cumo/model/builder.py

#    Copyright 2023 Haotian Liu
#
#    Licensed under the Apache License, Version 2.0 (the "License");
#    you may not use this file except in compliance with the License.
#    You may obtain a copy of the License at
#
#        http://www.apache.org/licenses/LICENSE-2.0
#
#    Unless required by applicable law or agreed to in writing, software
#    distributed under the License is distributed on an "AS IS" BASIS,
#    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#    See the License for the specific language governing permissions and
#    limitations under the License.
# ------------------------------------------------------------------------
# Modified from LLaVA (https://github.com/haotian-liu/LLaVA)
# Copyright 2024 Jiachen Li
# ------------------------------------------------------------------------


import os
import warnings
import shutil

from transformers import AutoTokenizer, AutoModelForCausalLM, AutoConfig, BitsAndBytesConfig
import torch
from cumo.model import *
from cumo.constants import DEFAULT_IMAGE_PATCH_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN

def load_pretrained_model(model_path, model_base, model_name, load_8bit=False, load_4bit=False, device_map="auto", device="cuda", use_flash_attn=False, **kwargs):
    kwargs = {"device_map": device_map, **kwargs}
    if device != "cuda":
        kwargs['device_map'] = {"": device}

    if load_8bit:
        kwargs['load_in_8bit'] = True
    elif load_4bit:
        kwargs['load_in_4bit'] = True
        kwargs['quantization_config'] = BitsAndBytesConfig(
            load_in_4bit=True,
            bnb_4bit_compute_dtype=torch.float16,
            bnb_4bit_use_double_quant=True,
            bnb_4bit_quant_type='nf4'
        )
    else:
        kwargs['torch_dtype'] = torch.float16
    if use_flash_attn:
        kwargs['attn_implementation'] = 'flash_attention_2'

    if 'lora' in model_name.lower() and model_base is None:
        warnings.warn('There is `lora` in model name but no `model_base` is provided. If you are loading a LoRA model, please provide the `model_base` argument. Detailed instruction: https://github.com/haotian-liu/LLaVA#launch-a-model-worker-lora-weights-unmerged.')
    if 'lora' in model_name.lower() and model_base is not None:
        if 'vicuna' in model_base:
            from cumo.model.language_model.llava_llama import LlavaConfig
            lora_cfg_pretrained = LlavaConfig.from_pretrained(model_path)
            tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=False)
            print('Loading from base model...')
            model = LlavaLlamaForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=lora_cfg_pretrained, **kwargs)
            token_num, tokem_dim = model.lm_head.out_features, model.lm_head.in_features
            if model.lm_head.weight.shape[0] != token_num:
                model.lm_head.weight = torch.nn.Parameter(torch.empty(token_num, tokem_dim, device=model.device, dtype=model.dtype))
                model.model.embed_tokens.weight = torch.nn.Parameter(torch.empty(token_num, tokem_dim, device=model.device, dtype=model.dtype))
        elif 'mistral' in model_base:
            if '8x' in model_base:
                from cumo.model.language_model.llava_mixtral import LlavaMixtralConfig
                lora_cfg_pretrained = LlavaMixtralConfig.from_pretrained(model_path)
                tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=False)
                print('Loading from base model...')
                model = LlavaMixtralForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=lora_cfg_pretrained, **kwargs)
                token_num, tokem_dim = model.lm_head.out_features, model.lm_head.in_features
                if model.lm_head.weight.shape[0] != token_num:
                    model.lm_head.weight = torch.nn.Parameter(torch.empty(token_num, tokem_dim, device=model.device, dtype=model.dtype))
                    model.model.embed_tokens.weight = torch.nn.Parameter(torch.empty(token_num, tokem_dim, device=model.device, dtype=model.dtype))
            else:
                from cumo.model.language_model.llava_mistral import LlavaMistralConfig
                lora_cfg_pretrained = LlavaMistralConfig.from_pretrained(model_path)
                tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=False)
                print('Loading from base model...')
                model = LlavaMistralForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=lora_cfg_pretrained, **kwargs)
                token_num, tokem_dim = model.lm_head.out_features, model.lm_head.in_features
                if model.lm_head.weight.shape[0] != token_num:
                    model.lm_head.weight = torch.nn.Parameter(torch.empty(token_num, tokem_dim, device=model.device, dtype=model.dtype))
                    model.model.embed_tokens.weight = torch.nn.Parameter(torch.empty(token_num, tokem_dim, device=model.device, dtype=model.dtype))
            
        print('Loading additional weights...')
        if os.path.exists(os.path.join(model_path, 'non_lora_trainables.bin')):
            non_lora_trainables = torch.load(os.path.join(model_path, 'non_lora_trainables.bin'), map_location='cpu')
        else:
            # this is probably from HF Hub
            from huggingface_hub import hf_hub_download
            def load_from_hf(repo_id, filename, subfolder=None):
                cache_file = hf_hub_download(
                    repo_id=repo_id,
                    filename=filename,
                    subfolder=subfolder)
                return torch.load(cache_file, map_location='cpu')
            non_lora_trainables = load_from_hf(model_path, 'non_lora_trainables.bin')
        non_lora_trainables = {(k[11:] if k.startswith('base_model.') else k): v for k, v in non_lora_trainables.items()}
        if any(k.startswith('model.model.') for k in non_lora_trainables):
            non_lora_trainables = {(k[6:] if k.startswith('model.') else k): v for k, v in non_lora_trainables.items()}
        model.load_state_dict(non_lora_trainables, strict=False)

        from peft import PeftModel
        print('Loading LoRA weights...')
        model = PeftModel.from_pretrained(model, model_path)
        print('Merging LoRA weights...')
        model = model.merge_and_unload()
        print('Model is loaded...')
    else:
        print('Loading from full model...')
        if 'mpt' in model_name.lower():
            tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=True)
            model = LlavaMptForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs)
        elif 'mistral' in model_name.lower() or 'mixtral' in model_name.lower():
            if '8x' in model_name:
                print('Loading CuMo 8x7b model...')
                tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False)
                tokenizer.pad_token = tokenizer.unk_token
                model = LlavaMixtralForCausalLM.from_pretrained(
                    model_path, 
                    low_cpu_mem_usage=True,
                    **kwargs
                )
            else:
                print('Loading CuMo 7b model...')
                tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False)
                tokenizer.pad_token = tokenizer.unk_token
                model = LlavaMistralForCausalLM.from_pretrained(
                    model_path,
                    low_cpu_mem_usage=True,
                    **kwargs
                )
        else:
            print('Loading LLaVA origin from full model...')
            tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False)
            model = LlavaLlamaForCausalLM.from_pretrained(
                model_path,
                low_cpu_mem_usage=True,
                **kwargs
            )

    image_processor = None
    mm_use_im_start_end = getattr(model.config, "mm_use_im_start_end", False)
    mm_use_im_patch_token = getattr(model.config, "mm_use_im_patch_token", True)
    if mm_use_im_patch_token:
        tokenizer.add_tokens([DEFAULT_IMAGE_PATCH_TOKEN], special_tokens=True)
    if mm_use_im_start_end:
        tokenizer.add_tokens([DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN], special_tokens=True)
    model.resize_token_embeddings(len(tokenizer))
    vision_tower = model.get_vision_tower()
    if device_map != 'auto':
        vision_tower.to(device=device_map, dtype=torch.float16)
    image_processor = vision_tower.image_processor

    if hasattr(model.config, "max_sequence_length"):
        context_len = model.config.max_sequence_length
    else:
        context_len = 2048

    return tokenizer, model, image_processor, context_len