import os import torch import torch.nn as nn import numpy as np import random from transformers import ( BartForConditionalGeneration, AutoModelForCausalLM, BertModel, Wav2Vec2Model, CLIPModel, AutoTokenizer ) class MultiModalModel(nn.Module): def __init__(self): super(MultiModalModel, self).__init__() # 初始化子模型 self.text_generator = BartForConditionalGeneration.from_pretrained('facebook/bart-base') self.code_generator = AutoModelForCausalLM.from_pretrained('gpt2') self.nlp_encoder = BertModel.from_pretrained('bert-base-uncased') self.speech_encoder = Wav2Vec2Model.from_pretrained('facebook/wav2vec2-base-960h') self.vision_encoder = CLIPModel.from_pretrained('openai/clip-vit-base-patch32') # 初始化分词器和处理器 self.text_tokenizer = AutoTokenizer.from_pretrained('facebook/bart-base') self.code_tokenizer = AutoTokenizer.from_pretrained('gpt2') self.nlp_tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased') self.speech_processor = AutoTokenizer.from_pretrained('facebook/wav2vec2-base-960h') self.vision_processor = AutoTokenizer.from_pretrained('openai/clip-vit-base-patch32') def forward(self, task, inputs): if task == 'text_generation': # 确保 attention_mask 在 inputs 中 attention_mask = inputs.get('attention_mask') print("输入数据:", inputs) outputs = self.text_generator.generate( inputs['input_ids'], max_new_tokens=100, # 增加生成的最大新令牌数 pad_token_id=self.text_tokenizer.eos_token_id, attention_mask=attention_mask, top_p=0.9, # 调整 top_p 值 top_k=50, # 保持 top_k 值 temperature=0.8, # 调整 temperature 值 do_sample=True ) print("生成的输出:", outputs) return self.text_tokenizer.decode(outputs[0], skip_special_tokens=True) # 根据需要添加其他任务的逻辑... # 主函数 if __name__ == "__main__": # 初始化模型 model = MultiModalModel() # 示例任务和输入数据 task = "text_generation" input_text = "This is a sample input." tokenizer = model.text_tokenizer inputs = tokenizer(input_text, return_tensors='pt') # 添加 attention_mask 键值对 inputs['attention_mask'] = torch.ones_like(inputs['input_ids']) # 模型推理 result = model(task, inputs) print("最终输出结果:", result) trust_remote_code=True