import os import warnings import torch from transformers import AutoTokenizer, AutoConfig, BitsAndBytesConfig, logging logging.set_verbosity_error() warnings.filterwarnings('ignore') from bunny.model import * def load_pretrained_model(model_path, model_base, model_name, model_type, load_8bit=False, load_4bit=False, device_map="auto", device="cuda", **kwargs): if model_type not in {'phi-1.5', 'phi-2', 'phi-3', 'stablelm-2', 'qwen1.5-1.8b', 'minicpm', 'llama3-8b'}: raise ValueError(f"Unknown Model Type {model_type}") 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 # Load Bunny model 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.') if 'lora' in model_name.lower() and model_base is not None: lora_cfg_pretrained = AutoConfig.from_pretrained(model_path) print('Loading Bunny from base model...') if model_type == 'phi-1.5' or model_type == 'phi-2': tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True) model = BunnyPhiForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=lora_cfg_pretrained, **kwargs) elif model_type == 'phi-3': tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True) model = BunnyPhi3ForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=lora_cfg_pretrained, **kwargs) elif model_type == 'stablelm-2': tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True, trust_remote_code=True) model = BunnyStableLMForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=lora_cfg_pretrained, **kwargs) elif model_type == 'qwen1.5-1.8b': tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True) model = BunnyQwen2ForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=lora_cfg_pretrained, **kwargs) elif model_type == 'minicpm': tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True) model = BunnyMiniCPMForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=lora_cfg_pretrained, **kwargs) elif model_type == 'llama3-8b': tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True) model = BunnyLlamaForCausalLM.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 Bunny 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...') elif model_base is not None: # this may be mm projector only print('Loading Bunny from base model...') cfg_pretrained = AutoConfig.from_pretrained(model_path) if model_type == 'phi-1.5' or model_type == 'phi-2': tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True) model = BunnyPhiForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=cfg_pretrained, **kwargs) elif model_type == 'phi-3': tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True) model = BunnyPhi3ForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=cfg_pretrained, **kwargs) elif model_type == 'stablelm-2': tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True, trust_remote_code=True) model = BunnyStableLMForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=cfg_pretrained, **kwargs) elif model_type == 'qwen1.5-1.8b': tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True) model = BunnyQwen2ForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=cfg_pretrained, **kwargs) elif model_type == 'minicpm': tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True) model = BunnyMiniCPMForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=cfg_pretrained, **kwargs) elif model_type == 'llama3-8b': tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True) model = BunnyLlamaForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=cfg_pretrained, **kwargs) mm_projector_weights = torch.load(os.path.join(model_path, 'mm_projector.bin'), map_location='cpu') mm_projector_weights = {k: v.to(torch.float16) for k, v in mm_projector_weights.items()} model.load_state_dict(mm_projector_weights, strict=False) else: if model_type == 'phi-1.5' or model_type == 'phi-2': tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=True) model = BunnyPhiForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs) elif model_type == 'phi-3': tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=True) model = BunnyPhi3ForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs) elif model_type == 'stablelm-2': tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=True, trust_remote_code=True) model = BunnyStableLMForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs) elif model_type == 'qwen1.5-1.8b': tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=True) model = BunnyQwen2ForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs) elif model_type == 'minicpm': tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=True) model = BunnyMiniCPMForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs) elif model_type == 'llama3-8b': tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=True) model = BunnyLlamaForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs) model.resize_token_embeddings(len(tokenizer)) vision_tower = model.get_vision_tower() if not vision_tower.is_loaded: vision_tower.load_model() # if getattr(model.config, "unfreeze_vision_tower", False): # if 'lora' in model_name.lower(): # assert model_base is not None # vision_non_lora_trainables = {k[19:]: v for k, v in non_lora_trainables.items() if # k.startswith('model.vision_tower.')} # vision_tower.load_state_dict(vision_non_lora_trainables, strict=False) # else: # assert model_base is None # from safetensors.torch import load_file # vision_weights = {} # for file_name in os.listdir(model_path): # if file_name.endswith('safetensors'): # vision_weights.update( # {k[19:]: v for k, v in load_file(os.path.join(model_path, file_name)).items() if # k.startswith('model.vision_tower.')}) # vision_tower.load_state_dict(vision_weights, strict=True) vision_tower.to(device=device, 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 if model_type == 'llama3-8b': tokenizer.eos_token_id = 128001 model.generation_config.pad_token_id = tokenizer.eos_token_id if model.generation_config.pad_token_id is None: model.generation_config.pad_token_id = model.generation_config.eos_token_id return tokenizer, model, image_processor, context_len