self.hf_module.device),
            attention_mask=batch_encoding["attention_mask"].to(self.hf_module.device),
            output_hidden_states=False,
        )
        return outputs[self.output_key]