threestudio/scripts/train_dreambooth.py [1097:1157]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    )
    optimizer = optimizer_class(
        params_to_optimize,
        lr=args.learning_rate,
        betas=(args.adam_beta1, args.adam_beta2),
        weight_decay=args.adam_weight_decay,
        eps=args.adam_epsilon,
    )

    if args.pre_compute_text_embeddings:

        def compute_text_embeddings(prompt):
            with torch.no_grad():
                text_inputs = tokenize_prompt(tokenizer, prompt, tokenizer_max_length=args.tokenizer_max_length)
                prompt_embeds = encode_prompt(
                    text_encoder,
                    text_inputs.input_ids,
                    text_inputs.attention_mask,
                    text_encoder_use_attention_mask=args.text_encoder_use_attention_mask,
                )

            return prompt_embeds

        pre_computed_encoder_hidden_states = compute_text_embeddings(args.instance_prompt)
        validation_prompt_negative_prompt_embeds = compute_text_embeddings("")

        if args.validation_prompt is not None:
            validation_prompt_encoder_hidden_states = compute_text_embeddings(args.validation_prompt)
        else:
            validation_prompt_encoder_hidden_states = None

        if args.class_prompt is not None:
            pre_computed_class_prompt_encoder_hidden_states = compute_text_embeddings(args.class_prompt)
        else:
            pre_computed_class_prompt_encoder_hidden_states = None

        text_encoder = None
        tokenizer = None

        gc.collect()
        torch.cuda.empty_cache()
    else:
        pre_computed_encoder_hidden_states = None
        validation_prompt_encoder_hidden_states = None
        validation_prompt_negative_prompt_embeds = None
        pre_computed_class_prompt_encoder_hidden_states = None

    # Dataset and DataLoaders creation:
    train_dataset = DreamBoothDataset(
        instance_data_root=args.instance_data_dir,
        instance_prompt=args.instance_prompt,
        class_data_root=args.class_data_dir if args.with_prior_preservation else None,
        class_prompt=args.class_prompt,
        class_num=args.num_class_images,
        tokenizer=tokenizer,
        size=args.resolution,
        center_crop=args.center_crop,
        encoder_hidden_states=pre_computed_encoder_hidden_states,
        class_prompt_encoder_hidden_states=pre_computed_class_prompt_encoder_hidden_states,
        tokenizer_max_length=args.tokenizer_max_length,
        use_view_dependent_prompt=args.use_view_dependent_prompt,
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



threestudio/scripts/train_dreambooth_lora.py [1020:1080]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    )
    optimizer = optimizer_class(
        params_to_optimize,
        lr=args.learning_rate,
        betas=(args.adam_beta1, args.adam_beta2),
        weight_decay=args.adam_weight_decay,
        eps=args.adam_epsilon,
    )

    if args.pre_compute_text_embeddings:

        def compute_text_embeddings(prompt):
            with torch.no_grad():
                text_inputs = tokenize_prompt(tokenizer, prompt, tokenizer_max_length=args.tokenizer_max_length)
                prompt_embeds = encode_prompt(
                    text_encoder,
                    text_inputs.input_ids,
                    text_inputs.attention_mask,
                    text_encoder_use_attention_mask=args.text_encoder_use_attention_mask,
                )

            return prompt_embeds

        pre_computed_encoder_hidden_states = compute_text_embeddings(args.instance_prompt)
        validation_prompt_negative_prompt_embeds = compute_text_embeddings("")

        if args.validation_prompt is not None:
            validation_prompt_encoder_hidden_states = compute_text_embeddings(args.validation_prompt)
        else:
            validation_prompt_encoder_hidden_states = None

        if args.class_prompt is not None:
            pre_computed_class_prompt_encoder_hidden_states = compute_text_embeddings(args.class_prompt)
        else:
            pre_computed_class_prompt_encoder_hidden_states = None

        text_encoder = None
        tokenizer = None

        gc.collect()
        torch.cuda.empty_cache()
    else:
        pre_computed_encoder_hidden_states = None
        validation_prompt_encoder_hidden_states = None
        validation_prompt_negative_prompt_embeds = None
        pre_computed_class_prompt_encoder_hidden_states = None

    # Dataset and DataLoaders creation:
    train_dataset = DreamBoothDataset(
        instance_data_root=args.instance_data_dir,
        instance_prompt=args.instance_prompt,
        class_data_root=args.class_data_dir if args.with_prior_preservation else None,
        class_prompt=args.class_prompt,
        class_num=args.num_class_images,
        tokenizer=tokenizer,
        size=args.resolution,
        center_crop=args.center_crop,
        encoder_hidden_states=pre_computed_encoder_hidden_states,
        class_prompt_encoder_hidden_states=pre_computed_class_prompt_encoder_hidden_states,
        tokenizer_max_length=args.tokenizer_max_length,
        use_view_dependent_prompt=args.use_view_dependent_prompt,
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -