from transformers import AutoProcessor from vllm import LLM, SamplingParams import os from PIL import Image from .conversations import internvl_conv class InternVL: def __init__(self,model_path,args): super().__init__() self.llm = LLM( model= model_path, trust_remote_code=True, tensor_parallel_size= int(os.environ.get("tensor_parallel_size",4)), enforce_eager=True, gpu_memory_utilization = 0.7, limit_mm_per_prompt = {"image": int(os.environ.get("max_image_num",1))}, ) self.processor = AutoProcessor.from_pretrained(model_path,trust_remote_code=True) self.sampling_params = SamplingParams( temperature=args.temperature, top_p=args.top_p, repetition_penalty=args.repetition_penalty, max_tokens= args.max_new_tokens, stop_token_ids=[], ) def process_messages(self,messages): conv = internvl_conv.copy() conv.messages = [] imgs = [] if "messages" in messages: messages = messages["messages"] for message in messages: role = message["role"] content = message["content"] if role == "user": conv.append_message(conv.roles[0],content) else: conv.append_message(conv.roles[1],content) else: if "system" in messages: conv.system_message = messages["system"] if "image" in messages: text = messages["prompt"] inp = "" + '\n' + text conv.append_message(conv.roles[0],inp) image = messages["image"] if isinstance(image,str): if os.path.exists(image): image = Image.open(image) elif isinstance(image,Image.Image): image = image.convert("RGB") imgs.append(image) elif "images" in messages: text = messages["prompt"] images = messages["images"] inp = "" for i,image in enumerate(images): inp = inp + f": " +"" + '\n' if isinstance(image,str): if os.path.exists(image): image = Image.open(image) elif isinstance(image,Image.Image): image = image.convert("RGB") imgs.append(image) inp = inp + text conv.append_message(conv.roles[0],inp) else: text = messages["prompt"] inp = text conv.append_message(conv.roles[0],inp) conv.append_message(conv.roles[1],None) prompt = conv.get_prompt() mm_data = {} if len(imgs) > 0: mm_data["image"] = imgs llm_inputs = { "prompt": prompt, "multi_modal_data": mm_data, } return llm_inputs def generate_output(self,messages): llm_inputs = self.process_messages(messages) outputs = self.llm.generate([llm_inputs], sampling_params=self.sampling_params) return outputs[0].outputs[0].text def generate_outputs(self,messages_list): llm_inputs_list = [self.process_messages(messages) for messages in messages_list] outputs = self.llm.generate(llm_inputs_list, sampling_params=self.sampling_params) res = [] for output in outputs: generated_text = output.outputs[0].text res.append(generated_text) return res